Compare commits

...

57 commits

Author SHA1 Message Date
Alex 94d723f27c Merge pull request 'Implement rpc_secret_file' (#466) from felix.scheinost/garage:feature/implement-rpc-secret-file into main
Reviewed-on: Deuxfleurs/garage#466
2023-01-11 16:04:35 +00:00
Alex be6b8f419d Merge pull request 'Implemented system metrics' (#472) from jpds/garage:system-metrics into main
Reviewed-on: Deuxfleurs/garage#472
Reviewed-by: Alex <alex@adnab.me>
2023-01-11 16:00:31 +00:00
Jonathan Davies 4cfb469d2b block/metrics.rs: Added compression_level metric. 2023-01-10 10:40:03 +00:00
Jonathan Davies df1d9a9873 system.rs: Integrated SystemMetrics into System implementation. 2023-01-10 10:39:50 +00:00
Jonathan Davies aac348fe93 Added system_metrics.rs file. 2023-01-10 10:38:50 +00:00
Felix Scheinost d6ea0cbefa Add tests for rpc_secret_file 2023-01-07 14:19:36 +01:00
Felix Scheinost 7b62fe3f0b Error on both rpc_secret and rpc_secret_file 2023-01-07 13:49:03 +01:00
Felix Scheinost f2106c2733 Implement rpc_secret_file 2023-01-04 18:35:10 +01:00
Alex 02e8eb167e Merge pull request 'PutObject: better cleanup when request is interrupted in the middle' (#462) from interrupted-cleanup into main
Reviewed-on: Deuxfleurs/garage#462
2023-01-04 14:43:45 +00:00
Alex 329c0e64f9 Merge pull request 'Improve garage worker set and add garage worker get' (#464) from worker-get into main
Reviewed-on: Deuxfleurs/garage#464
2023-01-04 13:47:42 +00:00
Alex 29dbcb8278
bg var operation on all nodes at once 2023-01-04 13:25:57 +01:00
Alex f3f27293df
Uniform framework for bg variable management 2023-01-04 13:07:13 +01:00
Alex 13c5549886
Remove token_bucket.rs 2023-01-04 11:47:56 +01:00
Alex 936b6cb563
When saving block, delete .tmp file if we could not complete 2023-01-03 17:34:26 +01:00
Alex 0650a43cf1
PutObject: better cleanup on Drop (incl. when request is interrupted in the middle) 2023-01-03 17:05:17 +01:00
Alex 4eb8ca3a52 Merge pull request 'Fix Consul & Kubernetes discovery with new way of doing background things' (#463) from fix-background into main
Reviewed-on: Deuxfleurs/garage#463
2023-01-03 16:04:40 +00:00
Alex 1fc220886a
Fix Consul & Kubernetes discovery with new way of doing background things 2023-01-03 16:55:59 +01:00
Alex 73ed9c7403 Merge pull request 'Refactor how things are migrated' (#461) from format-migration into main
Reviewed-on: Deuxfleurs/garage#461
2023-01-03 15:28:24 +00:00
Alex 1d5bdc17a4
use impossible enum type 2023-01-03 16:04:06 +01:00
Alex c106304b9c
more idiomatic and shorter 2023-01-03 16:00:19 +01:00
Alex 33f25d26c7
fix doc and add tests for migrate.rs 2023-01-03 15:53:13 +01:00
Alex d6d571d512
cargo fmt 2023-01-03 15:30:21 +01:00
Alex a54b67740d
move debug_serialize to garage_util::encode 2023-01-03 15:29:29 +01:00
Alex 8d5505514f
Make it explicit when using nonversioned encoding 2023-01-03 15:27:36 +01:00
Alex 426d8784da
cleanup 2023-01-03 15:08:37 +01:00
Alex a81200d345
Update cargo.nix 2023-01-03 14:45:47 +01:00
Alex cdb2a591e9
Refactor how things are migrated 2023-01-03 14:44:47 +01:00
Alex 582b076179 Merge pull request 'Some improvements to Garage internals' (#451) from internals-rework into main
Reviewed-on: Deuxfleurs/garage#451
2023-01-03 11:37:31 +00:00
Alex 939a6d67e8
Merge branch 'main' into internals-rework 2023-01-02 15:07:44 +01:00
Alex 76230f2028 Merge pull request 'Bump everything to v0.8.1' (#458) from up-v0.8.1 into main
Reviewed-on: Deuxfleurs/garage#458
2023-01-02 13:32:45 +00:00
Alex 6775569525
Bump everything to v0.8.1 2023-01-02 14:15:33 +01:00
Alex 6b857a9b8c
cargo fmt 2023-01-02 13:50:42 +01:00
Alex 1649002e2b Merge pull request 'Add a note about Peertube 5.0 private videos' (#456) from kaiyou/garage:docs-apps into main
Reviewed-on: Deuxfleurs/garage#456
2023-01-02 12:49:14 +00:00
Alex 822e344845 Merge pull request 'Add some docs about using Python Minio SDK' (#455) from kaiyou/garage:docs-s3-libs into main
Reviewed-on: Deuxfleurs/garage#455
2023-01-02 12:48:52 +00:00
Alex 7f7d53cfa9 Merge pull request 'improvements to CLI and new debug features' (#448) from cli-improvements into main
Reviewed-on: Deuxfleurs/garage#448
2023-01-02 12:42:24 +00:00
kaiyou 0c7ed0b0af Add some docs about using Python Minio SDK 2022-12-25 13:55:12 +01:00
Alex 1fcd0b371b
online repair workers: retry on error 2022-12-14 16:31:31 +01:00
Alex 13c8662126
factorize 2022-12-14 16:16:55 +01:00
Alex e6f14ab5cf
better error message handling 2022-12-14 16:11:19 +01:00
Alex 510b620108
Get rid of background::spawn 2022-12-14 16:08:05 +01:00
Alex dfc131850a
Simplified and more aggressive worker exit logic 2022-12-14 15:25:29 +01:00
Alex d4af27f920
Add missing notify 2022-12-14 13:54:21 +01:00
Alex 0d6b05bb6c
Update cargo.nix 2022-12-14 12:58:24 +01:00
Alex a19bfef508
Improve error message on rpc connection failure 2022-12-14 12:57:33 +01:00
Alex d56c472712
Refactor background runner and get rid of job worker 2022-12-14 12:51:42 +01:00
Alex 2183518edc
Spawn all background workers in a separate step 2022-12-14 12:28:07 +01:00
Alex 83c8467e23
Proper queueing for delayed inserts, now backed to disk 2022-12-14 11:58:06 +01:00
Alex f8e528c15d
Small refactor of tables internals 2022-12-14 10:48:49 +01:00
Alex d1279e04f3
Fix error messages 2022-12-13 16:18:01 +01:00
Alex 041b60ed1d
Add block.rc_size, table.size and table.merkle_tree_size metrics 2022-12-13 15:54:03 +01:00
Alex f8d5409894
cli: more info displayed on error in garage stats 2022-12-13 15:46:04 +01:00
Alex d6040e32a6
cli: prettier table in garage stats 2022-12-13 15:43:22 +01:00
Alex d7f90cabb0
Implement block retry-now and block purge 2022-12-13 15:02:42 +01:00
Alex 687660b27f
Implement block list-errors and block info 2022-12-13 14:23:45 +01:00
Alex 9d82196945
cli: new worker info command 2022-12-13 12:24:30 +01:00
Alex a51e8d94c6
cli: rename resync-n-workers into resync-worker-count 2022-12-13 11:44:11 +01:00
Alex de9d6cddf7
Prettier worker list table; remove useless CLI log messages 2022-12-12 17:17:05 +01:00
82 changed files with 2912 additions and 1945 deletions

1
.envrc Normal file
View file

@ -0,0 +1 @@
use flake

1
.gitignore vendored
View file

@ -3,3 +3,4 @@
/pki
**/*.rs.bk
*.swp
/.direnv

26
Cargo.lock generated
View file

@ -1048,7 +1048,7 @@ dependencies = [
[[package]]
name = "garage"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"assert-json-diff",
"async-trait",
@ -1080,7 +1080,6 @@ dependencies = [
"parse_duration",
"prometheus",
"rand 0.8.5",
"rmp-serde",
"serde",
"serde_bytes",
"serde_json",
@ -1096,7 +1095,7 @@ dependencies = [
[[package]]
name = "garage_api"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"async-trait",
"base64",
@ -1141,7 +1140,7 @@ dependencies = [
[[package]]
name = "garage_block"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"arc-swap",
"async-compression",
@ -1156,7 +1155,6 @@ dependencies = [
"hex",
"opentelemetry",
"rand 0.8.5",
"rmp-serde",
"serde",
"serde_bytes",
"tokio",
@ -1167,7 +1165,7 @@ dependencies = [
[[package]]
name = "garage_db"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"clap 3.1.18",
"err-derive",
@ -1182,7 +1180,7 @@ dependencies = [
[[package]]
name = "garage_model"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"arc-swap",
"async-trait",
@ -1200,7 +1198,6 @@ dependencies = [
"netapp",
"opentelemetry",
"rand 0.8.5",
"rmp-serde",
"serde",
"serde_bytes",
"tokio",
@ -1210,7 +1207,7 @@ dependencies = [
[[package]]
name = "garage_rpc"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"arc-swap",
"async-trait",
@ -1229,7 +1226,6 @@ dependencies = [
"pnet_datalink",
"rand 0.8.5",
"reqwest",
"rmp-serde",
"schemars",
"serde",
"serde_bytes",
@ -1241,8 +1237,9 @@ dependencies = [
[[package]]
name = "garage_table"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"arc-swap",
"async-trait",
"bytes",
"futures",
@ -1254,7 +1251,6 @@ dependencies = [
"hexdump",
"opentelemetry",
"rand 0.8.5",
"rmp-serde",
"serde",
"serde_bytes",
"tokio",
@ -1263,7 +1259,7 @@ dependencies = [
[[package]]
name = "garage_util"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"arc-swap",
"async-trait",
@ -1276,9 +1272,11 @@ dependencies = [
"garage_db",
"git-version",
"hex",
"hexdump",
"http",
"hyper",
"lazy_static",
"mktemp",
"netapp",
"opentelemetry",
"rand 0.8.5",
@ -1294,7 +1292,7 @@ dependencies = [
[[package]]
name = "garage_web"
version = "0.8.0"
version = "0.8.1"
dependencies = [
"err-derive",
"futures",

138
Cargo.nix
View file

@ -32,7 +32,7 @@ args@{
ignoreLockHash,
}:
let
nixifiedLockHash = "90b29705f5037c7e1b33f4650841f1266f2e86fa03d5d0c87ad80be7619985c7";
nixifiedLockHash = "8461dcfb984a8d042fecb5745d5da17912135dbf2a8ef7e6c3ae8e64c03d9744";
workspaceSrc = if args.workspaceSrc == null then ./. else args.workspaceSrc;
currentLockHash = builtins.hashFile "sha256" (workspaceSrc + /Cargo.lock);
lockHashIgnored = if ignoreLockHash
@ -56,15 +56,15 @@ in
{
cargo2nixVersion = "0.11.0";
workspace = {
garage_db = rustPackages.unknown.garage_db."0.8.0";
garage_util = rustPackages.unknown.garage_util."0.8.0";
garage_rpc = rustPackages.unknown.garage_rpc."0.8.0";
garage_table = rustPackages.unknown.garage_table."0.8.0";
garage_block = rustPackages.unknown.garage_block."0.8.0";
garage_model = rustPackages.unknown.garage_model."0.8.0";
garage_api = rustPackages.unknown.garage_api."0.8.0";
garage_web = rustPackages.unknown.garage_web."0.8.0";
garage = rustPackages.unknown.garage."0.8.0";
garage_db = rustPackages.unknown.garage_db."0.8.1";
garage_util = rustPackages.unknown.garage_util."0.8.1";
garage_rpc = rustPackages.unknown.garage_rpc."0.8.1";
garage_table = rustPackages.unknown.garage_table."0.8.1";
garage_block = rustPackages.unknown.garage_block."0.8.1";
garage_model = rustPackages.unknown.garage_model."0.8.1";
garage_api = rustPackages.unknown.garage_api."0.8.1";
garage_web = rustPackages.unknown.garage_web."0.8.1";
garage = rustPackages.unknown.garage."0.8.1";
k2v-client = rustPackages.unknown.k2v-client."0.0.1";
};
"registry+https://github.com/rust-lang/crates.io-index".addr2line."0.17.0" = overridableMkRustCrate (profileName: rec {
@ -1494,16 +1494,16 @@ in
};
});
"unknown".garage."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/garage");
features = builtins.concatLists [
(lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default") "bundled-libs")
(lib.optional (rootFeatures' ? "garage/consul-discovery") "consul-discovery")
(lib.optional (rootFeatures' ? "garage/default") "default")
(lib.optional (rootFeatures' ? "garage/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/kubernetes-discovery") "kubernetes-discovery")
(lib.optional (rootFeatures' ? "garage/lmdb") "lmdb")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics") "metrics")
@ -1522,14 +1522,14 @@ in
bytesize = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytesize."1.1.0" { inherit profileName; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_api = (rustPackages."unknown".garage_api."0.8.0" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.0" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.0" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.0" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.0" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_web = (rustPackages."unknown".garage_web."0.8.0" { inherit profileName; }).out;
garage_api = (rustPackages."unknown".garage_api."0.8.1" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.1" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.1" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.1" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.1" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
garage_web = (rustPackages."unknown".garage_web."0.8.1" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
sodiumoxide = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".kuska-sodiumoxide."0.2.5-0" { inherit profileName; }).out;
netapp = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".netapp."0.5.2" { inherit profileName; }).out;
@ -1539,7 +1539,6 @@ in
parse_duration = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".parse_duration."2.1.1" { inherit profileName; }).out;
${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage/prometheus" then "prometheus" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".prometheus."0.13.0" { inherit profileName; }).out;
rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.137" { inherit profileName; }).out;
serde_bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_bytes."0.11.5" { inherit profileName; }).out;
structopt = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".structopt."0.3.26" { inherit profileName; }).out;
@ -1563,13 +1562,13 @@ in
};
});
"unknown".garage_api."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_api."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_api";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/api");
features = builtins.concatLists [
(lib.optional (rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage_api/metrics") "metrics")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage_api/metrics" || rootFeatures' ? "garage_api/opentelemetry-prometheus") "opentelemetry-prometheus")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage_api/metrics" || rootFeatures' ? "garage_api/prometheus") "prometheus")
@ -1584,11 +1583,11 @@ in
form_urlencoded = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".form_urlencoded."1.0.1" { inherit profileName; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.0" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.0" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.0" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.1" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.1" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.1" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
hmac = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hmac."0.12.1" { inherit profileName; }).out;
http = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".http."0.2.8" { inherit profileName; }).out;
@ -1617,9 +1616,9 @@ in
};
});
"unknown".garage_block."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_block."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_block";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/block");
features = builtins.concatLists [
@ -1632,14 +1631,13 @@ in
bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytes."1.2.0" { inherit profileName; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.0" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.0" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.1" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.1" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
opentelemetry = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".opentelemetry."0.17.0" { inherit profileName; }).out;
rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.137" { inherit profileName; }).out;
serde_bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_bytes."0.11.5" { inherit profileName; }).out;
tokio = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".tokio."1.17.0" { inherit profileName; }).out;
@ -1649,9 +1647,9 @@ in
};
});
"unknown".garage_db."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_db."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_db";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/db");
features = builtins.concatLists [
@ -1681,14 +1679,14 @@ in
};
});
"unknown".garage_model."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_model."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_model";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/model");
features = builtins.concatLists [
[ "default" ]
(lib.optional (rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v" || rootFeatures' ? "garage_model/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v" || rootFeatures' ? "garage_model/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_model/lmdb") "lmdb")
[ "sled" ]
(lib.optional (rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_model/sqlite") "sqlite")
@ -1701,16 +1699,15 @@ in
err_derive = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.0" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.0" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.0" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_block = (rustPackages."unknown".garage_block."0.8.1" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.1" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.1" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
netapp = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".netapp."0.5.2" { inherit profileName; }).out;
opentelemetry = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".opentelemetry."0.17.0" { inherit profileName; }).out;
rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.137" { inherit profileName; }).out;
serde_bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_bytes."0.11.5" { inherit profileName; }).out;
tokio = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".tokio."1.17.0" { inherit profileName; }).out;
@ -1719,9 +1716,9 @@ in
};
});
"unknown".garage_rpc."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_rpc."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_rpc";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/rpc");
features = builtins.concatLists [
@ -1741,7 +1738,7 @@ in
${ if rootFeatures' ? "garage/consul-discovery" || rootFeatures' ? "garage_rpc/consul-discovery" || rootFeatures' ? "garage_rpc/err-derive" then "err_derive" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
gethostname = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".gethostname."0.2.3" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
${ if rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/k8s-openapi" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "k8s_openapi" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".k8s-openapi."0.16.0" { inherit profileName; }).out;
@ -1752,7 +1749,6 @@ in
pnet_datalink = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".pnet_datalink."0.28.0" { inherit profileName; }).out;
rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
${ if rootFeatures' ? "garage/consul-discovery" || rootFeatures' ? "garage_rpc/consul-discovery" || rootFeatures' ? "garage_rpc/reqwest" then "reqwest" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".reqwest."0.11.12" { inherit profileName; }).out;
rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
${ if rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/kubernetes-discovery" || rootFeatures' ? "garage_rpc/schemars" then "schemars" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".schemars."0.8.8" { inherit profileName; }).out;
serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.137" { inherit profileName; }).out;
serde_bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_bytes."0.11.5" { inherit profileName; }).out;
@ -1763,24 +1759,24 @@ in
};
});
"unknown".garage_table."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_table."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_table";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/table");
dependencies = {
arc_swap = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".arc-swap."1.5.0" { inherit profileName; }).out;
async_trait = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".async-trait."0.1.52" { profileName = "__noProfile"; }).out;
bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytes."1.2.0" { inherit profileName; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.21" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.0" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.1" { inherit profileName; }).out;
garage_rpc = (rustPackages."unknown".garage_rpc."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
hexdump = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hexdump."0.1.1" { inherit profileName; }).out;
opentelemetry = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".opentelemetry."0.17.0" { inherit profileName; }).out;
rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.137" { inherit profileName; }).out;
serde_bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_bytes."0.11.5" { inherit profileName; }).out;
tokio = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".tokio."1.17.0" { inherit profileName; }).out;
@ -1788,13 +1784,13 @@ in
};
});
"unknown".garage_util."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_util."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_util";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/util");
features = builtins.concatLists [
(lib.optional (rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v" || rootFeatures' ? "garage_model/k2v" || rootFeatures' ? "garage_util/k2v") "k2v")
(lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v" || rootFeatures' ? "garage_model/k2v" || rootFeatures' ? "garage_util/k2v") "k2v")
];
dependencies = {
arc_swap = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".arc-swap."1.5.0" { inherit profileName; }).out;
@ -1805,9 +1801,10 @@ in
digest = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".digest."0.10.3" { inherit profileName; }).out;
err_derive = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.0" { inherit profileName; }).out;
garage_db = (rustPackages."unknown".garage_db."0.8.1" { inherit profileName; }).out;
git_version = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".git-version."0.3.5" { inherit profileName; }).out;
hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
hexdump = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hexdump."0.1.1" { inherit profileName; }).out;
http = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".http."0.2.8" { inherit profileName; }).out;
hyper = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hyper."0.14.18" { inherit profileName; }).out;
lazy_static = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".lazy_static."1.4.0" { inherit profileName; }).out;
@ -1823,20 +1820,23 @@ in
tracing = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".tracing."0.1.32" { inherit profileName; }).out;
xxhash_rust = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".xxhash-rust."0.8.4" { inherit profileName; }).out;
};
devDependencies = {
mktemp = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".mktemp."0.4.1" { inherit profileName; }).out;
};
});
"unknown".garage_web."0.8.0" = overridableMkRustCrate (profileName: rec {
"unknown".garage_web."0.8.1" = overridableMkRustCrate (profileName: rec {
name = "garage_web";
version = "0.8.0";
version = "0.8.1";
registry = "unknown";
src = fetchCrateLocal (workspaceSrc + "/src/web");
dependencies = {
err_derive = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.21" { inherit profileName; }).out;
garage_api = (rustPackages."unknown".garage_api."0.8.0" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.0" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.0" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
garage_api = (rustPackages."unknown".garage_api."0.8.1" { inherit profileName; }).out;
garage_model = (rustPackages."unknown".garage_model."0.8.1" { inherit profileName; }).out;
garage_table = (rustPackages."unknown".garage_table."0.8.1" { inherit profileName; }).out;
garage_util = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
http = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".http."0.2.8" { inherit profileName; }).out;
hyper = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hyper."0.14.18" { inherit profileName; }).out;
opentelemetry = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".opentelemetry."0.17.0" { inherit profileName; }).out;
@ -2450,7 +2450,7 @@ in
dependencies = {
base64 = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".base64."0.13.0" { inherit profileName; }).out;
${ if rootFeatures' ? "k2v-client/clap" || rootFeatures' ? "k2v-client/cli" then "clap" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".clap."3.1.18" { inherit profileName; }).out;
${ if rootFeatures' ? "k2v-client/cli" || rootFeatures' ? "k2v-client/garage_util" then "garage_util" else null } = (rustPackages."unknown".garage_util."0.8.0" { inherit profileName; }).out;
${ if rootFeatures' ? "k2v-client/cli" || rootFeatures' ? "k2v-client/garage_util" then "garage_util" else null } = (rustPackages."unknown".garage_util."0.8.1" { inherit profileName; }).out;
http = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".http."0.2.8" { inherit profileName; }).out;
log = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".log."0.4.16" { inherit profileName; }).out;
rusoto_core = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rusoto_core."0.48.0" { inherit profileName; }).out;

View file

@ -5,16 +5,59 @@ weight = 20
## S3
### Using Minio SDK
First install the SDK:
```bash
pip3 install minio
```
Then instantiate a client object using garage root domain, api key and secret:
```python
import minio
client = minio.Minio(
"your.domain.tld",
"GKyourapikey",
"abcd[...]1234",
# Force the region, this is specific to garage
region="region",
)
```
Then use all the standard S3 endpoints as implemented by the Minio SDK:
```
# List buckets
print(client.list_buckets())
# Put an object containing 'content' to /path in bucket named 'bucket':
content = b"content"
client.put_object(
"bucket",
"path",
io.BytesIO(content),
len(content),
)
# Read the object back and check contents
data = client.get_object("bucket", "path").read()
assert data == content
```
For further documentation, see the Minio SDK
[Reference](https://docs.min.io/docs/python-client-api-reference.html)
### Using Amazon boto3
*Coming soon*
Some refs:
- Minio SDK
- [Reference](https://docs.min.io/docs/python-client-api-reference.html)
- Amazon boto3
- [Installation](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/quickstart.html)
- [Reference](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/s3.html)
- [Example](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/s3-uploading-files.html)
See the official documentation:
- [Installation](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/quickstart.html)
- [Reference](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/s3.html)
- [Example](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/s3-uploading-files.html)
## K2V

View file

@ -39,7 +39,7 @@ Now you can enter our nix-shell, all the required packages will be downloaded bu
nix-shell
```
You can use the traditionnal Rust development workflow:
You can use the traditional Rust development workflow:
```bash
cargo build # compile the project

View file

@ -96,7 +96,7 @@ Performance characteristics of the different DB engines are as follows:
- Sled: the default database engine, which tends to produce
large data files and also has performance issues, especially when the metadata folder
is on a traditionnal HDD and not on SSD.
is on a traditional HDD and not on SSD.
- LMDB: the recommended alternative on 64-bit systems,
much more space-efficiant and slightly faster. Note that the data format of LMDB is not portable
between architectures, so for instance the Garage database of an x86-64
@ -267,6 +267,10 @@ This key should be specified here in the form of a 32-byte hex-encoded
random string. Such a string can be generated with a command
such as `openssl rand -hex 32`.
### `rpc_secret_file`
Like `rpc_secret` above, just that this is the path to a file that Garage will try to read the secret from.
### `rpc_bind_addr`
The address and port on which to bind for inter-cluster communcations

View file

@ -55,6 +55,21 @@
"type": "github"
}
},
"flake-utils_2": {
"locked": {
"lastModified": 1667395993,
"narHash": "sha256-nuEHfE/LcWyuSWnS8t12N1wc105Qtau+/OdUAjtQ0rA=",
"owner": "numtide",
"repo": "flake-utils",
"rev": "5aed5285a952e0b949eb3ba02c12fa4fcfef535f",
"type": "github"
},
"original": {
"owner": "numtide",
"repo": "flake-utils",
"type": "github"
}
},
"nixpkgs": {
"locked": {
"lastModified": 1665657542,
@ -74,6 +89,7 @@
"root": {
"inputs": {
"cargo2nix": "cargo2nix",
"flake-utils": "flake-utils_2",
"nixpkgs": "nixpkgs"
}
},

View file

@ -7,22 +7,30 @@
url = "github:Alexis211/cargo2nix/a7a61179b66054904ef6a195d8da736eaaa06c36";
inputs.nixpkgs.follows = "nixpkgs";
};
inputs.flake-utils.url = "github:numtide/flake-utils";
outputs = { self, nixpkgs, cargo2nix }: let
git_version = self.lastModifiedDate;
compile = import ./nix/compile.nix;
forAllSystems = nixpkgs.lib.genAttrs nixpkgs.lib.systems.flakeExposed;
in
{
packages = forAllSystems (system: {
default = (compile {
inherit system git_version;
pkgsSrc = nixpkgs;
cargo2nixOverlay = cargo2nix.overlays.default;
release = true;
}).workspace.garage {
compileMode = "build";
};
});
};
outputs = { self, nixpkgs, cargo2nix, flake-utils }:
let
git_version = self.lastModifiedDate;
compile = import ./nix/compile.nix;
in flake-utils.lib.eachDefaultSystem (system:
let pkgs = nixpkgs.legacyPackages.${system};
in {
packages = {
default = (compile {
inherit system git_version;
pkgsSrc = nixpkgs;
cargo2nixOverlay = cargo2nix.overlays.default;
release = true;
}).workspace.garage { compileMode = "build"; };
};
devShell = ((compile {
inherit system git_version;
pkgsSrc = nixpkgs;
cargo2nixOverlay = cargo2nix.overlays.default;
release = false;
}).workspaceShell {
packages = [ pkgs.rustfmt cargo2nix.packages.${system}.default ];
});
});
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_api"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,11 +14,11 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_model = { version = "0.8.0", path = "../model" }
garage_table = { version = "0.8.0", path = "../table" }
garage_block = { version = "0.8.0", path = "../block" }
garage_util = { version = "0.8.0", path = "../util" }
garage_rpc = { version = "0.8.0", path = "../rpc" }
garage_model = { version = "0.8.1", path = "../model" }
garage_table = { version = "0.8.1", path = "../table" }
garage_block = { version = "0.8.1", path = "../block" }
garage_util = { version = "0.8.1", path = "../util" }
garage_rpc = { version = "0.8.1", path = "../rpc" }
async-trait = "0.1.7"
base64 = "0.13"

View file

@ -145,6 +145,7 @@ macro_rules! generateQueryParameters {
) => {
#[derive(Debug)]
#[allow(non_camel_case_types)]
#[allow(clippy::upper_case_acronyms)]
enum Keyword {
EMPTY,
$( $kw_name, )*

View file

@ -119,6 +119,17 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
return Ok((version_uuid, data_md5sum_hex));
}
// The following consists in many steps that can each fail.
// Keep track that some cleanup will be needed if things fail
// before everything is finished (cleanup is done using the Drop trait).
let mut interrupted_cleanup = InterruptedCleanup(Some((
garage.clone(),
bucket.id,
key.into(),
version_uuid,
version_timestamp,
)));
// Write version identifier in object table so that we have a trace
// that we are uploading something
let mut object_version = ObjectVersion {
@ -139,44 +150,27 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
// Transfer data and verify checksum
let first_block_hash = async_blake2sum(first_block.clone()).await;
let tx_result = (|| async {
let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
&garage,
&version,
1,
first_block,
first_block_hash,
&mut chunker,
)
.await?;
let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
&garage,
&version,
1,
first_block,
first_block_hash,
&mut chunker,
)
.await?;
ensure_checksum_matches(
data_md5sum.as_slice(),
data_sha256sum,
content_md5.as_deref(),
content_sha256,
)?;
ensure_checksum_matches(
data_md5sum.as_slice(),
data_sha256sum,
content_md5.as_deref(),
content_sha256,
)?;
check_quotas(&garage, bucket, key, total_size).await?;
Ok((total_size, data_md5sum))
})()
.await;
// If something went wrong, clean up
let (total_size, md5sum_arr) = match tx_result {
Ok(rv) => rv,
Err(e) => {
// Mark object as aborted, this will free the blocks further down
object_version.state = ObjectVersionState::Aborted;
let object = Object::new(bucket.id, key.into(), vec![object_version.clone()]);
garage.object_table.insert(&object).await?;
return Err(e);
}
};
check_quotas(&garage, bucket, key, total_size).await?;
// Save final object state, marked as Complete
let md5sum_hex = hex::encode(md5sum_arr);
let md5sum_hex = hex::encode(data_md5sum);
object_version.state = ObjectVersionState::Complete(ObjectVersionData::FirstBlock(
ObjectVersionMeta {
headers,
@ -188,6 +182,10 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
let object = Object::new(bucket.id, key.into(), vec![object_version]);
garage.object_table.insert(&object).await?;
// We were not interrupted, everything went fine.
// We won't have to clean up on drop.
interrupted_cleanup.cancel();
Ok((version_uuid, md5sum_hex))
}
@ -426,6 +424,33 @@ pub fn put_response(version_uuid: Uuid, md5sum_hex: String) -> Response<Body> {
.unwrap()
}
struct InterruptedCleanup(Option<(Arc<Garage>, Uuid, String, Uuid, u64)>);
impl InterruptedCleanup {
fn cancel(&mut self) {
drop(self.0.take());
}
}
impl Drop for InterruptedCleanup {
fn drop(&mut self) {
if let Some((garage, bucket_id, key, version_uuid, version_ts)) = self.0.take() {
tokio::spawn(async move {
let object_version = ObjectVersion {
uuid: version_uuid,
timestamp: version_ts,
state: ObjectVersionState::Aborted,
};
let object = Object::new(bucket_id, key, vec![object_version]);
if let Err(e) = garage.object_table.insert(&object).await {
warn!("Cannot cleanup after aborted PutObject: {}", e);
}
});
}
}
}
// ----
pub async fn handle_create_multipart_upload(
garage: Arc<Garage>,
req: &Request<Body>,

View file

@ -1,6 +1,6 @@
[package]
name = "garage_block"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,10 +14,10 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_db = { version = "0.8.0", path = "../db" }
garage_rpc = { version = "0.8.0", path = "../rpc" }
garage_util = { version = "0.8.0", path = "../util" }
garage_table = { version = "0.8.0", path = "../table" }
garage_db = { version = "0.8.1", path = "../db" }
garage_rpc = { version = "0.8.1", path = "../rpc" }
garage_util = { version = "0.8.1", path = "../util" }
garage_table = { version = "0.8.1", path = "../table" }
opentelemetry = "0.17"
@ -31,7 +31,6 @@ rand = "0.8"
async-compression = { version = "0.3", features = ["tokio", "zstd"] }
zstd = { version = "0.9", default-features = false }
rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"

View file

@ -3,8 +3,10 @@ use std::pin::Pin;
use std::sync::Arc;
use std::time::Duration;
use arc_swap::ArcSwapOption;
use async_trait::async_trait;
use bytes::Bytes;
use rand::prelude::*;
use serde::{Deserialize, Serialize};
use futures::Stream;
@ -22,9 +24,12 @@ use garage_rpc::rpc_helper::netapp::stream::{stream_asyncread, ByteStream};
use garage_db as db;
use garage_util::background::{vars, BackgroundRunner};
use garage_util::data::*;
use garage_util::error::*;
use garage_util::metrics::RecordDuration;
use garage_util::persister::PersisterShared;
use garage_util::time::msec_to_rfc3339;
use garage_rpc::rpc_helper::OrderTag;
use garage_rpc::system::System;
@ -87,7 +92,17 @@ pub struct BlockManager {
pub(crate) metrics: BlockManagerMetrics,
tx_scrub_command: mpsc::Sender<ScrubWorkerCommand>,
pub scrub_persister: PersisterShared<ScrubWorkerPersisted>,
tx_scrub_command: ArcSwapOption<mpsc::Sender<ScrubWorkerCommand>>,
}
#[derive(Serialize, Deserialize, Clone, Debug)]
pub struct BlockResyncErrorInfo {
pub hash: Hash,
pub refcount: u64,
pub error_count: u64,
pub last_try: u64,
pub next_try: u64,
}
// This custom struct contains functions that must only be ran
@ -114,9 +129,14 @@ impl BlockManager {
.netapp
.endpoint("garage_block/manager.rs/Rpc".to_string());
let metrics = BlockManagerMetrics::new(resync.queue.clone(), resync.errors.clone());
let metrics = BlockManagerMetrics::new(
compression_level,
rc.rc.clone(),
resync.queue.clone(),
resync.errors.clone(),
);
let (scrub_tx, scrub_rx) = mpsc::channel(1);
let scrub_persister = PersisterShared::new(&system.metadata_dir, "scrub_info");
let block_manager = Arc::new(Self {
replication,
@ -128,21 +148,46 @@ impl BlockManager {
system,
endpoint,
metrics,
tx_scrub_command: scrub_tx,
scrub_persister,
tx_scrub_command: ArcSwapOption::new(None),
});
block_manager.endpoint.set_handler(block_manager.clone());
block_manager
}
pub fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) {
// Spawn a bunch of resync workers
for index in 0..MAX_RESYNC_WORKERS {
let worker = ResyncWorker::new(index, block_manager.clone());
block_manager.system.background.spawn_worker(worker);
let worker = ResyncWorker::new(index, self.clone());
bg.spawn_worker(worker);
}
// Spawn scrub worker
let scrub_worker = ScrubWorker::new(block_manager.clone(), scrub_rx);
block_manager.system.background.spawn_worker(scrub_worker);
let (scrub_tx, scrub_rx) = mpsc::channel(1);
self.tx_scrub_command.store(Some(Arc::new(scrub_tx)));
bg.spawn_worker(ScrubWorker::new(
self.clone(),
scrub_rx,
self.scrub_persister.clone(),
));
}
block_manager
pub fn register_bg_vars(&self, vars: &mut vars::BgVars) {
self.resync.register_bg_vars(vars);
vars.register_rw(
&self.scrub_persister,
"scrub-tranquility",
|p| p.get_with(|x| x.tranquility),
|p, tranquility| p.set_with(|x| x.tranquility = tranquility),
);
vars.register_ro(&self.scrub_persister, "scrub-last-completed", |p| {
p.get_with(|x| msec_to_rfc3339(x.time_last_complete_scrub))
});
vars.register_ro(&self.scrub_persister, "scrub-corruptions_detected", |p| {
p.get_with(|x| x.corruptions_detected)
});
}
/// Ask nodes that might have a (possibly compressed) block for it
@ -309,9 +354,42 @@ impl BlockManager {
Ok(self.rc.rc.len()?)
}
/// Get number of items in the refcount table
pub fn rc_fast_len(&self) -> Result<Option<usize>, Error> {
Ok(self.rc.rc.fast_len()?)
}
/// Send command to start/stop/manager scrub worker
pub async fn send_scrub_command(&self, cmd: ScrubWorkerCommand) {
let _ = self.tx_scrub_command.send(cmd).await;
pub async fn send_scrub_command(&self, cmd: ScrubWorkerCommand) -> Result<(), Error> {
let tx = self.tx_scrub_command.load();
let tx = tx.as_ref().ok_or_message("scrub worker is not running")?;
tx.send(cmd).await.ok_or_message("send error")?;
Ok(())
}
/// Get the reference count of a block
pub fn get_block_rc(&self, hash: &Hash) -> Result<u64, Error> {
Ok(self.rc.get_block_rc(hash)?.as_u64())
}
/// List all resync errors
pub fn list_resync_errors(&self) -> Result<Vec<BlockResyncErrorInfo>, Error> {
let mut blocks = Vec::with_capacity(self.resync.errors.len());
for ent in self.resync.errors.iter()? {
let (hash, cnt) = ent?;
let cnt = ErrorCounter::decode(&cnt);
blocks.push(BlockResyncErrorInfo {
hash: Hash::try_from(&hash).unwrap(),
refcount: 0,
error_count: cnt.errors,
last_try: cnt.last_try,
next_try: cnt.next_try(),
});
}
for block in blocks.iter_mut() {
block.refcount = self.get_block_rc(&block.hash)?;
}
Ok(blocks)
}
//// ----- Managing the reference counter ----
@ -603,14 +681,21 @@ impl BlockManagerLocked {
}
};
let mut path2 = path.clone();
path2.set_extension("tmp");
let mut f = fs::File::create(&path2).await?;
let mut path_tmp = path.clone();
let tmp_extension = format!("tmp{}", hex::encode(thread_rng().gen::<[u8; 4]>()));
path_tmp.set_extension(tmp_extension);
let mut delete_on_drop = DeleteOnDrop(Some(path_tmp.clone()));
let mut f = fs::File::create(&path_tmp).await?;
f.write_all(data).await?;
f.sync_all().await?;
drop(f);
fs::rename(path2, path).await?;
fs::rename(path_tmp, path).await?;
delete_on_drop.cancel();
if let Some(to_delete) = to_delete {
fs::remove_file(to_delete).await?;
}
@ -676,3 +761,23 @@ async fn read_stream_to_end(mut stream: ByteStream) -> Result<Bytes, Error> {
.concat()
.into())
}
struct DeleteOnDrop(Option<PathBuf>);
impl DeleteOnDrop {
fn cancel(&mut self) {
drop(self.0.take());
}
}
impl Drop for DeleteOnDrop {
fn drop(&mut self) {
if let Some(path) = self.0.take() {
tokio::spawn(async move {
if let Err(e) = fs::remove_file(&path).await {
debug!("DeleteOnDrop failed for {}: {}", path.display(), e);
}
});
}
}
}

View file

@ -1,9 +1,12 @@
use opentelemetry::{global, metrics::*};
use garage_db as db;
use garage_db::counted_tree_hack::CountedTree;
/// TableMetrics reference all counter used for metrics
pub struct BlockManagerMetrics {
pub(crate) _compression_level: ValueObserver<u64>,
pub(crate) _rc_size: ValueObserver<u64>,
pub(crate) _resync_queue_len: ValueObserver<u64>,
pub(crate) _resync_errored_blocks: ValueObserver<u64>,
@ -23,9 +26,31 @@ pub struct BlockManagerMetrics {
}
impl BlockManagerMetrics {
pub fn new(resync_queue: CountedTree, resync_errors: CountedTree) -> Self {
pub fn new(
compression_level: Option<i32>,
rc_tree: db::Tree,
resync_queue: CountedTree,
resync_errors: CountedTree,
) -> Self {
let meter = global::meter("garage_model/block");
Self {
_compression_level: meter
.u64_value_observer("block.compression_level", move |observer| {
match compression_level {
Some(v) => observer.observe(v as u64, &[]),
None => observer.observe(0 as u64, &[]),
}
})
.with_description("Garage compression level for node")
.init(),
_rc_size: meter
.u64_value_observer("block.rc_size", move |observer| {
if let Ok(Some(v)) = rc_tree.fast_len() {
observer.observe(v as u64, &[])
}
})
.with_description("Number of blocks known to the reference counter")
.init(),
_resync_queue_len: meter
.u64_value_observer("block.resync_queue_length", move |observer| {
observer.observe(resync_queue.len() as u64, &[])

View file

@ -169,4 +169,11 @@ impl RcEntry {
pub(crate) fn is_needed(&self) -> bool {
!self.is_deletable()
}
pub(crate) fn as_u64(&self) -> u64 {
match self {
RcEntry::Present { count } => *count,
_ => 0,
}
}
}

View file

@ -13,7 +13,7 @@ use tokio::sync::watch;
use garage_util::background::*;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::persister::Persister;
use garage_util::persister::PersisterShared;
use garage_util::time::*;
use garage_util::tranquilizer::Tranquilizer;
@ -53,7 +53,7 @@ impl Worker for RepairWorker {
"Block repair worker".into()
}
fn info(&self) -> Option<String> {
fn status(&self) -> WorkerStatus {
match self.block_iter.as_ref() {
None => {
let idx_bytes = self
@ -66,9 +66,20 @@ impl Worker for RepairWorker {
} else {
idx_bytes
};
Some(format!("Phase 1: {}", hex::encode(idx_bytes)))
WorkerStatus {
progress: Some("0.00%".into()),
freeform: vec![format!(
"Currently in phase 1, iterator position: {}",
hex::encode(idx_bytes)
)],
..Default::default()
}
}
Some(bi) => Some(format!("Phase 2: {:.2}% done", bi.progress() * 100.)),
Some(bi) => WorkerStatus {
progress: Some(format!("{:.2}%", bi.progress() * 100.)),
freeform: vec!["Currently in phase 2".into()],
..Default::default()
},
}
}
@ -137,7 +148,7 @@ impl Worker for RepairWorker {
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
async fn wait_for_work(&mut self) -> WorkerState {
unreachable!()
}
}
@ -157,15 +168,24 @@ pub struct ScrubWorker {
work: ScrubWorkerState,
tranquilizer: Tranquilizer,
persister: Persister<ScrubWorkerPersisted>,
persisted: ScrubWorkerPersisted,
persister: PersisterShared<ScrubWorkerPersisted>,
}
#[derive(Serialize, Deserialize)]
struct ScrubWorkerPersisted {
tranquility: u32,
time_last_complete_scrub: u64,
corruptions_detected: u64,
pub struct ScrubWorkerPersisted {
pub tranquility: u32,
pub(crate) time_last_complete_scrub: u64,
pub(crate) corruptions_detected: u64,
}
impl garage_util::migrate::InitialFormat for ScrubWorkerPersisted {}
impl Default for ScrubWorkerPersisted {
fn default() -> Self {
ScrubWorkerPersisted {
time_last_complete_scrub: 0,
tranquility: INITIAL_SCRUB_TRANQUILITY,
corruptions_detected: 0,
}
}
}
enum ScrubWorkerState {
@ -186,27 +206,20 @@ pub enum ScrubWorkerCommand {
Pause(Duration),
Resume,
Cancel,
SetTranquility(u32),
}
impl ScrubWorker {
pub fn new(manager: Arc<BlockManager>, rx_cmd: mpsc::Receiver<ScrubWorkerCommand>) -> Self {
let persister = Persister::new(&manager.system.metadata_dir, "scrub_info");
let persisted = match persister.load() {
Ok(v) => v,
Err(_) => ScrubWorkerPersisted {
time_last_complete_scrub: 0,
tranquility: INITIAL_SCRUB_TRANQUILITY,
corruptions_detected: 0,
},
};
pub(crate) fn new(
manager: Arc<BlockManager>,
rx_cmd: mpsc::Receiver<ScrubWorkerCommand>,
persister: PersisterShared<ScrubWorkerPersisted>,
) -> Self {
Self {
manager,
rx_cmd,
work: ScrubWorkerState::Finished,
tranquilizer: Tranquilizer::new(30),
persister,
persisted,
}
}
@ -255,12 +268,6 @@ impl ScrubWorker {
}
}
}
ScrubWorkerCommand::SetTranquility(t) => {
self.persisted.tranquility = t;
if let Err(e) = self.persister.save_async(&self.persisted).await {
error!("Could not save new tranquilitiy value: {}", e);
}
}
}
}
}
@ -271,29 +278,37 @@ impl Worker for ScrubWorker {
"Block scrub worker".into()
}
fn info(&self) -> Option<String> {
let s = match &self.work {
ScrubWorkerState::Running(bsi) => format!(
"{:.2}% done (tranquility = {})",
bsi.progress() * 100.,
self.persisted.tranquility
),
ScrubWorkerState::Paused(bsi, rt) => {
format!(
"Paused, {:.2}% done, resumes at {}",
bsi.progress() * 100.,
msec_to_rfc3339(*rt)
fn status(&self) -> WorkerStatus {
let (corruptions_detected, tranquility, time_last_complete_scrub) =
self.persister.get_with(|p| {
(
p.corruptions_detected,
p.tranquility,
p.time_last_complete_scrub,
)
}
ScrubWorkerState::Finished => format!(
"Last completed scrub: {}",
msec_to_rfc3339(self.persisted.time_last_complete_scrub)
),
});
let mut s = WorkerStatus {
persistent_errors: Some(corruptions_detected),
tranquility: Some(tranquility),
..Default::default()
};
Some(format!(
"{} ; corruptions detected: {}",
s, self.persisted.corruptions_detected
))
match &self.work {
ScrubWorkerState::Running(bsi) => {
s.progress = Some(format!("{:.2}%", bsi.progress() * 100.));
}
ScrubWorkerState::Paused(bsi, rt) => {
s.progress = Some(format!("{:.2}%", bsi.progress() * 100.));
s.freeform = vec![format!("Scrub paused, resumes at {}", msec_to_rfc3339(*rt))];
}
ScrubWorkerState::Finished => {
s.freeform = vec![format!(
"Last scrub completed at {}",
msec_to_rfc3339(time_last_complete_scrub)
)];
}
}
s
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
@ -310,18 +325,17 @@ impl Worker for ScrubWorker {
match self.manager.read_block(&hash).await {
Err(Error::CorruptData(_)) => {
error!("Found corrupt data block during scrub: {:?}", hash);
self.persisted.corruptions_detected += 1;
self.persister.save_async(&self.persisted).await?;
self.persister.set_with(|p| p.corruptions_detected += 1)?;
}
Err(e) => return Err(e),
_ => (),
};
Ok(self
.tranquilizer
.tranquilize_worker(self.persisted.tranquility))
.tranquilize_worker(self.persister.get_with(|p| p.tranquility)))
} else {
self.persisted.time_last_complete_scrub = now_msec();
self.persister.save_async(&self.persisted).await?;
self.persister
.set_with(|p| p.time_last_complete_scrub = now_msec())?;
self.work = ScrubWorkerState::Finished;
self.tranquilizer.clear();
Ok(WorkerState::Idle)
@ -331,12 +345,13 @@ impl Worker for ScrubWorker {
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
async fn wait_for_work(&mut self) -> WorkerState {
let (wait_until, command) = match &self.work {
ScrubWorkerState::Running(_) => return WorkerState::Busy,
ScrubWorkerState::Paused(_, resume_time) => (*resume_time, ScrubWorkerCommand::Resume),
ScrubWorkerState::Finished => (
self.persisted.time_last_complete_scrub + SCRUB_INTERVAL.as_millis() as u64,
self.persister.get_with(|p| p.time_last_complete_scrub)
+ SCRUB_INTERVAL.as_millis() as u64,
ScrubWorkerCommand::Start,
),
};

View file

@ -3,7 +3,6 @@ use std::convert::TryInto;
use std::sync::{Arc, Mutex};
use std::time::Duration;
use arc_swap::ArcSwap;
use async_trait::async_trait;
use serde::{Deserialize, Serialize};
@ -22,7 +21,7 @@ use garage_util::background::*;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::metrics::RecordDuration;
use garage_util::persister::Persister;
use garage_util::persister::PersisterShared;
use garage_util::time::*;
use garage_util::tranquilizer::Tranquilizer;
@ -49,13 +48,12 @@ const INITIAL_RESYNC_TRANQUILITY: u32 = 2;
pub struct BlockResyncManager {
pub(crate) queue: CountedTree,
pub(crate) notify: Notify,
pub(crate) notify: Arc<Notify>,
pub(crate) errors: CountedTree,
busy_set: BusySet,
persister: Persister<ResyncPersistedConfig>,
persisted: ArcSwap<ResyncPersistedConfig>,
persister: PersisterShared<ResyncPersistedConfig>,
}
#[derive(Serialize, Deserialize, Clone, Copy)]
@ -63,6 +61,15 @@ struct ResyncPersistedConfig {
n_workers: usize,
tranquility: u32,
}
impl garage_util::migrate::InitialFormat for ResyncPersistedConfig {}
impl Default for ResyncPersistedConfig {
fn default() -> Self {
ResyncPersistedConfig {
n_workers: 1,
tranquility: INITIAL_RESYNC_TRANQUILITY,
}
}
}
enum ResyncIterResult {
BusyDidSomething,
@ -90,22 +97,14 @@ impl BlockResyncManager {
.expect("Unable to open block_local_resync_errors tree");
let errors = CountedTree::new(errors).expect("Could not count block_local_resync_errors");
let persister = Persister::new(&system.metadata_dir, "resync_cfg");
let persisted = match persister.load() {
Ok(v) => v,
Err(_) => ResyncPersistedConfig {
n_workers: 1,
tranquility: INITIAL_RESYNC_TRANQUILITY,
},
};
let persister = PersisterShared::new(&system.metadata_dir, "resync_cfg");
Self {
queue,
notify: Notify::new(),
notify: Arc::new(Notify::new()),
errors,
busy_set: Arc::new(Mutex::new(HashSet::new())),
persister,
persisted: ArcSwap::new(Arc::new(persisted)),
}
}
@ -123,6 +122,56 @@ impl BlockResyncManager {
Ok(self.errors.len())
}
/// Clear the error counter for a block and put it in queue immediately
pub fn clear_backoff(&self, hash: &Hash) -> Result<(), Error> {
let now = now_msec();
if let Some(ec) = self.errors.get(hash)? {
let mut ec = ErrorCounter::decode(&ec);
if ec.errors > 0 {
ec.last_try = now - ec.delay_msec();
self.errors.insert(hash, ec.encode())?;
self.put_to_resync_at(hash, now)?;
return Ok(());
}
}
Err(Error::Message(format!(
"Block {:?} was not in an errored state",
hash
)))
}
pub fn register_bg_vars(&self, vars: &mut vars::BgVars) {
let notify = self.notify.clone();
vars.register_rw(
&self.persister,
"resync-worker-count",
|p| p.get_with(|x| x.n_workers),
move |p, n_workers| {
if !(1..=MAX_RESYNC_WORKERS).contains(&n_workers) {
return Err(Error::Message(format!(
"Invalid number of resync workers, must be between 1 and {}",
MAX_RESYNC_WORKERS
)));
}
p.set_with(|x| x.n_workers = n_workers)?;
notify.notify_waiters();
Ok(())
},
);
let notify = self.notify.clone();
vars.register_rw(
&self.persister,
"resync-tranquility",
|p| p.get_with(|x| x.tranquility),
move |p, tranquility| {
p.set_with(|x| x.tranquility = tranquility)?;
notify.notify_waiters();
Ok(())
},
);
}
// ---- Resync loop ----
// This part manages a queue of blocks that need to be
@ -257,7 +306,7 @@ impl BlockResyncManager {
if let Err(e) = &res {
manager.metrics.resync_error_counter.add(1);
warn!("Error when resyncing {:?}: {}", hash, e);
error!("Error when resyncing {:?}: {}", hash, e);
let err_counter = match self.errors.get(hash.as_slice())? {
Some(ec) => ErrorCounter::decode(&ec).add1(now + 1),
@ -417,33 +466,6 @@ impl BlockResyncManager {
Ok(())
}
async fn update_persisted(
&self,
update: impl Fn(&mut ResyncPersistedConfig),
) -> Result<(), Error> {
let mut cfg: ResyncPersistedConfig = *self.persisted.load().as_ref();
update(&mut cfg);
self.persister.save_async(&cfg).await?;
self.persisted.store(Arc::new(cfg));
self.notify.notify_waiters();
Ok(())
}
pub async fn set_n_workers(&self, n_workers: usize) -> Result<(), Error> {
if !(1..=MAX_RESYNC_WORKERS).contains(&n_workers) {
return Err(Error::Message(format!(
"Invalid number of resync workers, must be between 1 and {}",
MAX_RESYNC_WORKERS
)));
}
self.update_persisted(|cfg| cfg.n_workers = n_workers).await
}
pub async fn set_tranquility(&self, tranquility: u32) -> Result<(), Error> {
self.update_persisted(|cfg| cfg.tranquility = tranquility)
.await
}
}
impl Drop for BusyBlock {
@ -458,15 +480,18 @@ pub(crate) struct ResyncWorker {
manager: Arc<BlockManager>,
tranquilizer: Tranquilizer,
next_delay: Duration,
persister: PersisterShared<ResyncPersistedConfig>,
}
impl ResyncWorker {
pub(crate) fn new(index: usize, manager: Arc<BlockManager>) -> Self {
let persister = manager.resync.persister.clone();
Self {
index,
manager,
tranquilizer: Tranquilizer::new(30),
next_delay: Duration::from_secs(10),
persister,
}
}
}
@ -477,39 +502,36 @@ impl Worker for ResyncWorker {
format!("Block resync worker #{}", self.index + 1)
}
fn info(&self) -> Option<String> {
let persisted = self.manager.resync.persisted.load();
fn status(&self) -> WorkerStatus {
let (n_workers, tranquility) = self.persister.get_with(|x| (x.n_workers, x.tranquility));
if self.index >= persisted.n_workers {
return Some("(unused)".into());
if self.index >= n_workers {
return WorkerStatus {
freeform: vec!["This worker is currently disabled".into()],
..Default::default()
};
}
let mut ret = vec![];
ret.push(format!("tranquility = {}", persisted.tranquility));
let qlen = self.manager.resync.queue_len().unwrap_or(0);
if qlen > 0 {
ret.push(format!("{} blocks in queue", qlen));
WorkerStatus {
queue_length: Some(self.manager.resync.queue_len().unwrap_or(0) as u64),
tranquility: Some(tranquility),
persistent_errors: Some(self.manager.resync.errors_len().unwrap_or(0) as u64),
..Default::default()
}
let elen = self.manager.resync.errors_len().unwrap_or(0);
if elen > 0 {
ret.push(format!("{} blocks in error state", elen));
}
Some(ret.join(", "))
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
if self.index >= self.manager.resync.persisted.load().n_workers {
let (n_workers, tranquility) = self.persister.get_with(|x| (x.n_workers, x.tranquility));
if self.index >= n_workers {
return Ok(WorkerState::Idle);
}
self.tranquilizer.reset();
match self.manager.resync.resync_iter(&self.manager).await {
Ok(ResyncIterResult::BusyDidSomething) => Ok(self
.tranquilizer
.tranquilize_worker(self.manager.resync.persisted.load().tranquility)),
Ok(ResyncIterResult::BusyDidSomething) => {
Ok(self.tranquilizer.tranquilize_worker(tranquility))
}
Ok(ResyncIterResult::BusyDidNothing) => Ok(WorkerState::Busy),
Ok(ResyncIterResult::IdleFor(delay)) => {
self.next_delay = delay;
@ -527,8 +549,8 @@ impl Worker for ResyncWorker {
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
while self.index >= self.manager.resync.persisted.load().n_workers {
async fn wait_for_work(&mut self) -> WorkerState {
while self.index >= self.persister.get_with(|x| x.n_workers) {
self.manager.resync.notify.notified().await
}
@ -545,9 +567,9 @@ impl Worker for ResyncWorker {
/// and the time of the last try.
/// Used to implement exponential backoff.
#[derive(Clone, Copy, Debug)]
struct ErrorCounter {
errors: u64,
last_try: u64,
pub(crate) struct ErrorCounter {
pub(crate) errors: u64,
pub(crate) last_try: u64,
}
impl ErrorCounter {
@ -558,12 +580,13 @@ impl ErrorCounter {
}
}
fn decode(data: &[u8]) -> Self {
pub(crate) fn decode(data: &[u8]) -> Self {
Self {
errors: u64::from_be_bytes(data[0..8].try_into().unwrap()),
last_try: u64::from_be_bytes(data[8..16].try_into().unwrap()),
}
}
fn encode(&self) -> Vec<u8> {
[
u64::to_be_bytes(self.errors),
@ -583,7 +606,8 @@ impl ErrorCounter {
(RESYNC_RETRY_DELAY.as_millis() as u64)
<< std::cmp::min(self.errors - 1, RESYNC_RETRY_DELAY_MAX_BACKOFF_POWER)
}
fn next_try(&self) -> u64 {
pub(crate) fn next_try(&self) -> u64 {
self.last_try + self.delay_msec()
}
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_db"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"

View file

@ -181,6 +181,10 @@ impl Tree {
pub fn len(&self) -> Result<usize> {
self.0.len(self.1)
}
#[inline]
pub fn fast_len(&self) -> Result<Option<usize>> {
self.0.fast_len(self.1)
}
#[inline]
pub fn first(&self) -> Result<Option<(Value, Value)>> {
@ -323,6 +327,9 @@ pub(crate) trait IDb: Send + Sync {
fn get(&self, tree: usize, key: &[u8]) -> Result<Option<Value>>;
fn len(&self, tree: usize) -> Result<usize>;
fn fast_len(&self, _tree: usize) -> Result<Option<usize>> {
Ok(None)
}
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>>;
fn remove(&self, tree: usize, key: &[u8]) -> Result<Option<Value>>;

View file

@ -121,6 +121,10 @@ impl IDb for LmdbDb {
Ok(tree.len(&tx)?.try_into().unwrap())
}
fn fast_len(&self, tree: usize) -> Result<Option<usize>> {
Ok(Some(self.len(tree)?))
}
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>> {
let tree = self.get_tree(tree)?;
let mut tx = self.db.write_txn()?;

View file

@ -144,6 +144,10 @@ impl IDb for SqliteDb {
}
}
fn fast_len(&self, tree: usize) -> Result<Option<usize>> {
Ok(Some(self.len(tree)?))
}
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>> {
trace!("insert {}: lock db", tree);
let this = self.0.lock().unwrap();

View file

@ -1,6 +1,6 @@
[package]
name = "garage"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -21,14 +21,14 @@ path = "tests/lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_db = { version = "0.8.0", path = "../db" }
garage_api = { version = "0.8.0", path = "../api" }
garage_block = { version = "0.8.0", path = "../block" }
garage_model = { version = "0.8.0", path = "../model" }
garage_rpc = { version = "0.8.0", path = "../rpc" }
garage_table = { version = "0.8.0", path = "../table" }
garage_util = { version = "0.8.0", path = "../util" }
garage_web = { version = "0.8.0", path = "../web" }
garage_db = { version = "0.8.1", path = "../db" }
garage_api = { version = "0.8.1", path = "../api" }
garage_block = { version = "0.8.1", path = "../block" }
garage_model = { version = "0.8.1", path = "../model" }
garage_rpc = { version = "0.8.1", path = "../rpc" }
garage_table = { version = "0.8.1", path = "../table" }
garage_util = { version = "0.8.1", path = "../util" }
garage_web = { version = "0.8.1", path = "../web" }
backtrace = "0.3"
bytes = "1.0"
@ -42,7 +42,6 @@ rand = "0.8"
async-trait = "0.1.7"
sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"
structopt = { version = "0.3", default-features = false }
@ -74,7 +73,7 @@ base64 = "0.13"
[features]
default = [ "bundled-libs", "metrics", "sled" ]
default = [ "bundled-libs", "metrics", "sled", "k2v" ]
k2v = [ "garage_util/k2v", "garage_api/k2v" ]

View file

@ -5,9 +5,11 @@ use std::sync::Arc;
use async_trait::async_trait;
use serde::{Deserialize, Serialize};
use garage_util::background::BackgroundRunner;
use garage_util::crdt::*;
use garage_util::data::*;
use garage_util::error::Error as GarageError;
use garage_util::formater::format_table_to_string;
use garage_util::time::*;
use garage_table::replication::*;
@ -15,7 +17,7 @@ use garage_table::*;
use garage_rpc::*;
use garage_block::repair::ScrubWorkerCommand;
use garage_block::manager::BlockResyncErrorInfo;
use garage_model::bucket_alias_table::*;
use garage_model::bucket_table::*;
@ -24,6 +26,8 @@ use garage_model::helper::error::{Error, OkOrBadRequest};
use garage_model::key_table::*;
use garage_model::migrate::Migrate;
use garage_model::permission::*;
use garage_model::s3::object_table::*;
use garage_model::s3::version_table::Version;
use crate::cli::*;
use crate::repair::online::launch_online_repair;
@ -38,7 +42,8 @@ pub enum AdminRpc {
LaunchRepair(RepairOpt),
Migrate(MigrateOpt),
Stats(StatsOpt),
Worker(WorkerOpt),
Worker(WorkerOperation),
BlockOperation(BlockOperation),
// Replies
Ok(String),
@ -54,6 +59,14 @@ pub enum AdminRpc {
HashMap<usize, garage_util::background::WorkerInfo>,
WorkerListOpt,
),
WorkerVars(Vec<(Uuid, String, String)>),
WorkerInfo(usize, garage_util::background::WorkerInfo),
BlockErrorList(Vec<BlockResyncErrorInfo>),
BlockInfo {
hash: Hash,
refcount: u64,
versions: Vec<Result<Version, Uuid>>,
},
}
impl Rpc for AdminRpc {
@ -62,17 +75,24 @@ impl Rpc for AdminRpc {
pub struct AdminRpcHandler {
garage: Arc<Garage>,
background: Arc<BackgroundRunner>,
endpoint: Arc<Endpoint<AdminRpc, Self>>,
}
impl AdminRpcHandler {
pub fn new(garage: Arc<Garage>) -> Arc<Self> {
pub fn new(garage: Arc<Garage>, background: Arc<BackgroundRunner>) -> Arc<Self> {
let endpoint = garage.system.netapp.endpoint(ADMIN_RPC_PATH.into());
let admin = Arc::new(Self { garage, endpoint });
let admin = Arc::new(Self {
garage,
background,
endpoint,
});
admin.endpoint.set_handler(admin.clone());
admin
}
// ================ BUCKET COMMANDS ====================
async fn handle_bucket_cmd(&self, cmd: &BucketOperation) -> Result<AdminRpc, Error> {
match cmd {
BucketOperation::List => self.handle_list_buckets().await,
@ -551,6 +571,8 @@ impl AdminRpcHandler {
Ok(AdminRpc::Ok(ret))
}
// ================ KEY COMMANDS ====================
async fn handle_key_cmd(&self, cmd: &KeyOperation) -> Result<AdminRpc, Error> {
match cmd {
KeyOperation::List => self.handle_list_keys().await,
@ -688,6 +710,8 @@ impl AdminRpcHandler {
Ok(AdminRpc::KeyInfo(key, relevant_buckets))
}
// ================ MIGRATION COMMANDS ====================
async fn handle_migrate(self: &Arc<Self>, opt: MigrateOpt) -> Result<AdminRpc, Error> {
if !opt.yes {
return Err(Error::BadRequest(
@ -704,6 +728,8 @@ impl AdminRpcHandler {
Ok(AdminRpc::Ok("Migration successfull.".into()))
}
// ================ REPAIR COMMANDS ====================
async fn handle_launch_repair(self: &Arc<Self>, opt: RepairOpt) -> Result<AdminRpc, Error> {
if !opt.yes {
return Err(Error::BadRequest(
@ -739,7 +765,7 @@ impl AdminRpcHandler {
)))
}
} else {
launch_online_repair(self.garage.clone(), opt).await;
launch_online_repair(&self.garage, &self.background, opt).await?;
Ok(AdminRpc::Ok(format!(
"Repair launched on {:?}",
self.garage.system.id
@ -747,6 +773,8 @@ impl AdminRpcHandler {
}
}
// ================ STATS COMMANDS ====================
async fn handle_stats(&self, opt: StatsOpt) -> Result<AdminRpc, Error> {
if opt.all_nodes {
let mut ret = String::new();
@ -763,11 +791,12 @@ impl AdminRpcHandler {
match self
.endpoint
.call(&node_id, AdminRpc::Stats(opt), PRIO_NORMAL)
.await?
.await
{
Ok(AdminRpc::Ok(s)) => writeln!(&mut ret, "{}", s).unwrap(),
Ok(x) => writeln!(&mut ret, "Bad answer: {:?}", x).unwrap(),
Err(e) => writeln!(&mut ret, "Error: {}", e).unwrap(),
Ok(Ok(AdminRpc::Ok(s))) => writeln!(&mut ret, "{}", s).unwrap(),
Ok(Ok(x)) => writeln!(&mut ret, "Bad answer: {:?}", x).unwrap(),
Ok(Err(e)) => writeln!(&mut ret, "Remote error: {}", e).unwrap(),
Err(e) => writeln!(&mut ret, "Network error: {}", e).unwrap(),
}
}
Ok(AdminRpc::Ok(ret))
@ -787,6 +816,7 @@ impl AdminRpcHandler {
.unwrap_or_else(|| "(unknown)".into()),
)
.unwrap();
writeln!(&mut ret, "\nDatabase engine: {}", self.garage.db.engine()).unwrap();
// Gather ring statistics
@ -805,21 +835,38 @@ impl AdminRpcHandler {
writeln!(&mut ret, " {:?} {}", n, c).unwrap();
}
self.gather_table_stats(&mut ret, &self.garage.bucket_table, &opt)?;
self.gather_table_stats(&mut ret, &self.garage.key_table, &opt)?;
self.gather_table_stats(&mut ret, &self.garage.object_table, &opt)?;
self.gather_table_stats(&mut ret, &self.garage.version_table, &opt)?;
self.gather_table_stats(&mut ret, &self.garage.block_ref_table, &opt)?;
// Gather table statistics
let mut table = vec![" Table\tItems\tMklItems\tMklTodo\tGcTodo".into()];
table.push(self.gather_table_stats(&self.garage.bucket_table, opt.detailed)?);
table.push(self.gather_table_stats(&self.garage.key_table, opt.detailed)?);
table.push(self.gather_table_stats(&self.garage.object_table, opt.detailed)?);
table.push(self.gather_table_stats(&self.garage.version_table, opt.detailed)?);
table.push(self.gather_table_stats(&self.garage.block_ref_table, opt.detailed)?);
write!(
&mut ret,
"\nTable stats:\n{}",
format_table_to_string(table)
)
.unwrap();
// Gather block manager statistics
writeln!(&mut ret, "\nBlock manager stats:").unwrap();
if opt.detailed {
writeln!(
&mut ret,
" number of RC entries (~= number of blocks): {}",
self.garage.block_manager.rc_len()?
)
.unwrap();
}
let rc_len = if opt.detailed {
self.garage.block_manager.rc_len()?.to_string()
} else {
self.garage
.block_manager
.rc_fast_len()?
.map(|x| x.to_string())
.unwrap_or_else(|| "NC".into())
};
writeln!(
&mut ret,
" number of RC entries (~= number of blocks): {}",
rc_len
)
.unwrap();
writeln!(
&mut ret,
" resync queue length: {}",
@ -833,79 +880,305 @@ impl AdminRpcHandler {
)
.unwrap();
if !opt.detailed {
writeln!(&mut ret, "\nIf values are missing (marked as NC), consider adding the --detailed flag - this will be slow.").unwrap();
}
Ok(ret)
}
fn gather_table_stats<F, R>(
&self,
to: &mut String,
t: &Arc<Table<F, R>>,
opt: &StatsOpt,
) -> Result<(), Error>
detailed: bool,
) -> Result<String, Error>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
writeln!(to, "\nTable stats for {}", F::TABLE_NAME).unwrap();
if opt.detailed {
writeln!(
to,
" number of items: {}",
t.data.store.len().map_err(GarageError::from)?
let (data_len, mkl_len) = if detailed {
(
t.data.store.len().map_err(GarageError::from)?.to_string(),
t.merkle_updater.merkle_tree_len()?.to_string(),
)
.unwrap();
writeln!(
to,
" Merkle tree size: {}",
t.merkle_updater.merkle_tree_len()?
} else {
(
t.data
.store
.fast_len()
.map_err(GarageError::from)?
.map(|x| x.to_string())
.unwrap_or_else(|| "NC".into()),
t.merkle_updater
.merkle_tree_fast_len()?
.map(|x| x.to_string())
.unwrap_or_else(|| "NC".into()),
)
.unwrap();
}
writeln!(
to,
" Merkle updater todo queue length: {}",
t.merkle_updater.todo_len()?
)
.unwrap();
writeln!(to, " GC todo queue length: {}", t.data.gc_todo_len()?).unwrap();
};
Ok(())
Ok(format!(
" {}\t{}\t{}\t{}\t{}",
F::TABLE_NAME,
data_len,
mkl_len,
t.merkle_updater.todo_len()?,
t.data.gc_todo_len()?
))
}
// ----
// ================ WORKER COMMANDS ====================
async fn handle_worker_cmd(&self, opt: WorkerOpt) -> Result<AdminRpc, Error> {
match opt.cmd {
WorkerCmd::List { opt } => {
let workers = self.garage.background.get_worker_info();
Ok(AdminRpc::WorkerList(workers, opt))
async fn handle_worker_cmd(&self, cmd: &WorkerOperation) -> Result<AdminRpc, Error> {
match cmd {
WorkerOperation::List { opt } => {
let workers = self.background.get_worker_info();
Ok(AdminRpc::WorkerList(workers, *opt))
}
WorkerCmd::Set { opt } => match opt {
WorkerSetCmd::ScrubTranquility { tranquility } => {
let scrub_command = ScrubWorkerCommand::SetTranquility(tranquility);
self.garage
.block_manager
.send_scrub_command(scrub_command)
.await;
Ok(AdminRpc::Ok("Scrub tranquility updated".into()))
WorkerOperation::Info { tid } => {
let info = self
.background
.get_worker_info()
.get(tid)
.ok_or_bad_request(format!("No worker with TID {}", tid))?
.clone();
Ok(AdminRpc::WorkerInfo(*tid, info))
}
WorkerOperation::Get {
all_nodes,
variable,
} => self.handle_get_var(*all_nodes, variable).await,
WorkerOperation::Set {
all_nodes,
variable,
value,
} => self.handle_set_var(*all_nodes, variable, value).await,
}
}
async fn handle_get_var(
&self,
all_nodes: bool,
variable: &Option<String>,
) -> Result<AdminRpc, Error> {
if all_nodes {
let mut ret = vec![];
let ring = self.garage.system.ring.borrow().clone();
for node in ring.layout.node_ids().iter() {
let node = (*node).into();
match self
.endpoint
.call(
&node,
AdminRpc::Worker(WorkerOperation::Get {
all_nodes: false,
variable: variable.clone(),
}),
PRIO_NORMAL,
)
.await??
{
AdminRpc::WorkerVars(v) => ret.extend(v),
m => return Err(GarageError::unexpected_rpc_message(m).into()),
}
WorkerSetCmd::ResyncNWorkers { n_workers } => {
self.garage
.block_manager
.resync
.set_n_workers(n_workers)
}
Ok(AdminRpc::WorkerVars(ret))
} else {
#[allow(clippy::collapsible_else_if)]
if let Some(v) = variable {
Ok(AdminRpc::WorkerVars(vec![(
self.garage.system.id,
v.clone(),
self.garage.bg_vars.get(v)?,
)]))
} else {
let mut vars = self.garage.bg_vars.get_all();
vars.sort();
Ok(AdminRpc::WorkerVars(
vars.into_iter()
.map(|(k, v)| (self.garage.system.id, k.to_string(), v))
.collect(),
))
}
}
}
async fn handle_set_var(
&self,
all_nodes: bool,
variable: &str,
value: &str,
) -> Result<AdminRpc, Error> {
if all_nodes {
let mut ret = vec![];
let ring = self.garage.system.ring.borrow().clone();
for node in ring.layout.node_ids().iter() {
let node = (*node).into();
match self
.endpoint
.call(
&node,
AdminRpc::Worker(WorkerOperation::Set {
all_nodes: false,
variable: variable.to_string(),
value: value.to_string(),
}),
PRIO_NORMAL,
)
.await??
{
AdminRpc::WorkerVars(v) => ret.extend(v),
m => return Err(GarageError::unexpected_rpc_message(m).into()),
}
}
Ok(AdminRpc::WorkerVars(ret))
} else {
self.garage.bg_vars.set(variable, value)?;
Ok(AdminRpc::WorkerVars(vec![(
self.garage.system.id,
variable.to_string(),
value.to_string(),
)]))
}
}
// ================ BLOCK COMMANDS ====================
async fn handle_block_cmd(&self, cmd: &BlockOperation) -> Result<AdminRpc, Error> {
match cmd {
BlockOperation::ListErrors => Ok(AdminRpc::BlockErrorList(
self.garage.block_manager.list_resync_errors()?,
)),
BlockOperation::Info { hash } => {
let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?;
let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?;
let refcount = self.garage.block_manager.get_block_rc(&hash)?;
let block_refs = self
.garage
.block_ref_table
.get_range(&hash, None, None, 10000, Default::default())
.await?;
let mut versions = vec![];
for br in block_refs {
if let Some(v) = self
.garage
.version_table
.get(&br.version, &EmptyKey)
.await?
{
versions.push(Ok(v));
} else {
versions.push(Err(br.version));
}
}
Ok(AdminRpc::BlockInfo {
hash,
refcount,
versions,
})
}
BlockOperation::RetryNow { all, blocks } => {
if *all {
if !blocks.is_empty() {
return Err(Error::BadRequest(
"--all was specified, cannot also specify blocks".into(),
));
}
let blocks = self.garage.block_manager.list_resync_errors()?;
for b in blocks.iter() {
self.garage.block_manager.resync.clear_backoff(&b.hash)?;
}
Ok(AdminRpc::Ok(format!(
"{} blocks returned in queue for a retry now (check logs to see results)",
blocks.len()
)))
} else {
for hash in blocks {
let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?;
let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?;
self.garage.block_manager.resync.clear_backoff(&hash)?;
}
Ok(AdminRpc::Ok(format!(
"{} blocks returned in queue for a retry now (check logs to see results)",
blocks.len()
)))
}
}
BlockOperation::Purge { yes, blocks } => {
if !yes {
return Err(Error::BadRequest(
"Pass the --yes flag to confirm block purge operation.".into(),
));
}
let mut obj_dels = 0;
let mut ver_dels = 0;
for hash in blocks {
let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?;
let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?;
let block_refs = self
.garage
.block_ref_table
.get_range(&hash, None, None, 10000, Default::default())
.await?;
Ok(AdminRpc::Ok("Number of resync workers updated".into()))
for br in block_refs {
let version = match self
.garage
.version_table
.get(&br.version, &EmptyKey)
.await?
{
Some(v) => v,
None => continue,
};
if let Some(object) = self
.garage
.object_table
.get(&version.bucket_id, &version.key)
.await?
{
let ov = object.versions().iter().rev().find(|v| v.is_complete());
if let Some(ov) = ov {
if ov.uuid == br.version {
let del_uuid = gen_uuid();
let deleted_object = Object::new(
version.bucket_id,
version.key.clone(),
vec![ObjectVersion {
uuid: del_uuid,
timestamp: ov.timestamp + 1,
state: ObjectVersionState::Complete(
ObjectVersionData::DeleteMarker,
),
}],
);
self.garage.object_table.insert(&deleted_object).await?;
obj_dels += 1;
}
}
}
if !version.deleted.get() {
let deleted_version = Version::new(
version.uuid,
version.bucket_id,
version.key.clone(),
true,
);
self.garage.version_table.insert(&deleted_version).await?;
ver_dels += 1;
}
}
}
WorkerSetCmd::ResyncTranquility { tranquility } => {
self.garage
.block_manager
.resync
.set_tranquility(tranquility)
.await?;
Ok(AdminRpc::Ok("Resync tranquility updated".into()))
}
},
Ok(AdminRpc::Ok(format!(
"{} blocks were purged: {} object deletion markers added, {} versions marked deleted",
blocks.len(),
obj_dels,
ver_dels
)))
}
}
}
}
@ -923,7 +1196,8 @@ impl EndpointHandler<AdminRpc> for AdminRpcHandler {
AdminRpc::Migrate(opt) => self.handle_migrate(opt.clone()).await,
AdminRpc::LaunchRepair(opt) => self.handle_launch_repair(opt.clone()).await,
AdminRpc::Stats(opt) => self.handle_stats(opt.clone()).await,
AdminRpc::Worker(opt) => self.handle_worker_cmd(opt.clone()).await,
AdminRpc::Worker(wo) => self.handle_worker_cmd(wo).await,
AdminRpc::BlockOperation(bo) => self.handle_block_cmd(bo).await,
m => Err(GarageError::unexpected_rpc_message(m).into()),
}
}

View file

@ -41,6 +41,9 @@ pub async fn cli_command_dispatch(
}
Command::Stats(so) => cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::Stats(so)).await,
Command::Worker(wo) => cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::Worker(wo)).await,
Command::Block(bo) => {
cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::BlockOperation(bo)).await
}
_ => unreachable!(),
}
}
@ -186,7 +189,23 @@ pub async fn cmd_admin(
print_key_info(&key, &rb);
}
AdminRpc::WorkerList(wi, wlo) => {
print_worker_info(wi, wlo);
print_worker_list(wi, wlo);
}
AdminRpc::WorkerVars(wv) => {
print_worker_vars(wv);
}
AdminRpc::WorkerInfo(tid, wi) => {
print_worker_info(tid, wi);
}
AdminRpc::BlockErrorList(el) => {
print_block_error_list(el);
}
AdminRpc::BlockInfo {
hash,
refcount,
versions,
} => {
print_block_info(hash, refcount, versions);
}
r => {
error!("Unexpected response: {:?}", r);

View file

@ -49,7 +49,11 @@ pub enum Command {
/// Manage background workers
#[structopt(name = "worker", version = garage_version())]
Worker(WorkerOpt),
Worker(WorkerOperation),
/// Low-level debug operations on data blocks
#[structopt(name = "block", version = garage_version())]
Block(BlockOperation),
}
#[derive(StructOpt, Debug)]
@ -502,25 +506,36 @@ pub struct StatsOpt {
pub detailed: bool,
}
#[derive(Serialize, Deserialize, StructOpt, Debug, Clone)]
pub struct WorkerOpt {
#[structopt(subcommand)]
pub cmd: WorkerCmd,
}
#[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
pub enum WorkerCmd {
pub enum WorkerOperation {
/// List all workers on Garage node
#[structopt(name = "list", version = garage_version())]
List {
#[structopt(flatten)]
opt: WorkerListOpt,
},
/// Get detailed information about a worker
#[structopt(name = "info", version = garage_version())]
Info { tid: usize },
/// Get worker parameter
#[structopt(name = "get", version = garage_version())]
Get {
/// Gather variable values from all nodes
#[structopt(short = "a", long = "all-nodes")]
all_nodes: bool,
/// Variable name to get, or none to get all variables
variable: Option<String>,
},
/// Set worker parameter
#[structopt(name = "set", version = garage_version())]
Set {
#[structopt(subcommand)]
opt: WorkerSetCmd,
/// Set variable values on all nodes
#[structopt(short = "a", long = "all-nodes")]
all_nodes: bool,
/// Variable node to set
variable: String,
/// Value to set the variable to
value: String,
},
}
@ -535,14 +550,33 @@ pub struct WorkerListOpt {
}
#[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
pub enum WorkerSetCmd {
/// Set tranquility of scrub operations
#[structopt(name = "scrub-tranquility", version = garage_version())]
ScrubTranquility { tranquility: u32 },
/// Set number of concurrent block resync workers
#[structopt(name = "resync-n-workers", version = garage_version())]
ResyncNWorkers { n_workers: usize },
/// Set tranquility of block resync operations
#[structopt(name = "resync-tranquility", version = garage_version())]
ResyncTranquility { tranquility: u32 },
pub enum BlockOperation {
/// List all blocks that currently have a resync error
#[structopt(name = "list-errors", version = garage_version())]
ListErrors,
/// Get detailed information about a single block
#[structopt(name = "info", version = garage_version())]
Info {
/// Hash of the block for which to retrieve information
hash: String,
},
/// Retry now the resync of one or many blocks
#[structopt(name = "retry-now", version = garage_version())]
RetryNow {
/// Retry all blocks that have a resync error
#[structopt(long = "all")]
all: bool,
/// Hashes of the block to retry to resync now
blocks: Vec<String>,
},
/// Delete all objects referencing a missing block
#[structopt(name = "purge", version = garage_version())]
Purge {
/// Mandatory to confirm this operation
#[structopt(long = "yes")]
yes: bool,
/// Hashes of the block to purge
#[structopt(required = true)]
blocks: Vec<String>,
},
}

View file

@ -3,14 +3,17 @@ use std::time::Duration;
use garage_util::background::*;
use garage_util::crdt::*;
use garage_util::data::Uuid;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::formater::format_table;
use garage_util::time::*;
use garage_block::manager::BlockResyncErrorInfo;
use garage_model::bucket_table::*;
use garage_model::key_table::*;
use garage_model::s3::object_table::{BYTES, OBJECTS, UNFINISHED_UPLOADS};
use garage_model::s3::version_table::Version;
use crate::cli::structs::WorkerListOpt;
@ -241,7 +244,7 @@ pub fn find_matching_node(
}
}
pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) {
pub fn print_worker_list(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) {
let mut wi = wi.into_iter().collect::<Vec<_>>();
wi.sort_by_key(|(tid, info)| {
(
@ -254,7 +257,7 @@ pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) {
)
});
let mut table = vec![];
let mut table = vec!["TID\tState\tName\tTranq\tDone\tQueue\tErrors\tConsec\tLast".to_string()];
for (tid, info) in wi.iter() {
if wlo.busy && !matches!(info.state, WorkerState::Busy | WorkerState::Throttled(_)) {
continue;
@ -263,33 +266,155 @@ pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) {
continue;
}
table.push(format!("{}\t{}\t{}", tid, info.state, info.name));
if let Some(i) = &info.info {
table.push(format!("\t\t {}", i));
}
let tf = timeago::Formatter::new();
let (err_ago, err_msg) = info
let err_ago = info
.last_error
.as_ref()
.map(|(m, t)| {
(
tf.convert(Duration::from_millis(now_msec() - t)),
m.as_str(),
)
})
.unwrap_or(("(?) ago".into(), "(?)"));
if info.consecutive_errors > 0 {
.map(|(_, t)| tf.convert(Duration::from_millis(now_msec() - t)))
.unwrap_or_default();
let (total_err, consec_err) = if info.errors > 0 {
(info.errors.to_string(), info.consecutive_errors.to_string())
} else {
("-".into(), "-".into())
};
table.push(format!(
"{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}",
tid,
info.state,
info.name,
info.status
.tranquility
.as_ref()
.map(ToString::to_string)
.unwrap_or_else(|| "-".into()),
info.status.progress.as_deref().unwrap_or("-"),
info.status
.queue_length
.as_ref()
.map(ToString::to_string)
.unwrap_or_else(|| "-".into()),
total_err,
consec_err,
err_ago,
));
}
format_table(table);
}
pub fn print_worker_info(tid: usize, info: WorkerInfo) {
let mut table = vec![];
table.push(format!("Task id:\t{}", tid));
table.push(format!("Worker name:\t{}", info.name));
match info.state {
WorkerState::Throttled(t) => {
table.push(format!(
"\t\t {} consecutive errors ({} total), last {}",
info.consecutive_errors, info.errors, err_ago,
"Worker state:\tBusy (throttled, paused for {:.3}s)",
t
));
table.push(format!("\t\t {}", err_msg));
} else if info.errors > 0 {
table.push(format!("\t\t ({} errors, last {})", info.errors, err_ago,));
if wlo.errors {
table.push(format!("\t\t {}", err_msg));
}
s => {
table.push(format!("Worker state:\t{}", s));
}
};
if let Some(tql) = info.status.tranquility {
table.push(format!("Tranquility:\t{}", tql));
}
table.push("".into());
table.push(format!("Total errors:\t{}", info.errors));
table.push(format!("Consecutive errs:\t{}", info.consecutive_errors));
if let Some((s, t)) = info.last_error {
table.push(format!("Last error:\t{}", s));
let tf = timeago::Formatter::new();
table.push(format!(
"Last error time:\t{}",
tf.convert(Duration::from_millis(now_msec() - t))
));
}
table.push("".into());
if let Some(p) = info.status.progress {
table.push(format!("Progress:\t{}", p));
}
if let Some(ql) = info.status.queue_length {
table.push(format!("Queue length:\t{}", ql));
}
if let Some(pe) = info.status.persistent_errors {
table.push(format!("Persistent errors:\t{}", pe));
}
for (i, s) in info.status.freeform.iter().enumerate() {
if i == 0 {
if table.last() != Some(&"".into()) {
table.push("".into());
}
table.push(format!("Message:\t{}", s));
} else {
table.push(format!("\t{}", s));
}
}
format_table(table);
}
pub fn print_worker_vars(wv: Vec<(Uuid, String, String)>) {
let table = wv
.into_iter()
.map(|(n, k, v)| format!("{:?}\t{}\t{}", n, k, v))
.collect::<Vec<_>>();
format_table(table);
}
pub fn print_block_error_list(el: Vec<BlockResyncErrorInfo>) {
let now = now_msec();
let tf = timeago::Formatter::new();
let mut tf2 = timeago::Formatter::new();
tf2.ago("");
let mut table = vec!["Hash\tRC\tErrors\tLast error\tNext try".into()];
for e in el {
table.push(format!(
"{}\t{}\t{}\t{}\tin {}",
hex::encode(e.hash.as_slice()),
e.refcount,
e.error_count,
tf.convert(Duration::from_millis(now - e.last_try)),
tf2.convert(Duration::from_millis(e.next_try - now))
));
}
format_table(table);
}
pub fn print_block_info(hash: Hash, refcount: u64, versions: Vec<Result<Version, Uuid>>) {
println!("Block hash: {}", hex::encode(hash.as_slice()));
println!("Refcount: {}", refcount);
println!();
let mut table = vec!["Version\tBucket\tKey\tDeleted".into()];
let mut nondeleted_count = 0;
for v in versions.iter() {
match v {
Ok(ver) => {
table.push(format!(
"{:?}\t{:?}\t{}\t{:?}",
ver.uuid,
ver.bucket_id,
ver.key,
ver.deleted.get()
));
if !ver.deleted.get() {
nondeleted_count += 1;
}
}
Err(vh) => {
table.push(format!("{:?}\t\t\tyes", vh));
}
}
}
format_table(table);
if refcount != nondeleted_count {
println!();
println!("Warning: refcount does not match number of non-deleted versions");
}
}

View file

@ -127,9 +127,16 @@ async fn main() {
std::process::abort();
}));
// Parse arguments and dispatch command line
let opt = Opt::from_clap(&Opt::clap().version(version.as_str()).get_matches());
// Initialize logging as well as other libraries used in Garage
if std::env::var("RUST_LOG").is_err() {
std::env::set_var("RUST_LOG", "netapp=info,garage=info")
let default_log = match &opt.cmd {
Command::Server => "netapp=info,garage=info",
_ => "netapp=warn,garage=warn",
};
std::env::set_var("RUST_LOG", default_log)
}
tracing_subscriber::fmt()
.with_writer(std::io::stderr)
@ -137,9 +144,6 @@ async fn main() {
.init();
sodiumoxide::init().expect("Unable to init sodiumoxide");
// Parse arguments and dispatch command line
let opt = Opt::from_clap(&Opt::clap().version(version.as_str()).get_matches());
let res = match opt.cmd {
Command::Server => server::run_server(opt.config_file).await,
Command::OfflineRepair(repair_opt) => {
@ -169,7 +173,7 @@ async fn cli_command(opt: Opt) -> Result<(), Error> {
let net_key_hex_str = opt
.rpc_secret
.as_ref()
.or_else(|| config.as_ref().map(|c| &c.rpc_secret))
.or_else(|| config.as_ref().and_then(|c| c.rpc_secret.as_ref()))
.ok_or("No RPC secret provided")?;
let network_key = NetworkKey::from_slice(
&hex::decode(net_key_hex_str).err_context("Invalid RPC secret key (bad hex)")?[..],
@ -182,9 +186,9 @@ async fn cli_command(opt: Opt) -> Result<(), Error> {
let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, sk);
// Find and parse the address of the target host
let (id, addr) = if let Some(h) = opt.rpc_host {
let (id, addr, is_default_addr) = if let Some(h) = opt.rpc_host {
let (id, addrs) = parse_and_resolve_peer_addr(&h).ok_or_else(|| format!("Invalid RPC remote node identifier: {}. Expected format is <pubkey>@<IP or hostname>:<port>.", h))?;
(id, addrs[0])
(id, addrs[0], false)
} else {
let node_id = garage_rpc::system::read_node_id(&config.as_ref().unwrap().metadata_dir)
.err_context(READ_KEY_ERROR)?;
@ -195,24 +199,26 @@ async fn cli_command(opt: Opt) -> Result<(), Error> {
.ok_or_message("unable to resolve rpc_public_addr specified in config file")?
.next()
.ok_or_message("unable to resolve rpc_public_addr specified in config file")?;
(node_id, a)
(node_id, a, false)
} else {
let default_addr = SocketAddr::new(
"127.0.0.1".parse().unwrap(),
config.as_ref().unwrap().rpc_bind_addr.port(),
);
warn!(
"Trying to contact Garage node at default address {}",
default_addr
);
warn!("If this doesn't work, consider adding rpc_public_addr in your config file or specifying the -h command line parameter.");
(node_id, default_addr)
(node_id, default_addr, true)
}
};
// Connect to target host
netapp.clone().try_connect(addr, id).await
.err_context("Unable to connect to destination RPC host. Check that you are using the same value of rpc_secret as them, and that you have their correct public key.")?;
if let Err(e) = netapp.clone().try_connect(addr, id).await {
if is_default_addr {
warn!(
"Tried to contact Garage node at default address {}, which didn't work. If that address is wrong, consider setting rpc_public_addr in your config file.",
addr
);
}
Err(e).err_context("Unable to connect to destination RPC host. Check that you are using the same value of rpc_secret as them, and that you have their correct public key.")?;
}
let system_rpc_endpoint = netapp.endpoint::<SystemRpc, ()>(SYSTEM_RPC_PATH.into());
let admin_rpc_endpoint = netapp.endpoint::<AdminRpc, ()>(ADMIN_RPC_PATH.into());

View file

@ -1,8 +1,5 @@
use std::path::PathBuf;
use tokio::sync::watch;
use garage_util::background::*;
use garage_util::config::*;
use garage_util::error::*;
@ -20,12 +17,8 @@ pub async fn offline_repair(config_file: PathBuf, opt: OfflineRepairOpt) -> Resu
info!("Loading configuration...");
let config = read_config(config_file)?;
info!("Initializing background runner...");
let (done_tx, done_rx) = watch::channel(false);
let (background, await_background_done) = BackgroundRunner::new(16, done_rx);
info!("Initializing Garage main data store...");
let garage = Garage::new(config.clone(), background)?;
let garage = Garage::new(config)?;
info!("Launching repair operation...");
match opt.what {
@ -43,13 +36,7 @@ pub async fn offline_repair(config_file: PathBuf, opt: OfflineRepairOpt) -> Resu
}
}
info!("Repair operation finished, shutting down Garage internals...");
done_tx.send(true).unwrap();
drop(garage);
await_background_done.await?;
info!("Cleaning up...");
info!("Repair operation finished, shutting down...");
Ok(())
}

View file

@ -12,38 +12,37 @@ use garage_model::s3::version_table::*;
use garage_table::*;
use garage_util::background::*;
use garage_util::error::Error;
use garage_util::migrate::Migrate;
use crate::*;
pub async fn launch_online_repair(garage: Arc<Garage>, opt: RepairOpt) {
pub async fn launch_online_repair(
garage: &Arc<Garage>,
bg: &BackgroundRunner,
opt: RepairOpt,
) -> Result<(), Error> {
match opt.what {
RepairWhat::Tables => {
info!("Launching a full sync of tables");
garage.bucket_table.syncer.add_full_sync();
garage.object_table.syncer.add_full_sync();
garage.version_table.syncer.add_full_sync();
garage.block_ref_table.syncer.add_full_sync();
garage.key_table.syncer.add_full_sync();
garage.bucket_table.syncer.add_full_sync()?;
garage.object_table.syncer.add_full_sync()?;
garage.version_table.syncer.add_full_sync()?;
garage.block_ref_table.syncer.add_full_sync()?;
garage.key_table.syncer.add_full_sync()?;
}
RepairWhat::Versions => {
info!("Repairing the versions table");
garage
.background
.spawn_worker(RepairVersionsWorker::new(garage.clone()));
bg.spawn_worker(RepairVersionsWorker::new(garage.clone()));
}
RepairWhat::BlockRefs => {
info!("Repairing the block refs table");
garage
.background
.spawn_worker(RepairBlockrefsWorker::new(garage.clone()));
bg.spawn_worker(RepairBlockrefsWorker::new(garage.clone()));
}
RepairWhat::Blocks => {
info!("Repairing the stored blocks");
garage
.background
.spawn_worker(garage_block::repair::RepairWorker::new(
garage.block_manager.clone(),
));
bg.spawn_worker(garage_block::repair::RepairWorker::new(
garage.block_manager.clone(),
));
}
RepairWhat::Scrub { cmd } => {
let cmd = match cmd {
@ -52,13 +51,18 @@ pub async fn launch_online_repair(garage: Arc<Garage>, opt: RepairOpt) {
ScrubCmd::Resume => ScrubWorkerCommand::Resume,
ScrubCmd::Cancel => ScrubWorkerCommand::Cancel,
ScrubCmd::SetTranquility { tranquility } => {
ScrubWorkerCommand::SetTranquility(tranquility)
garage
.block_manager
.scrub_persister
.set_with(|x| x.tranquility = tranquility)?;
return Ok(());
}
};
info!("Sending command to scrub worker: {:?}", cmd);
garage.block_manager.send_scrub_command(cmd).await;
garage.block_manager.send_scrub_command(cmd).await?;
}
}
Ok(())
}
// ----
@ -85,25 +89,23 @@ impl Worker for RepairVersionsWorker {
"Version repair worker".into()
}
fn info(&self) -> Option<String> {
Some(format!("{} items done", self.counter))
fn status(&self) -> WorkerStatus {
WorkerStatus {
progress: Some(self.counter.to_string()),
..Default::default()
}
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
let item_bytes = match self.garage.version_table.data.store.get_gt(&self.pos)? {
Some((k, v)) => {
self.pos = k;
v
}
let (item_bytes, next_pos) = match self.garage.version_table.data.store.get_gt(&self.pos)? {
Some((k, v)) => (v, k),
None => {
info!("repair_versions: finished, done {}", self.counter);
return Ok(WorkerState::Done);
}
};
self.counter += 1;
let version = rmp_serde::decode::from_read_ref::<_, Version>(&item_bytes)?;
let version = Version::decode(&item_bytes).ok_or_message("Cannot decode Version")?;
if !version.deleted.get() {
let object = self
.garage
@ -131,10 +133,13 @@ impl Worker for RepairVersionsWorker {
}
}
self.counter += 1;
self.pos = next_pos;
Ok(WorkerState::Busy)
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
async fn wait_for_work(&mut self) -> WorkerState {
unreachable!()
}
}
@ -163,25 +168,24 @@ impl Worker for RepairBlockrefsWorker {
"Block refs repair worker".into()
}
fn info(&self) -> Option<String> {
Some(format!("{} items done", self.counter))
fn status(&self) -> WorkerStatus {
WorkerStatus {
progress: Some(self.counter.to_string()),
..Default::default()
}
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
let item_bytes = match self.garage.block_ref_table.data.store.get_gt(&self.pos)? {
Some((k, v)) => {
self.pos = k;
v
}
None => {
info!("repair_block_ref: finished, done {}", self.counter);
return Ok(WorkerState::Done);
}
};
let (item_bytes, next_pos) =
match self.garage.block_ref_table.data.store.get_gt(&self.pos)? {
Some((k, v)) => (v, k),
None => {
info!("repair_block_ref: finished, done {}", self.counter);
return Ok(WorkerState::Done);
}
};
self.counter += 1;
let block_ref = rmp_serde::decode::from_read_ref::<_, BlockRef>(&item_bytes)?;
let block_ref = BlockRef::decode(&item_bytes).ok_or_message("Cannot decode BlockRef")?;
if !block_ref.deleted.get() {
let version = self
.garage
@ -206,10 +210,13 @@ impl Worker for RepairBlockrefsWorker {
}
}
self.counter += 1;
self.pos = next_pos;
Ok(WorkerState::Busy)
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
async fn wait_for_work(&mut self) -> WorkerState {
unreachable!()
}
}

View file

@ -35,12 +35,15 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> {
#[cfg(feature = "metrics")]
let metrics_exporter = opentelemetry_prometheus::exporter().init();
info!("Initializing Garage main data store...");
let garage = Garage::new(config.clone())?;
info!("Initializing background runner...");
let watch_cancel = watch_shutdown_signal();
let (background, await_background_done) = BackgroundRunner::new(16, watch_cancel.clone());
let (background, await_background_done) = BackgroundRunner::new(watch_cancel.clone());
info!("Initializing Garage main data store...");
let garage = Garage::new(config.clone(), background)?;
info!("Spawning Garage workers...");
garage.spawn_workers(&background);
if config.admin.trace_sink.is_some() {
info!("Initialize tracing...");
@ -63,7 +66,7 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> {
let run_system = tokio::spawn(garage.system.clone().run(watch_cancel.clone()));
info!("Create admin RPC handler...");
AdminRpcHandler::new(garage.clone());
AdminRpcHandler::new(garage.clone(), background.clone());
// ---- Launch public-facing API servers ----

View file

@ -22,7 +22,7 @@ tokio = "1.17.0"
# cli deps
clap = { version = "3.1.18", optional = true, features = ["derive", "env"] }
garage_util = { version = "0.8.0", path = "../util", optional = true }
garage_util = { version = "0.8.1", path = "../util", optional = true }
[features]

View file

@ -1,6 +1,6 @@
[package]
name = "garage_model"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,11 +14,11 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_db = { version = "0.8.0", default-features = false, path = "../db" }
garage_rpc = { version = "0.8.0", path = "../rpc" }
garage_table = { version = "0.8.0", path = "../table" }
garage_block = { version = "0.8.0", path = "../block" }
garage_util = { version = "0.8.0", path = "../util" }
garage_db = { version = "0.8.1", default-features = false, path = "../db" }
garage_rpc = { version = "0.8.1", path = "../rpc" }
garage_table = { version = "0.8.1", path = "../table" }
garage_block = { version = "0.8.1", path = "../block" }
garage_util = { version = "0.8.1", path = "../util" }
async-trait = "0.1.7"
arc-swap = "1.0"
@ -30,7 +30,6 @@ tracing = "0.1.30"
rand = "0.8"
zstd = { version = "0.9", default-features = false }
rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"

View file

@ -1,18 +1,26 @@
use serde::{Deserialize, Serialize};
use garage_util::data::*;
use garage_table::crdt::*;
use garage_table::*;
/// The bucket alias table holds the names given to buckets
/// in the global namespace.
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BucketAlias {
name: String,
pub state: crdt::Lww<Option<Uuid>>,
mod v08 {
use garage_util::crdt;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
/// The bucket alias table holds the names given to buckets
/// in the global namespace.
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BucketAlias {
pub(super) name: String,
pub state: crdt::Lww<Option<Uuid>>,
}
impl garage_util::migrate::InitialFormat for BucketAlias {}
}
pub use v08::*;
impl BucketAlias {
pub fn new(name: String, ts: u64, bucket_id: Option<Uuid>) -> Option<Self> {
if !is_valid_bucket_name(&name) {

View file

@ -1,5 +1,3 @@
use serde::{Deserialize, Serialize};
use garage_table::crdt::*;
use garage_table::*;
use garage_util::data::*;
@ -7,71 +5,82 @@ use garage_util::time::*;
use crate::permission::BucketKeyPerm;
/// A bucket is a collection of objects
///
/// Its parameters are not directly accessible as:
/// - It must be possible to merge paramaters, hence the use of a LWW CRDT.
/// - A bucket has 2 states, Present or Deleted and parameters make sense only if present.
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Bucket {
/// ID of the bucket
pub id: Uuid,
/// State, and configuration if not deleted, of the bucket
pub state: crdt::Deletable<BucketParams>,
mod v08 {
use crate::permission::BucketKeyPerm;
use garage_util::crdt;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
/// A bucket is a collection of objects
///
/// Its parameters are not directly accessible as:
/// - It must be possible to merge paramaters, hence the use of a LWW CRDT.
/// - A bucket has 2 states, Present or Deleted and parameters make sense only if present.
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Bucket {
/// ID of the bucket
pub id: Uuid,
/// State, and configuration if not deleted, of the bucket
pub state: crdt::Deletable<BucketParams>,
}
/// Configuration for a bucket
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BucketParams {
/// Bucket's creation date
pub creation_date: u64,
/// Map of key with access to the bucket, and what kind of access they give
pub authorized_keys: crdt::Map<String, BucketKeyPerm>,
/// Map of aliases that are or have been given to this bucket
/// in the global namespace
/// (not authoritative: this is just used as an indication to
/// map back to aliases when doing ListBuckets)
pub aliases: crdt::LwwMap<String, bool>,
/// Map of aliases that are or have been given to this bucket
/// in namespaces local to keys
/// key = (access key id, alias name)
pub local_aliases: crdt::LwwMap<(String, String), bool>,
/// Whether this bucket is allowed for website access
/// (under all of its global alias names),
/// and if so, the website configuration XML document
pub website_config: crdt::Lww<Option<WebsiteConfig>>,
/// CORS rules
pub cors_config: crdt::Lww<Option<Vec<CorsRule>>>,
/// Bucket quotas
#[serde(default)]
pub quotas: crdt::Lww<BucketQuotas>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct WebsiteConfig {
pub index_document: String,
pub error_document: Option<String>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct CorsRule {
pub id: Option<String>,
pub max_age_seconds: Option<u64>,
pub allow_origins: Vec<String>,
pub allow_methods: Vec<String>,
pub allow_headers: Vec<String>,
pub expose_headers: Vec<String>,
}
#[derive(Default, PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct BucketQuotas {
/// Maximum size in bytes (bucket size = sum of sizes of objects in the bucket)
pub max_size: Option<u64>,
/// Maximum number of non-deleted objects in the bucket
pub max_objects: Option<u64>,
}
impl garage_util::migrate::InitialFormat for Bucket {}
}
/// Configuration for a bucket
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BucketParams {
/// Bucket's creation date
pub creation_date: u64,
/// Map of key with access to the bucket, and what kind of access they give
pub authorized_keys: crdt::Map<String, BucketKeyPerm>,
/// Map of aliases that are or have been given to this bucket
/// in the global namespace
/// (not authoritative: this is just used as an indication to
/// map back to aliases when doing ListBuckets)
pub aliases: crdt::LwwMap<String, bool>,
/// Map of aliases that are or have been given to this bucket
/// in namespaces local to keys
/// key = (access key id, alias name)
pub local_aliases: crdt::LwwMap<(String, String), bool>,
/// Whether this bucket is allowed for website access
/// (under all of its global alias names),
/// and if so, the website configuration XML document
pub website_config: crdt::Lww<Option<WebsiteConfig>>,
/// CORS rules
pub cors_config: crdt::Lww<Option<Vec<CorsRule>>>,
/// Bucket quotas
#[serde(default)]
pub quotas: crdt::Lww<BucketQuotas>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct WebsiteConfig {
pub index_document: String,
pub error_document: Option<String>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct CorsRule {
pub id: Option<String>,
pub max_age_seconds: Option<u64>,
pub allow_origins: Vec<String>,
pub allow_methods: Vec<String>,
pub allow_headers: Vec<String>,
pub expose_headers: Vec<String>,
}
#[derive(Default, PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct BucketQuotas {
/// Maximum size in bytes (bucket size = sum of sizes of objects in the bucket)
pub max_size: Option<u64>,
/// Maximum number of non-deleted objects in the bucket
pub max_objects: Option<u64>,
}
pub use v08::*;
impl AutoCrdt for BucketQuotas {
const WARN_IF_DIFFERENT: bool = true;

View file

@ -33,14 +33,14 @@ use crate::k2v::{item_table::*, poll::*, rpc::*};
pub struct Garage {
/// The parsed configuration Garage is running
pub config: Config,
/// The set of background variables that can be viewed/modified at runtime
pub bg_vars: vars::BgVars,
/// The replication mode of this cluster
pub replication_mode: ReplicationMode,
/// The local database
pub db: db::Db,
/// A background job runner
pub background: Arc<BackgroundRunner>,
/// The membership manager
pub system: Arc<System>,
/// The block manager
@ -78,7 +78,7 @@ pub struct GarageK2V {
impl Garage {
/// Create and run garage
pub fn new(config: Config, background: Arc<BackgroundRunner>) -> Result<Arc<Self>, Error> {
pub fn new(config: Config) -> Result<Arc<Self>, Error> {
// Create meta dir and data dir if they don't exist already
std::fs::create_dir_all(&config.metadata_dir)
.ok_or_message("Unable to create Garage metadata directory")?;
@ -159,7 +159,7 @@ impl Garage {
};
let network_key = NetworkKey::from_slice(
&hex::decode(&config.rpc_secret).expect("Invalid RPC secret key")[..],
&hex::decode(&config.rpc_secret.as_ref().unwrap()).expect("Invalid RPC secret key")[..],
)
.expect("Invalid RPC secret key");
@ -167,7 +167,7 @@ impl Garage {
.expect("Invalid replication_mode in config file.");
info!("Initialize membership management system...");
let system = System::new(network_key, background.clone(), replication_mode, &config)?;
let system = System::new(network_key, replication_mode, &config)?;
let data_rep_param = TableShardedReplication {
system: system.clone(),
@ -225,7 +225,6 @@ impl Garage {
info!("Initialize version_table...");
let version_table = Table::new(
VersionTable {
background: background.clone(),
block_ref_table: block_ref_table.clone(),
},
meta_rep_param.clone(),
@ -240,7 +239,6 @@ impl Garage {
#[allow(clippy::redundant_clone)]
let object_table = Table::new(
ObjectTable {
background: background.clone(),
version_table: version_table.clone(),
object_counter_table: object_counter_table.clone(),
},
@ -253,12 +251,16 @@ impl Garage {
#[cfg(feature = "k2v")]
let k2v = GarageK2V::new(system.clone(), &db, meta_rep_param);
// Initialize bg vars
let mut bg_vars = vars::BgVars::new();
block_manager.register_bg_vars(&mut bg_vars);
// -- done --
Ok(Arc::new(Self {
config,
bg_vars,
replication_mode,
db,
background,
system,
block_manager,
bucket_table,
@ -273,6 +275,22 @@ impl Garage {
}))
}
pub fn spawn_workers(&self, bg: &BackgroundRunner) {
self.block_manager.spawn_workers(bg);
self.bucket_table.spawn_workers(bg);
self.bucket_alias_table.spawn_workers(bg);
self.key_table.spawn_workers(bg);
self.object_table.spawn_workers(bg);
self.object_counter_table.spawn_workers(bg);
self.version_table.spawn_workers(bg);
self.block_ref_table.spawn_workers(bg);
#[cfg(feature = "k2v")]
self.k2v.spawn_workers(bg);
}
pub fn bucket_helper(&self) -> helper::bucket::BucketHelper {
helper::bucket::BucketHelper(self)
}
@ -307,4 +325,9 @@ impl GarageK2V {
rpc,
}
}
pub fn spawn_workers(&self, bg: &BackgroundRunner) {
self.item_table.spawn_workers(bg);
self.counter_table.spawn_workers(bg);
}
}

View file

@ -1,19 +1,18 @@
use core::ops::Bound;
use std::collections::{hash_map, BTreeMap, HashMap};
use std::collections::{BTreeMap, HashMap};
use std::marker::PhantomData;
use std::sync::Arc;
use async_trait::async_trait;
use serde::{Deserialize, Serialize};
use tokio::sync::{mpsc, watch};
use garage_db as db;
use garage_rpc::ring::Ring;
use garage_rpc::system::System;
use garage_util::background::*;
use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::migrate::Migrate;
use garage_util::time::*;
use garage_table::crdt::*;
@ -31,14 +30,44 @@ pub trait CountedItem: Clone + PartialEq + Send + Sync + 'static {
fn counts(&self) -> Vec<(&'static str, i64)>;
}
/// A counter entry in the global table
#[derive(Clone, PartialEq, Debug, Serialize, Deserialize)]
pub struct CounterEntry<T: CountedItem> {
pub pk: T::CP,
pub sk: T::CS,
pub values: BTreeMap<String, CounterValue>,
mod v08 {
use super::CountedItem;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
// ---- Global part (the table everyone queries) ----
/// A counter entry in the global table
#[derive(Clone, PartialEq, Debug, Serialize, Deserialize)]
pub struct CounterEntry<T: CountedItem> {
pub pk: T::CP,
pub sk: T::CS,
pub values: BTreeMap<String, CounterValue>,
}
/// A counter entry in the global table
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct CounterValue {
pub node_values: BTreeMap<Uuid, (u64, i64)>,
}
impl<T: CountedItem> garage_util::migrate::InitialFormat for CounterEntry<T> {}
// ---- Local part (the counter we maintain transactionnaly on each node) ----
#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
pub(super) struct LocalCounterEntry<T: CountedItem> {
pub(super) pk: T::CP,
pub(super) sk: T::CS,
pub(super) values: BTreeMap<String, (u64, i64)>,
}
impl<T: CountedItem> garage_util::migrate::InitialFormat for LocalCounterEntry<T> {}
}
pub use v08::*;
impl<T: CountedItem> Entry<T::CP, T::CS> for CounterEntry<T> {
fn partition_key(&self) -> &T::CP {
&self.pk
@ -80,12 +109,6 @@ impl<T: CountedItem> CounterEntry<T> {
}
}
/// A counter entry in the global table
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct CounterValue {
pub node_values: BTreeMap<Uuid, (u64, i64)>,
}
impl<T: CountedItem> Crdt for CounterEntry<T> {
fn merge(&mut self, other: &Self) {
for (name, e2) in other.values.iter() {
@ -142,7 +165,6 @@ impl<T: CountedItem> TableSchema for CounterTable<T> {
pub struct IndexCounter<T: CountedItem> {
this_node: Uuid,
local_counter: db::Tree,
propagate_tx: mpsc::UnboundedSender<(T::CP, T::CS, LocalCounterEntry<T>)>,
pub table: Arc<Table<CounterTable<T>, TableShardedReplication>>,
}
@ -152,16 +174,11 @@ impl<T: CountedItem> IndexCounter<T> {
replication: TableShardedReplication,
db: &db::Db,
) -> Arc<Self> {
let background = system.background.clone();
let (propagate_tx, propagate_rx) = mpsc::unbounded_channel();
let this = Arc::new(Self {
Arc::new(Self {
this_node: system.id,
local_counter: db
.open_tree(format!("local_counter_v2:{}", T::COUNTER_TABLE_NAME))
.expect("Unable to open local counter tree"),
propagate_tx,
table: Table::new(
CounterTable {
_phantom_t: Default::default(),
@ -170,16 +187,11 @@ impl<T: CountedItem> IndexCounter<T> {
system,
db,
),
});
})
}
background.spawn_worker(IndexPropagatorWorker {
index_counter: this.clone(),
propagate_rx,
buf: HashMap::new(),
errors: 0,
});
this
pub fn spawn_workers(&self, bg: &BackgroundRunner) {
self.table.spawn_workers(bg);
}
pub fn count(
@ -208,11 +220,9 @@ impl<T: CountedItem> IndexCounter<T> {
let tree_key = self.table.data.tree_key(pk, sk);
let mut entry = match tx.get(&self.local_counter, &tree_key[..])? {
Some(old_bytes) => {
rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>(&old_bytes)
.map_err(Error::RmpDecode)
.map_err(db::TxError::Abort)?
}
Some(old_bytes) => LocalCounterEntry::<T>::decode(&old_bytes)
.ok_or_message("Cannot decode local counter entry")
.map_err(db::TxError::Abort)?,
None => LocalCounterEntry {
pk: pk.clone(),
sk: sk.clone(),
@ -227,17 +237,14 @@ impl<T: CountedItem> IndexCounter<T> {
ent.1 += *inc;
}
let new_entry_bytes = rmp_to_vec_all_named(&entry)
let new_entry_bytes = entry
.encode()
.map_err(Error::RmpEncode)
.map_err(db::TxError::Abort)?;
tx.insert(&self.local_counter, &tree_key[..], new_entry_bytes)?;
if let Err(e) = self.propagate_tx.send((pk.clone(), sk.clone(), entry)) {
error!(
"Could not propagate updated counter values, failed to send to channel: {}",
e
);
}
let dist_entry = entry.into_counter_entry(self.this_node);
self.table.queue_insert(tx, &dist_entry)?;
Ok(())
}
@ -250,23 +257,6 @@ impl<T: CountedItem> IndexCounter<T> {
TS: TableSchema<E = T>,
TR: TableReplication,
{
let save_counter_entry = |entry: CounterEntry<T>| -> Result<(), Error> {
let entry_k = self
.table
.data
.tree_key(entry.partition_key(), entry.sort_key());
self.table
.data
.update_entry_with(&entry_k, |ent| match ent {
Some(mut ent) => {
ent.merge(&entry);
ent
}
None => entry.clone(),
})?;
Ok(())
};
// 1. Set all old local counters to zero
let now = now_msec();
let mut next_start: Option<Vec<u8>> = None;
@ -289,20 +279,22 @@ impl<T: CountedItem> IndexCounter<T> {
info!("zeroing old counters... ({})", hex::encode(&batch[0].0));
for (local_counter_k, local_counter) in batch {
let mut local_counter =
rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>(&local_counter)?;
let mut local_counter = LocalCounterEntry::<T>::decode(&local_counter)
.ok_or_message("Cannot decode local counter entry")?;
for (_, tv) in local_counter.values.iter_mut() {
tv.0 = std::cmp::max(tv.0 + 1, now);
tv.1 = 0;
}
let local_counter_bytes = rmp_to_vec_all_named(&local_counter)?;
let local_counter_bytes = local_counter.encode()?;
self.local_counter
.insert(&local_counter_k, &local_counter_bytes)?;
let counter_entry = local_counter.into_counter_entry(self.this_node);
save_counter_entry(counter_entry)?;
self.local_counter
.db()
.transaction(|mut tx| self.table.queue_insert(&mut tx, &counter_entry))?;
next_start = Some(local_counter_k);
}
@ -343,9 +335,8 @@ impl<T: CountedItem> IndexCounter<T> {
let local_counter_key = self.table.data.tree_key(pk, sk);
let mut local_counter = match self.local_counter.get(&local_counter_key)? {
Some(old_bytes) => {
let ent = rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>(
&old_bytes,
)?;
let ent = LocalCounterEntry::<T>::decode(&old_bytes)
.ok_or_message("Cannot decode local counter entry")?;
assert!(ent.pk == *pk);
assert!(ent.sk == *sk);
ent
@ -362,12 +353,14 @@ impl<T: CountedItem> IndexCounter<T> {
tv.1 += v;
}
let local_counter_bytes = rmp_to_vec_all_named(&local_counter)?;
let local_counter_bytes = local_counter.encode()?;
self.local_counter
.insert(&local_counter_key, local_counter_bytes)?;
let counter_entry = local_counter.into_counter_entry(self.this_node);
save_counter_entry(counter_entry)?;
self.local_counter
.db()
.transaction(|mut tx| self.table.queue_insert(&mut tx, &counter_entry))?;
next_start = Some(counted_entry_k);
}
@ -378,104 +371,7 @@ impl<T: CountedItem> IndexCounter<T> {
}
}
struct IndexPropagatorWorker<T: CountedItem> {
index_counter: Arc<IndexCounter<T>>,
propagate_rx: mpsc::UnboundedReceiver<(T::CP, T::CS, LocalCounterEntry<T>)>,
buf: HashMap<Vec<u8>, CounterEntry<T>>,
errors: usize,
}
impl<T: CountedItem> IndexPropagatorWorker<T> {
fn add_ent(&mut self, pk: T::CP, sk: T::CS, counters: LocalCounterEntry<T>) {
let tree_key = self.index_counter.table.data.tree_key(&pk, &sk);
let dist_entry = counters.into_counter_entry(self.index_counter.this_node);
match self.buf.entry(tree_key) {
hash_map::Entry::Vacant(e) => {
e.insert(dist_entry);
}
hash_map::Entry::Occupied(mut e) => {
e.get_mut().merge(&dist_entry);
}
}
}
}
#[async_trait]
impl<T: CountedItem> Worker for IndexPropagatorWorker<T> {
fn name(&self) -> String {
format!("{} index counter propagator", T::COUNTER_TABLE_NAME)
}
fn info(&self) -> Option<String> {
if !self.buf.is_empty() {
Some(format!("{} items in queue", self.buf.len()))
} else {
None
}
}
async fn work(&mut self, must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
// This loop batches updates to counters to be sent all at once.
// They are sent once the propagate_rx channel has been emptied (or is closed).
let closed = loop {
match self.propagate_rx.try_recv() {
Ok((pk, sk, counters)) => {
self.add_ent(pk, sk, counters);
}
Err(mpsc::error::TryRecvError::Empty) => break false,
Err(mpsc::error::TryRecvError::Disconnected) => break true,
}
};
if !self.buf.is_empty() {
let entries_k = self.buf.keys().take(100).cloned().collect::<Vec<_>>();
let entries = entries_k.iter().map(|k| self.buf.get(k).unwrap());
if let Err(e) = self.index_counter.table.insert_many(entries).await {
self.errors += 1;
if self.errors >= 2 && *must_exit.borrow() {
error!("({}) Could not propagate {} counter values: {}, these counters will not be updated correctly.", T::COUNTER_TABLE_NAME, self.buf.len(), e);
return Ok(WorkerState::Done);
}
// Propagate error up to worker manager, it will log it, increment a counter,
// and sleep for a certain delay (with exponential backoff), waiting for
// things to go back to normal
return Err(e);
} else {
for k in entries_k {
self.buf.remove(&k);
}
self.errors = 0;
}
return Ok(WorkerState::Busy);
} else if closed {
return Ok(WorkerState::Done);
} else {
return Ok(WorkerState::Idle);
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
match self.propagate_rx.recv().await {
Some((pk, sk, counters)) => {
self.add_ent(pk, sk, counters);
WorkerState::Busy
}
None => match self.buf.is_empty() {
false => WorkerState::Busy,
true => WorkerState::Done,
},
}
}
}
#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
struct LocalCounterEntry<T: CountedItem> {
pk: T::CP,
sk: T::CS,
values: BTreeMap<String, (u64, i64)>,
}
// ----
impl<T: CountedItem> LocalCounterEntry<T> {
fn into_counter_entry(self, this_node: Uuid) -> CounterEntry<T> {

View file

@ -1,7 +1,8 @@
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
use std::sync::Arc;
use serde::{Deserialize, Serialize};
use garage_db as db;
use garage_util::data::*;
@ -17,31 +18,42 @@ pub const CONFLICTS: &str = "conflicts";
pub const VALUES: &str = "values";
pub const BYTES: &str = "bytes";
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct K2VItem {
pub partition: K2VItemPartition,
pub sort_key: String,
mod v08 {
use crate::k2v::causality::K2VNodeId;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
items: BTreeMap<K2VNodeId, DvvsEntry>,
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct K2VItem {
pub partition: K2VItemPartition,
pub sort_key: String,
pub(super) items: BTreeMap<K2VNodeId, DvvsEntry>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize, Hash)]
pub struct K2VItemPartition {
pub bucket_id: Uuid,
pub partition_key: String,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct DvvsEntry {
pub(super) t_discard: u64,
pub(super) values: Vec<(u64, DvvsValue)>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub enum DvvsValue {
Value(#[serde(with = "serde_bytes")] Vec<u8>),
Deleted,
}
impl garage_util::migrate::InitialFormat for K2VItem {}
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize, Hash)]
pub struct K2VItemPartition {
pub bucket_id: Uuid,
pub partition_key: String,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
struct DvvsEntry {
t_discard: u64,
values: Vec<(u64, DvvsValue)>,
}
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub enum DvvsValue {
Value(#[serde(with = "serde_bytes")] Vec<u8>),
Deleted,
}
pub use v08::*;
impl K2VItem {
/// Creates a new K2VItem when no previous entry existed in the db

View file

@ -273,14 +273,9 @@ impl K2VRpcHandler {
}
fn local_insert(&self, item: &InsertedItem) -> Result<Option<K2VItem>, Error> {
let tree_key = self
.item_table
.data
.tree_key(&item.partition, &item.sort_key);
self.item_table
.data
.update_entry_with(&tree_key[..], |ent| {
.update_entry_with(&item.partition, &item.sort_key, |ent| {
let mut ent = ent.unwrap_or_else(|| {
K2VItem::new(
item.partition.bucket_id,

View file

@ -1,45 +1,121 @@
use serde::{Deserialize, Serialize};
use garage_table::crdt::*;
use garage_table::*;
use garage_util::crdt::{self, Crdt};
use garage_util::data::*;
use garage_table::{DeletedFilter, EmptyKey, Entry, TableSchema};
use crate::permission::BucketKeyPerm;
use crate::prev::v051::key_table as old;
pub(crate) mod v05 {
use garage_util::crdt;
use serde::{Deserialize, Serialize};
/// An api key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Key {
/// The id of the key (immutable), used as partition key
pub key_id: String,
/// An api key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Key {
/// The id of the key (immutable), used as partition key
pub key_id: String,
/// Internal state of the key
pub state: crdt::Deletable<KeyParams>,
/// The secret_key associated
pub secret_key: String,
/// Name for the key
pub name: crdt::Lww<String>,
/// Is the key deleted
pub deleted: crdt::Bool,
/// Buckets in which the key is authorized. Empty if `Key` is deleted
// CRDT interaction: deleted implies authorized_buckets is empty
pub authorized_buckets: crdt::LwwMap<String, PermissionSet>,
}
/// Permission given to a key in a bucket
#[derive(PartialOrd, Ord, PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct PermissionSet {
/// The key can be used to read the bucket
pub allow_read: bool,
/// The key can be used to write in the bucket
pub allow_write: bool,
}
impl crdt::AutoCrdt for PermissionSet {
const WARN_IF_DIFFERENT: bool = true;
}
impl garage_util::migrate::InitialFormat for Key {}
}
/// Configuration for a key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct KeyParams {
/// The secret_key associated (immutable)
pub secret_key: String,
mod v08 {
use super::v05;
use crate::permission::BucketKeyPerm;
use garage_util::crdt;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
/// Name for the key
pub name: crdt::Lww<String>,
/// An api key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Key {
/// The id of the key (immutable), used as partition key
pub key_id: String,
/// Flag to allow users having this key to create buckets
pub allow_create_bucket: crdt::Lww<bool>,
/// Internal state of the key
pub state: crdt::Deletable<KeyParams>,
}
/// If the key is present: it gives some permissions,
/// a map of bucket IDs (uuids) to permissions.
/// Otherwise no permissions are granted to key
pub authorized_buckets: crdt::Map<Uuid, BucketKeyPerm>,
/// Configuration for a key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct KeyParams {
/// The secret_key associated (immutable)
pub secret_key: String,
/// A key can have a local view of buckets names it is
/// the only one to see, this is the namespace for these aliases
pub local_aliases: crdt::LwwMap<String, Option<Uuid>>,
/// Name for the key
pub name: crdt::Lww<String>,
/// Flag to allow users having this key to create buckets
pub allow_create_bucket: crdt::Lww<bool>,
/// If the key is present: it gives some permissions,
/// a map of bucket IDs (uuids) to permissions.
/// Otherwise no permissions are granted to key
pub authorized_buckets: crdt::Map<Uuid, BucketKeyPerm>,
/// A key can have a local view of buckets names it is
/// the only one to see, this is the namespace for these aliases
pub local_aliases: crdt::LwwMap<String, Option<Uuid>>,
}
impl garage_util::migrate::Migrate for Key {
type Previous = v05::Key;
fn migrate(old_k: v05::Key) -> Key {
let name = crdt::Lww::raw(old_k.name.timestamp(), old_k.name.get().clone());
let state = if old_k.deleted.get() {
crdt::Deletable::Deleted
} else {
// Authorized buckets is ignored here,
// migration is performed in specific migration code in
// garage/migrate.rs
crdt::Deletable::Present(KeyParams {
secret_key: old_k.secret_key,
name,
allow_create_bucket: crdt::Lww::new(false),
authorized_buckets: crdt::Map::new(),
local_aliases: crdt::LwwMap::new(),
})
};
Key {
key_id: old_k.key_id,
state,
}
}
}
}
pub use v08::*;
impl KeyParams {
fn new(secret_key: &str, name: &str) -> Self {
KeyParams {
@ -173,28 +249,4 @@ impl TableSchema for KeyTable {
}
}
}
fn try_migrate(bytes: &[u8]) -> Option<Self::E> {
let old_k = rmp_serde::decode::from_read_ref::<_, old::Key>(bytes).ok()?;
let name = crdt::Lww::raw(old_k.name.timestamp(), old_k.name.get().clone());
let state = if old_k.deleted.get() {
crdt::Deletable::Deleted
} else {
// Authorized buckets is ignored here,
// migration is performed in specific migration code in
// garage/migrate.rs
crdt::Deletable::Present(KeyParams {
secret_key: old_k.secret_key,
name,
allow_create_bucket: crdt::Lww::new(false),
authorized_buckets: crdt::Map::new(),
local_aliases: crdt::LwwMap::new(),
})
};
Some(Key {
key_id: old_k.key_id,
state,
})
}
}

View file

@ -2,6 +2,7 @@ use std::sync::Arc;
use garage_util::crdt::*;
use garage_util::data::*;
use garage_util::encode::nonversioned_decode;
use garage_util::error::Error as GarageError;
use garage_util::time::*;
@ -28,8 +29,8 @@ impl Migrate {
let mut old_buckets = vec![];
for res in tree.iter().map_err(GarageError::from)? {
let (_k, v) = res.map_err(GarageError::from)?;
let bucket = rmp_serde::decode::from_read_ref::<_, old_bucket::Bucket>(&v[..])
.map_err(GarageError::from)?;
let bucket =
nonversioned_decode::<old_bucket::Bucket>(&v[..]).map_err(GarageError::from)?;
old_buckets.push(bucket);
}

View file

@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize};
use garage_table::crdt::Crdt;
use garage_table::*;
use super::key_table::PermissionSet;
use crate::key_table::v05::PermissionSet;
/// A bucket is a collection of objects
///

View file

@ -1,50 +0,0 @@
use serde::{Deserialize, Serialize};
use garage_table::crdt::*;
use garage_table::*;
/// An api key
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Key {
/// The id of the key (immutable), used as partition key
pub key_id: String,
/// The secret_key associated
pub secret_key: String,
/// Name for the key
pub name: crdt::Lww<String>,
/// Is the key deleted
pub deleted: crdt::Bool,
/// Buckets in which the key is authorized. Empty if `Key` is deleted
// CRDT interaction: deleted implies authorized_buckets is empty
pub authorized_buckets: crdt::LwwMap<String, PermissionSet>,
}
/// Permission given to a key in a bucket
#[derive(PartialOrd, Ord, PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct PermissionSet {
/// The key can be used to read the bucket
pub allow_read: bool,
/// The key can be used to write in the bucket
pub allow_write: bool,
}
impl AutoCrdt for PermissionSet {
const WARN_IF_DIFFERENT: bool = true;
}
impl Crdt for Key {
fn merge(&mut self, other: &Self) {
self.name.merge(&other.name);
self.deleted.merge(&other.deleted);
if self.deleted.get() {
self.authorized_buckets.clear();
} else {
self.authorized_buckets.merge(&other.authorized_buckets);
}
}
}

View file

@ -1,4 +1 @@
pub(crate) mod bucket_table;
pub(crate) mod key_table;
pub(crate) mod object_table;
pub(crate) mod version_table;

View file

@ -1,149 +0,0 @@
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
use garage_util::data::*;
use garage_table::crdt::*;
/// An object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Object {
/// The bucket in which the object is stored, used as partition key
pub bucket: String,
/// The key at which the object is stored in its bucket, used as sorting key
pub key: String,
/// The list of currenty stored versions of the object
versions: Vec<ObjectVersion>,
}
impl Object {
/// Get a list of currently stored versions of `Object`
pub fn versions(&self) -> &[ObjectVersion] {
&self.versions[..]
}
}
/// Informations about a version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersion {
/// Id of the version
pub uuid: Uuid,
/// Timestamp of when the object was created
pub timestamp: u64,
/// State of the version
pub state: ObjectVersionState,
}
/// State of an object version
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionState {
/// The version is being received
Uploading(ObjectVersionHeaders),
/// The version is fully received
Complete(ObjectVersionData),
/// The version uploaded containded errors or the upload was explicitly aborted
Aborted,
}
impl Crdt for ObjectVersionState {
fn merge(&mut self, other: &Self) {
use ObjectVersionState::*;
match other {
Aborted => {
*self = Aborted;
}
Complete(b) => match self {
Aborted => {}
Complete(a) => {
a.merge(b);
}
Uploading(_) => {
*self = Complete(b.clone());
}
},
Uploading(_) => {}
}
}
}
/// Data stored in object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionData {
/// The object was deleted, this Version is a tombstone to mark it as such
DeleteMarker,
/// The object is short, it's stored inlined
Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>),
/// The object is not short, Hash of first block is stored here, next segments hashes are
/// stored in the version table
FirstBlock(ObjectVersionMeta, Hash),
}
impl AutoCrdt for ObjectVersionData {
const WARN_IF_DIFFERENT: bool = true;
}
/// Metadata about the object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionMeta {
/// Headers to send to the client
pub headers: ObjectVersionHeaders,
/// Size of the object
pub size: u64,
/// etag of the object
pub etag: String,
}
/// Additional headers for an object
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionHeaders {
/// Content type of the object
pub content_type: String,
/// Any other http headers to send
pub other: BTreeMap<String, String>,
}
impl ObjectVersion {
fn cmp_key(&self) -> (u64, Uuid) {
(self.timestamp, self.uuid)
}
/// Is the object version completely received
pub fn is_complete(&self) -> bool {
matches!(self.state, ObjectVersionState::Complete(_))
}
}
impl Crdt for Object {
fn merge(&mut self, other: &Self) {
// Merge versions from other into here
for other_v in other.versions.iter() {
match self
.versions
.binary_search_by(|v| v.cmp_key().cmp(&other_v.cmp_key()))
{
Ok(i) => {
self.versions[i].state.merge(&other_v.state);
}
Err(i) => {
self.versions.insert(i, other_v.clone());
}
}
}
// Remove versions which are obsolete, i.e. those that come
// before the last version which .is_complete().
let last_complete = self
.versions
.iter()
.enumerate()
.rev()
.find(|(_, v)| v.is_complete())
.map(|(vi, _)| vi);
if let Some(last_vi) = last_complete {
self.versions = self.versions.drain(last_vi..).collect::<Vec<_>>();
}
}
}

View file

@ -1,79 +0,0 @@
use serde::{Deserialize, Serialize};
use garage_util::data::*;
use garage_table::crdt::*;
use garage_table::*;
/// A version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Version {
/// UUID of the version, used as partition key
pub uuid: Uuid,
// Actual data: the blocks for this version
// In the case of a multipart upload, also store the etags
// of individual parts and check them when doing CompleteMultipartUpload
/// Is this version deleted
pub deleted: crdt::Bool,
/// list of blocks of data composing the version
pub blocks: crdt::Map<VersionBlockKey, VersionBlock>,
/// Etag of each part in case of a multipart upload, empty otherwise
pub parts_etags: crdt::Map<u64, String>,
// Back link to bucket+key so that we can figure if
// this was deleted later on
/// Bucket in which the related object is stored
pub bucket: String,
/// Key in which the related object is stored
pub key: String,
}
#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlockKey {
/// Number of the part
pub part_number: u64,
/// Offset of this sub-segment in its part
pub offset: u64,
}
impl Ord for VersionBlockKey {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.part_number
.cmp(&other.part_number)
.then(self.offset.cmp(&other.offset))
}
}
impl PartialOrd for VersionBlockKey {
fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
Some(self.cmp(other))
}
}
/// Informations about a single block
#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlock {
/// Blake2 sum of the block
pub hash: Hash,
/// Size of the block
pub size: u64,
}
impl AutoCrdt for VersionBlock {
const WARN_IF_DIFFERENT: bool = true;
}
impl Crdt for Version {
fn merge(&mut self, other: &Self) {
self.deleted.merge(&other.deleted);
if self.deleted.get() {
self.blocks.clear();
self.parts_etags.clear();
} else {
self.blocks.merge(&other.blocks);
self.parts_etags.merge(&other.parts_etags);
}
}
}

View file

@ -1,4 +1,3 @@
use serde::{Deserialize, Serialize};
use std::sync::Arc;
use garage_db as db;
@ -10,19 +9,29 @@ use garage_table::*;
use garage_block::manager::*;
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BlockRef {
/// Hash (blake2 sum) of the block, used as partition key
pub block: Hash,
mod v08 {
use garage_util::crdt;
use garage_util::data::{Hash, Uuid};
use serde::{Deserialize, Serialize};
/// Id of the Version for the object containing this block, used as sorting key
pub version: Uuid,
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct BlockRef {
/// Hash (blake2 sum) of the block, used as partition key
pub block: Hash,
// Keep track of deleted status
/// Is the Version that contains this block deleted
pub deleted: crdt::Bool,
/// Id of the Version for the object containing this block, used as sorting key
pub version: Uuid,
// Keep track of deleted status
/// Is the Version that contains this block deleted
pub deleted: crdt::Bool,
}
impl garage_util::migrate::InitialFormat for BlockRef {}
}
pub use v08::*;
impl Entry<Hash, Uuid> for BlockRef {
fn partition_key(&self) -> &Hash {
&self.block

View file

@ -1,10 +1,8 @@
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
use std::sync::Arc;
use garage_db as db;
use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_table::crdt::*;
@ -14,25 +12,126 @@ use garage_table::*;
use crate::index_counter::*;
use crate::s3::version_table::*;
use crate::prev::v051::object_table as old;
pub const OBJECTS: &str = "objects";
pub const UNFINISHED_UPLOADS: &str = "unfinished_uploads";
pub const BYTES: &str = "bytes";
/// An object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Object {
/// The bucket in which the object is stored, used as partition key
pub bucket_id: Uuid,
mod v05 {
use garage_util::data::{Hash, Uuid};
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;
/// The key at which the object is stored in its bucket, used as sorting key
pub key: String,
/// An object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Object {
/// The bucket in which the object is stored, used as partition key
pub bucket: String,
/// The list of currenty stored versions of the object
versions: Vec<ObjectVersion>,
/// The key at which the object is stored in its bucket, used as sorting key
pub key: String,
/// The list of currenty stored versions of the object
pub(super) versions: Vec<ObjectVersion>,
}
/// Informations about a version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersion {
/// Id of the version
pub uuid: Uuid,
/// Timestamp of when the object was created
pub timestamp: u64,
/// State of the version
pub state: ObjectVersionState,
}
/// State of an object version
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionState {
/// The version is being received
Uploading(ObjectVersionHeaders),
/// The version is fully received
Complete(ObjectVersionData),
/// The version uploaded containded errors or the upload was explicitly aborted
Aborted,
}
/// Data stored in object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionData {
/// The object was deleted, this Version is a tombstone to mark it as such
DeleteMarker,
/// The object is short, it's stored inlined
Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>),
/// The object is not short, Hash of first block is stored here, next segments hashes are
/// stored in the version table
FirstBlock(ObjectVersionMeta, Hash),
}
/// Metadata about the object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionMeta {
/// Headers to send to the client
pub headers: ObjectVersionHeaders,
/// Size of the object
pub size: u64,
/// etag of the object
pub etag: String,
}
/// Additional headers for an object
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionHeaders {
/// Content type of the object
pub content_type: String,
/// Any other http headers to send
pub other: BTreeMap<String, String>,
}
impl garage_util::migrate::InitialFormat for Object {}
}
mod v08 {
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
use super::v05;
pub use v05::{
ObjectVersion, ObjectVersionData, ObjectVersionHeaders, ObjectVersionMeta,
ObjectVersionState,
};
/// An object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Object {
/// The bucket in which the object is stored, used as partition key
pub bucket_id: Uuid,
/// The key at which the object is stored in its bucket, used as sorting key
pub key: String,
/// The list of currenty stored versions of the object
pub(super) versions: Vec<ObjectVersion>,
}
impl garage_util::migrate::Migrate for Object {
type Previous = v05::Object;
fn migrate(old: v05::Object) -> Object {
use garage_util::data::blake2sum;
Object {
bucket_id: blake2sum(old.bucket.as_bytes()),
key: old.key,
versions: old.versions,
}
}
}
}
pub use v08::*;
impl Object {
/// Initialize an Object struct from parts
pub fn new(bucket_id: Uuid, key: String, versions: Vec<ObjectVersion>) -> Self {
@ -69,28 +168,6 @@ impl Object {
}
}
/// Informations about a version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersion {
/// Id of the version
pub uuid: Uuid,
/// Timestamp of when the object was created
pub timestamp: u64,
/// State of the version
pub state: ObjectVersionState,
}
/// State of an object version
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionState {
/// The version is being received
Uploading(ObjectVersionHeaders),
/// The version is fully received
Complete(ObjectVersionData),
/// The version uploaded containded errors or the upload was explicitly aborted
Aborted,
}
impl Crdt for ObjectVersionState {
fn merge(&mut self, other: &Self) {
use ObjectVersionState::*;
@ -112,42 +189,10 @@ impl Crdt for ObjectVersionState {
}
}
/// Data stored in object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub enum ObjectVersionData {
/// The object was deleted, this Version is a tombstone to mark it as such
DeleteMarker,
/// The object is short, it's stored inlined
Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>),
/// The object is not short, Hash of first block is stored here, next segments hashes are
/// stored in the version table
FirstBlock(ObjectVersionMeta, Hash),
}
impl AutoCrdt for ObjectVersionData {
const WARN_IF_DIFFERENT: bool = true;
}
/// Metadata about the object version
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionMeta {
/// Headers to send to the client
pub headers: ObjectVersionHeaders,
/// Size of the object
pub size: u64,
/// etag of the object
pub etag: String,
}
/// Additional headers for an object
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct ObjectVersionHeaders {
/// Content type of the object
pub content_type: String,
/// Any other http headers to send
pub other: BTreeMap<String, String>,
}
impl ObjectVersion {
fn cmp_key(&self) -> (u64, Uuid) {
(self.timestamp, self.uuid)
@ -221,7 +266,6 @@ impl Crdt for Object {
}
pub struct ObjectTable {
pub background: Arc<BackgroundRunner>,
pub version_table: Arc<Table<VersionTable, TableShardedReplication>>,
pub object_counter_table: Arc<IndexCounter<Object>>,
}
@ -255,34 +299,34 @@ impl TableSchema for ObjectTable {
);
}
// 2. Spawn threads that propagates deletions to version table
let version_table = self.version_table.clone();
let old = old.cloned();
let new = new.cloned();
self.background.spawn(async move {
if let (Some(old_v), Some(new_v)) = (old, new) {
// Propagate deletion of old versions
for v in old_v.versions.iter() {
let newly_deleted = match new_v
.versions
.binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key()))
{
Err(_) => true,
Ok(i) => {
new_v.versions[i].state == ObjectVersionState::Aborted
&& v.state != ObjectVersionState::Aborted
}
};
if newly_deleted {
let deleted_version =
Version::new(v.uuid, old_v.bucket_id, old_v.key.clone(), true);
version_table.insert(&deleted_version).await?;
// 2. Enqueue propagation deletions to version table
if let (Some(old_v), Some(new_v)) = (old, new) {
// Propagate deletion of old versions
for v in old_v.versions.iter() {
let newly_deleted = match new_v
.versions
.binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key()))
{
Err(_) => true,
Ok(i) => {
new_v.versions[i].state == ObjectVersionState::Aborted
&& v.state != ObjectVersionState::Aborted
}
};
if newly_deleted {
let deleted_version =
Version::new(v.uuid, old_v.bucket_id, old_v.key.clone(), true);
let res = self.version_table.queue_insert(tx, &deleted_version);
if let Err(e) = db::unabort(res)? {
error!(
"Unable to enqueue version deletion propagation: {}. A repair will be needed.",
e
);
}
}
}
Ok(())
});
}
Ok(())
}
@ -292,11 +336,6 @@ impl TableSchema for ObjectTable {
ObjectFilter::IsUploading => entry.versions.iter().any(|v| v.is_uploading()),
}
}
fn try_migrate(bytes: &[u8]) -> Option<Self::E> {
let old_obj = rmp_serde::decode::from_read_ref::<_, old::Object>(bytes).ok()?;
Some(migrate_object(old_obj))
}
}
impl CountedItem for Object {
@ -341,64 +380,3 @@ impl CountedItem for Object {
]
}
}
// vvvvvvvv migration code, stupid stuff vvvvvvvvvvvv
// (we just want to change bucket into bucket_id by hashing it)
fn migrate_object(o: old::Object) -> Object {
let versions = o
.versions()
.iter()
.cloned()
.map(migrate_object_version)
.collect();
Object {
bucket_id: blake2sum(o.bucket.as_bytes()),
key: o.key,
versions,
}
}
fn migrate_object_version(v: old::ObjectVersion) -> ObjectVersion {
ObjectVersion {
uuid: Uuid::try_from(v.uuid.as_slice()).unwrap(),
timestamp: v.timestamp,
state: match v.state {
old::ObjectVersionState::Uploading(h) => {
ObjectVersionState::Uploading(migrate_object_version_headers(h))
}
old::ObjectVersionState::Complete(d) => {
ObjectVersionState::Complete(migrate_object_version_data(d))
}
old::ObjectVersionState::Aborted => ObjectVersionState::Aborted,
},
}
}
fn migrate_object_version_headers(h: old::ObjectVersionHeaders) -> ObjectVersionHeaders {
ObjectVersionHeaders {
content_type: h.content_type,
other: h.other,
}
}
fn migrate_object_version_data(d: old::ObjectVersionData) -> ObjectVersionData {
match d {
old::ObjectVersionData::DeleteMarker => ObjectVersionData::DeleteMarker,
old::ObjectVersionData::Inline(m, b) => {
ObjectVersionData::Inline(migrate_object_version_meta(m), b)
}
old::ObjectVersionData::FirstBlock(m, h) => ObjectVersionData::FirstBlock(
migrate_object_version_meta(m),
Hash::try_from(h.as_slice()).unwrap(),
),
}
}
fn migrate_object_version_meta(m: old::ObjectVersionMeta) -> ObjectVersionMeta {
ObjectVersionMeta {
headers: migrate_object_version_headers(m.headers),
size: m.size,
etag: m.etag,
}
}

View file

@ -1,9 +1,7 @@
use serde::{Deserialize, Serialize};
use std::sync::Arc;
use garage_db as db;
use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_table::crdt::*;
@ -12,32 +10,108 @@ use garage_table::*;
use crate::s3::block_ref_table::*;
use crate::prev::v051::version_table as old;
mod v05 {
use garage_util::crdt;
use garage_util::data::{Hash, Uuid};
use serde::{Deserialize, Serialize};
/// A version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Version {
/// UUID of the version, used as partition key
pub uuid: Uuid,
/// A version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Version {
/// UUID of the version, used as partition key
pub uuid: Uuid,
// Actual data: the blocks for this version
// In the case of a multipart upload, also store the etags
// of individual parts and check them when doing CompleteMultipartUpload
/// Is this version deleted
pub deleted: crdt::Bool,
/// list of blocks of data composing the version
pub blocks: crdt::Map<VersionBlockKey, VersionBlock>,
/// Etag of each part in case of a multipart upload, empty otherwise
pub parts_etags: crdt::Map<u64, String>,
// Actual data: the blocks for this version
// In the case of a multipart upload, also store the etags
// of individual parts and check them when doing CompleteMultipartUpload
/// Is this version deleted
pub deleted: crdt::Bool,
/// list of blocks of data composing the version
pub blocks: crdt::Map<VersionBlockKey, VersionBlock>,
/// Etag of each part in case of a multipart upload, empty otherwise
pub parts_etags: crdt::Map<u64, String>,
// Back link to bucket+key so that we can figure if
// this was deleted later on
/// Bucket in which the related object is stored
pub bucket_id: Uuid,
/// Key in which the related object is stored
pub key: String,
// Back link to bucket+key so that we can figure if
// this was deleted later on
/// Bucket in which the related object is stored
pub bucket: String,
/// Key in which the related object is stored
pub key: String,
}
#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlockKey {
/// Number of the part
pub part_number: u64,
/// Offset of this sub-segment in its part
pub offset: u64,
}
/// Informations about a single block
#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlock {
/// Blake2 sum of the block
pub hash: Hash,
/// Size of the block
pub size: u64,
}
impl garage_util::migrate::InitialFormat for Version {}
}
mod v08 {
use garage_util::crdt;
use garage_util::data::Uuid;
use serde::{Deserialize, Serialize};
use super::v05;
/// A version of an object
#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
pub struct Version {
/// UUID of the version, used as partition key
pub uuid: Uuid,
// Actual data: the blocks for this version
// In the case of a multipart upload, also store the etags
// of individual parts and check them when doing CompleteMultipartUpload
/// Is this version deleted
pub deleted: crdt::Bool,
/// list of blocks of data composing the version
pub blocks: crdt::Map<VersionBlockKey, VersionBlock>,
/// Etag of each part in case of a multipart upload, empty otherwise
pub parts_etags: crdt::Map<u64, String>,
// Back link to bucket+key so that we can figure if
// this was deleted later on
/// Bucket in which the related object is stored
pub bucket_id: Uuid,
/// Key in which the related object is stored
pub key: String,
}
pub use v05::{VersionBlock, VersionBlockKey};
impl garage_util::migrate::Migrate for Version {
type Previous = v05::Version;
fn migrate(old: v05::Version) -> Version {
use garage_util::data::blake2sum;
Version {
uuid: old.uuid,
deleted: old.deleted,
blocks: old.blocks,
parts_etags: old.parts_etags,
bucket_id: blake2sum(old.bucket.as_bytes()),
key: old.key,
}
}
}
}
pub use v08::*;
impl Version {
pub fn new(uuid: Uuid, bucket_id: Uuid, key: String, deleted: bool) -> Self {
Self {
@ -65,14 +139,6 @@ impl Version {
}
}
#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlockKey {
/// Number of the part
pub part_number: u64,
/// Offset of this sub-segment in its part
pub offset: u64,
}
impl Ord for VersionBlockKey {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.part_number
@ -87,15 +153,6 @@ impl PartialOrd for VersionBlockKey {
}
}
/// Informations about a single block
#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)]
pub struct VersionBlock {
/// Blake2 sum of the block
pub hash: Hash,
/// Size of the block
pub size: u64,
}
impl AutoCrdt for VersionBlock {
const WARN_IF_DIFFERENT: bool = true;
}
@ -127,7 +184,6 @@ impl Crdt for Version {
}
pub struct VersionTable {
pub background: Arc<BackgroundRunner>,
pub block_ref_table: Arc<Table<BlockRefTable, TableShardedReplication>>,
}
@ -141,33 +197,26 @@ impl TableSchema for VersionTable {
fn updated(
&self,
_tx: &mut db::Transaction,
tx: &mut db::Transaction,
old: Option<&Self::E>,
new: Option<&Self::E>,
) -> db::TxOpResult<()> {
let block_ref_table = self.block_ref_table.clone();
let old = old.cloned();
let new = new.cloned();
self.background.spawn(async move {
if let (Some(old_v), Some(new_v)) = (old, new) {
// Propagate deletion of version blocks
if new_v.deleted.get() && !old_v.deleted.get() {
let deleted_block_refs = old_v
.blocks
.items()
.iter()
.map(|(_k, vb)| BlockRef {
block: vb.hash,
version: old_v.uuid,
deleted: true.into(),
})
.collect::<Vec<_>>();
block_ref_table.insert_many(&deleted_block_refs[..]).await?;
if let (Some(old_v), Some(new_v)) = (old, new) {
// Propagate deletion of version blocks
if new_v.deleted.get() && !old_v.deleted.get() {
let deleted_block_refs = old_v.blocks.items().iter().map(|(_k, vb)| BlockRef {
block: vb.hash,
version: old_v.uuid,
deleted: true.into(),
});
for block_ref in deleted_block_refs {
let res = self.block_ref_table.queue_insert(tx, &block_ref);
if let Err(e) = db::unabort(res)? {
error!("Unable to enqueue block ref deletion propagation: {}. A repair will be needed.", e);
}
}
}
Ok(())
});
}
Ok(())
}
@ -175,42 +224,4 @@ impl TableSchema for VersionTable {
fn matches_filter(entry: &Self::E, filter: &Self::Filter) -> bool {
filter.apply(entry.deleted.get())
}
fn try_migrate(bytes: &[u8]) -> Option<Self::E> {
let old = rmp_serde::decode::from_read_ref::<_, old::Version>(bytes).ok()?;
let blocks = old
.blocks
.items()
.iter()
.map(|(k, v)| {
(
VersionBlockKey {
part_number: k.part_number,
offset: k.offset,
},
VersionBlock {
hash: Hash::try_from(v.hash.as_slice()).unwrap(),
size: v.size,
},
)
})
.collect::<crdt::Map<_, _>>();
let parts_etags = old
.parts_etags
.items()
.iter()
.map(|(k, v)| (*k, v.clone()))
.collect::<crdt::Map<_, _>>();
Some(Version {
uuid: Hash::try_from(old.uuid.as_slice()).unwrap(),
deleted: crdt::Bool::new(old.deleted.get()),
blocks,
parts_etags,
bucket_id: blake2sum(old.bucket.as_bytes()),
key: old.key,
})
}
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_rpc"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,7 +14,7 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_util = { version = "0.8.0", path = "../util" }
garage_util = { version = "0.8.1", path = "../util" }
arc-swap = "1.0"
bytes = "1.0"
@ -25,7 +25,6 @@ rand = "0.8"
sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
async-trait = "0.1.7"
rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"
serde_json = "1.0"

View file

@ -5,6 +5,7 @@ use serde::{Deserialize, Serialize};
use garage_util::crdt::{AutoCrdt, Crdt, LwwMap};
use garage_util::data::*;
use garage_util::encode::nonversioned_encode;
use garage_util::error::*;
use crate::ring::*;
@ -35,6 +36,8 @@ pub struct ClusterLayout {
pub staging_hash: Hash,
}
impl garage_util::migrate::InitialFormat for ClusterLayout {}
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
pub struct NodeRoleV(pub Option<NodeRole>);
@ -68,7 +71,7 @@ impl NodeRole {
impl ClusterLayout {
pub fn new(replication_factor: usize) -> Self {
let empty_lwwmap = LwwMap::new();
let empty_lwwmap_hash = blake2sum(&rmp_to_vec_all_named(&empty_lwwmap).unwrap()[..]);
let empty_lwwmap_hash = blake2sum(&nonversioned_encode(&empty_lwwmap).unwrap()[..]);
ClusterLayout {
version: 0,
@ -90,7 +93,7 @@ impl ClusterLayout {
Ordering::Equal => {
self.staging.merge(&other.staging);
let new_staging_hash = blake2sum(&rmp_to_vec_all_named(&self.staging).unwrap()[..]);
let new_staging_hash = blake2sum(&nonversioned_encode(&self.staging).unwrap()[..]);
let changed = new_staging_hash != self.staging_hash;
self.staging_hash = new_staging_hash;
@ -125,7 +128,7 @@ To know the correct value of the new layout version, invoke `garage layout show`
}
self.staging.clear();
self.staging_hash = blake2sum(&rmp_to_vec_all_named(&self.staging).unwrap()[..]);
self.staging_hash = blake2sum(&nonversioned_encode(&self.staging).unwrap()[..]);
self.version += 1;
@ -149,7 +152,7 @@ To know the correct value of the new layout version, invoke `garage layout show`
}
self.staging.clear();
self.staging_hash = blake2sum(&rmp_to_vec_all_named(&self.staging).unwrap()[..]);
self.staging_hash = blake2sum(&nonversioned_encode(&self.staging).unwrap()[..]);
self.version += 1;
@ -178,7 +181,7 @@ To know the correct value of the new layout version, invoke `garage layout show`
/// returns true if consistent, false if error
pub fn check(&self) -> bool {
// Check that the hash of the staging data is correct
let staging_hash = blake2sum(&rmp_to_vec_all_named(&self.staging).unwrap()[..]);
let staging_hash = blake2sum(&nonversioned_encode(&self.staging).unwrap()[..]);
if staging_hash != self.staging_hash {
return false;
}

View file

@ -17,3 +17,5 @@ mod metrics;
pub mod rpc_helper;
pub use rpc_helper::*;
pub mod system_metrics;

View file

@ -5,7 +5,6 @@ use std::time::Duration;
use futures::future::join_all;
use futures::stream::futures_unordered::FuturesUnordered;
use futures::stream::StreamExt;
use futures_util::future::FutureExt;
use tokio::select;
use tokio::sync::watch;
@ -24,7 +23,6 @@ pub use netapp::message::{
use netapp::peering::fullmesh::FullMeshPeeringStrategy;
pub use netapp::{self, NetApp, NodeID};
use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_util::error::Error;
use garage_util::metrics::RecordDuration;
@ -94,7 +92,6 @@ pub struct RpcHelper(Arc<RpcHelperInner>);
struct RpcHelperInner {
our_node_id: Uuid,
fullmesh: Arc<FullMeshPeeringStrategy>,
background: Arc<BackgroundRunner>,
ring: watch::Receiver<Arc<Ring>>,
metrics: RpcMetrics,
rpc_timeout: Duration,
@ -104,7 +101,6 @@ impl RpcHelper {
pub(crate) fn new(
our_node_id: Uuid,
fullmesh: Arc<FullMeshPeeringStrategy>,
background: Arc<BackgroundRunner>,
ring: watch::Receiver<Arc<Ring>>,
rpc_timeout: Option<Duration>,
) -> Self {
@ -113,7 +109,6 @@ impl RpcHelper {
Self(Arc::new(RpcHelperInner {
our_node_id,
fullmesh,
background,
ring,
metrics,
rpc_timeout: rpc_timeout.unwrap_or(DEFAULT_TIMEOUT),
@ -377,16 +372,13 @@ impl RpcHelper {
if !resp_stream.is_empty() {
// Continue remaining requests in background.
// Continue the remaining requests immediately using tokio::spawn
// but enqueue a task in the background runner
// to ensure that the process won't exit until the requests are done
// (if we had just enqueued the resp_stream.collect directly in the background runner,
// the requests might have been put on hold in the background runner's queue,
// in which case they might timeout or otherwise fail)
let wait_finished_fut = tokio::spawn(async move {
// Note: these requests can get interrupted on process shutdown,
// we must not count on them being executed for certain.
// For all background things that have to happen with certainty,
// they have to be put in a proper queue that is persisted to disk.
tokio::spawn(async move {
resp_stream.collect::<Vec<Result<_, _>>>().await;
});
self.0.background.spawn(wait_finished_fut.map(|_| Ok(())));
}
}

View file

@ -21,7 +21,6 @@ use netapp::peering::fullmesh::FullMeshPeeringStrategy;
use netapp::util::parse_and_resolve_peer_addr_async;
use netapp::{NetApp, NetworkKey, NodeID, NodeKey};
use garage_util::background::BackgroundRunner;
use garage_util::config::Config;
#[cfg(feature = "kubernetes-discovery")]
use garage_util::config::KubernetesDiscoveryConfig;
@ -39,6 +38,9 @@ use crate::replication_mode::*;
use crate::ring::*;
use crate::rpc_helper::*;
#[cfg(feature = "metrics")]
use crate::system_metrics::*;
const DISCOVERY_INTERVAL: Duration = Duration::from_secs(60);
const STATUS_EXCHANGE_INTERVAL: Duration = Duration::from_secs(10);
@ -50,8 +52,6 @@ pub const GARAGE_VERSION_TAG: u64 = 0x6761726167650008; // garage 0x0008
/// RPC endpoint used for calls related to membership
pub const SYSTEM_RPC_PATH: &str = "garage_rpc/membership.rs/SystemRpc";
pub const CONNECT_ERROR_MESSAGE: &str = "Error establishing RPC connection to remote node. This can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret";
/// RPC messages related to membership
#[derive(Debug, Serialize, Deserialize, Clone)]
pub enum SystemRpc {
@ -76,13 +76,17 @@ impl Rpc for SystemRpc {
type Response = Result<SystemRpc, Error>;
}
#[derive(Serialize, Deserialize)]
pub struct PeerList(Vec<(Uuid, SocketAddr)>);
impl garage_util::migrate::InitialFormat for PeerList {}
/// This node's membership manager
pub struct System {
/// The id of this node
pub id: Uuid,
persist_cluster_layout: Persister<ClusterLayout>,
persist_peer_list: Persister<Vec<(Uuid, SocketAddr)>>,
persist_peer_list: Persister<PeerList>,
local_status: ArcSwap<NodeStatus>,
node_status: RwLock<HashMap<Uuid, (u64, NodeStatus)>>,
@ -102,6 +106,8 @@ pub struct System {
consul_discovery: Option<ConsulDiscovery>,
#[cfg(feature = "kubernetes-discovery")]
kubernetes_discovery: Option<KubernetesDiscoveryConfig>,
#[cfg(feature = "metrics")]
metrics: SystemMetrics,
replication_mode: ReplicationMode,
replication_factor: usize,
@ -110,9 +116,6 @@ pub struct System {
pub ring: watch::Receiver<Arc<Ring>>,
update_ring: Mutex<watch::Sender<Arc<Ring>>>,
/// The job runner of this node
pub background: Arc<BackgroundRunner>,
/// Path to metadata directory
pub metadata_dir: PathBuf,
}
@ -232,7 +235,6 @@ impl System {
/// Create this node's membership manager
pub fn new(
network_key: NetworkKey,
background: Arc<BackgroundRunner>,
replication_mode: ReplicationMode,
config: &Config,
) -> Result<Arc<Self>, Error> {
@ -278,6 +280,9 @@ impl System {
cluster_layout_staging_hash: cluster_layout.staging_hash,
};
#[cfg(feature = "metrics")]
let metrics = SystemMetrics::new(replication_factor);
let ring = Ring::new(cluster_layout, replication_factor);
let (update_ring, ring) = watch::channel(Arc::new(ring));
@ -354,7 +359,6 @@ impl System {
rpc: RpcHelper::new(
netapp.id.into(),
fullmesh,
background.clone(),
ring.clone(),
config.rpc_timeout_msec.map(Duration::from_millis),
),
@ -369,10 +373,11 @@ impl System {
consul_discovery,
#[cfg(feature = "kubernetes-discovery")]
kubernetes_discovery: config.kubernetes_discovery.clone(),
#[cfg(feature = "metrics")]
metrics,
ring,
update_ring: Mutex::new(update_ring),
background,
metadata_dir: config.metadata_dir.clone(),
});
sys.system_endpoint.set_handler(sys.clone());
@ -444,17 +449,14 @@ impl System {
))
})?;
let mut errors = vec![];
for ip in addrs.iter() {
match self
.netapp
.clone()
.try_connect(*ip, pubkey)
.await
.err_context(CONNECT_ERROR_MESSAGE)
{
for addr in addrs.iter() {
match self.netapp.clone().try_connect(*addr, pubkey).await {
Ok(()) => return Ok(()),
Err(e) => {
errors.push((*ip, e));
errors.push((
*addr,
Error::Message(connect_error_message(*addr, pubkey, e)),
));
}
}
}
@ -529,56 +531,61 @@ impl System {
// ---- INTERNALS ----
#[cfg(feature = "consul-discovery")]
async fn advertise_to_consul(self: Arc<Self>) -> Result<(), Error> {
async fn advertise_to_consul(self: Arc<Self>) {
let c = match &self.consul_discovery {
Some(c) => c,
_ => return Ok(()),
_ => return,
};
let rpc_public_addr = match self.rpc_public_addr {
Some(addr) => addr,
None => {
warn!("Not advertising to Consul because rpc_public_addr is not defined in config file and could not be autodetected.");
return Ok(());
return;
}
};
c.publish_consul_service(
self.netapp.id,
&self.local_status.load_full().hostname,
rpc_public_addr,
)
.await
.err_context("Error while publishing Consul service")
if let Err(e) = c
.publish_consul_service(
self.netapp.id,
&self.local_status.load_full().hostname,
rpc_public_addr,
)
.await
{
error!("Error while publishing Consul service: {}", e);
}
}
#[cfg(feature = "kubernetes-discovery")]
async fn advertise_to_kubernetes(self: Arc<Self>) -> Result<(), Error> {
async fn advertise_to_kubernetes(self: Arc<Self>) {
let k = match &self.kubernetes_discovery {
Some(k) => k,
_ => return Ok(()),
_ => return,
};
let rpc_public_addr = match self.rpc_public_addr {
Some(addr) => addr,
None => {
warn!("Not advertising to Kubernetes because rpc_public_addr is not defined in config file and could not be autodetected.");
return Ok(());
return;
}
};
publish_kubernetes_node(
if let Err(e) = publish_kubernetes_node(
k,
self.netapp.id,
&self.local_status.load_full().hostname,
rpc_public_addr,
)
.await
.err_context("Error while publishing node to kubernetes")
{
error!("Error while publishing node to Kubernetes: {}", e);
}
}
/// Save network configuration to disc
async fn save_cluster_layout(self: Arc<Self>) -> Result<(), Error> {
async fn save_cluster_layout(&self) -> Result<(), Error> {
let ring: Arc<Ring> = self.ring.borrow().clone();
self.persist_cluster_layout
.save_async(&ring.layout)
@ -630,11 +637,7 @@ impl System {
if info.cluster_layout_version > local_info.cluster_layout_version
|| info.cluster_layout_staging_hash != local_info.cluster_layout_staging_hash
{
let self2 = self.clone();
self.background.spawn_cancellable(async move {
self2.pull_cluster_layout(from).await;
Ok(())
});
tokio::spawn(self.clone().pull_cluster_layout(from));
}
self.node_status
@ -676,18 +679,21 @@ impl System {
drop(update_ring);
let self2 = self.clone();
self.background.spawn_cancellable(async move {
self2
tokio::spawn(async move {
if let Err(e) = self2
.rpc
.broadcast(
&self2.system_endpoint,
SystemRpc::AdvertiseClusterLayout(layout),
RequestStrategy::with_priority(PRIO_HIGH),
)
.await?;
Ok(())
.await
{
warn!("Error while broadcasting new cluster layout: {}", e);
}
});
self.background.spawn(self.clone().save_cluster_layout());
self.save_cluster_layout().await?;
}
Ok(SystemRpc::Ok)
@ -734,7 +740,7 @@ impl System {
// Add peer list from list stored on disk
if let Ok(peers) = self.persist_peer_list.load_async().await {
ping_list.extend(peers.iter().map(|(id, addr)| ((*id).into(), *addr)))
ping_list.extend(peers.0.iter().map(|(id, addr)| ((*id).into(), *addr)))
}
// Fetch peer list from Consul
@ -773,12 +779,12 @@ impl System {
}
for (node_id, node_addr) in ping_list {
tokio::spawn(
self.netapp
.clone()
.try_connect(node_addr, node_id)
.map(|r| r.err_context(CONNECT_ERROR_MESSAGE)),
);
let self2 = self.clone();
tokio::spawn(async move {
if let Err(e) = self2.netapp.clone().try_connect(node_addr, node_id).await {
error!("{}", connect_error_message(node_addr, node_id, e));
}
});
}
}
@ -787,11 +793,10 @@ impl System {
}
#[cfg(feature = "consul-discovery")]
self.background.spawn(self.clone().advertise_to_consul());
tokio::spawn(self.clone().advertise_to_consul());
#[cfg(feature = "kubernetes-discovery")]
self.background
.spawn(self.clone().advertise_to_kubernetes());
tokio::spawn(self.clone().advertise_to_kubernetes());
let restart_at = tokio::time::sleep(DISCOVERY_INTERVAL);
select! {
@ -815,12 +820,16 @@ impl System {
// and append it to the list we are about to save,
// so that no peer ID gets lost in the process.
if let Ok(mut prev_peer_list) = self.persist_peer_list.load_async().await {
prev_peer_list.retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id));
peer_list.extend(prev_peer_list);
prev_peer_list
.0
.retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id));
peer_list.extend(prev_peer_list.0);
}
// Save new peer list to file
self.persist_peer_list.save_async(&peer_list).await
self.persist_peer_list
.save_async(&PeerList(peer_list))
.await
}
async fn pull_cluster_layout(self: Arc<Self>, peer: Uuid) {
@ -881,3 +890,11 @@ async fn resolve_peers(peers: &[String]) -> Vec<(NodeID, SocketAddr)> {
ret
}
fn connect_error_message(
addr: SocketAddr,
pubkey: ed25519::PublicKey,
e: netapp::error::Error,
) -> String {
format!("Error establishing RPC connection to remote node: {}@{}.\nThis can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret.\n{}", hex::encode(pubkey), addr, e)
}

33
src/rpc/system_metrics.rs Normal file
View file

@ -0,0 +1,33 @@
use opentelemetry::{global, metrics::*, KeyValue};
/// TableMetrics reference all counter used for metrics
pub struct SystemMetrics {
pub(crate) _garage_build_info: ValueObserver<u64>,
pub(crate) _replication_factor: ValueObserver<u64>,
}
impl SystemMetrics {
pub fn new(replication_factor: usize) -> Self {
let meter = global::meter("garage_system");
Self {
_garage_build_info: meter
.u64_value_observer("garage_build_info", move |observer| {
observer.observe(
1,
&[KeyValue::new(
"version",
garage_util::version::garage_version(),
)],
)
})
.with_description("Garage build info")
.init(),
_replication_factor: meter
.u64_value_observer("garage_replication_factor", move |observer| {
observer.observe(replication_factor as u64, &[])
})
.with_description("Garage replication factor setting")
.init(),
}
}
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_table"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,20 +14,20 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_db = { version = "0.8.0", path = "../db" }
garage_rpc = { version = "0.8.0", path = "../rpc" }
garage_util = { version = "0.8.0", path = "../util" }
garage_db = { version = "0.8.1", path = "../db" }
garage_rpc = { version = "0.8.1", path = "../rpc" }
garage_util = { version = "0.8.1", path = "../util" }
opentelemetry = "0.17"
async-trait = "0.1.7"
arc-swap = "1.0"
bytes = "1.0"
hex = "0.4"
hexdump = "0.1"
tracing = "0.1.30"
rand = "0.8"
rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"

View file

@ -10,6 +10,7 @@ use garage_db::counted_tree_hack::CountedTree;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::migrate::Migrate;
use garage_rpc::system::System;
@ -31,16 +32,16 @@ pub struct TableData<F: TableSchema, R: TableReplication> {
pub(crate) merkle_tree: db::Tree,
pub(crate) merkle_todo: db::Tree,
pub(crate) merkle_todo_notify: Notify,
pub(crate) insert_queue: db::Tree,
pub(crate) insert_queue_notify: Notify,
pub(crate) gc_todo: CountedTree,
pub(crate) metrics: TableMetrics,
}
impl<F, R> TableData<F, R>
where
F: TableSchema,
R: TableReplication,
{
impl<F: TableSchema, R: TableReplication> TableData<F, R> {
pub fn new(system: Arc<System>, instance: F, replication: R, db: &db::Db) -> Arc<Self> {
let store = db
.open_tree(&format!("{}:table", F::TABLE_NAME))
@ -53,12 +54,22 @@ where
.open_tree(&format!("{}:merkle_todo", F::TABLE_NAME))
.expect("Unable to open DB Merkle TODO tree");
let insert_queue = db
.open_tree(&format!("{}:insert_queue", F::TABLE_NAME))
.expect("Unable to open insert queue DB tree");
let gc_todo = db
.open_tree(&format!("{}:gc_todo_v2", F::TABLE_NAME))
.expect("Unable to open DB tree");
.expect("Unable to open GC DB tree");
let gc_todo = CountedTree::new(gc_todo).expect("Cannot count gc_todo_v2");
let metrics = TableMetrics::new(F::TABLE_NAME, merkle_todo.clone(), gc_todo.clone());
let metrics = TableMetrics::new(
F::TABLE_NAME,
store.clone(),
merkle_tree.clone(),
merkle_todo.clone(),
gc_todo.clone(),
);
Arc::new(Self {
system,
@ -68,6 +79,8 @@ where
merkle_tree,
merkle_todo,
merkle_todo_notify: Notify::new(),
insert_queue,
insert_queue_notify: Notify::new(),
gc_todo,
metrics,
})
@ -167,9 +180,8 @@ where
pub(crate) fn update_entry(&self, update_bytes: &[u8]) -> Result<(), Error> {
let update = self.decode_entry(update_bytes)?;
let tree_key = self.tree_key(update.partition_key(), update.sort_key());
self.update_entry_with(&tree_key[..], |ent| match ent {
self.update_entry_with(update.partition_key(), update.sort_key(), |ent| match ent {
Some(mut ent) => {
ent.merge(&update);
ent
@ -181,11 +193,14 @@ where
pub fn update_entry_with(
&self,
tree_key: &[u8],
partition_key: &F::P,
sort_key: &F::S,
f: impl Fn(Option<F::E>) -> F::E,
) -> Result<Option<F::E>, Error> {
let tree_key = self.tree_key(partition_key, sort_key);
let changed = self.store.db().transaction(|mut tx| {
let (old_entry, old_bytes, new_entry) = match tx.get(&self.store, tree_key)? {
let (old_entry, old_bytes, new_entry) = match tx.get(&self.store, &tree_key)? {
Some(old_bytes) => {
let old_entry = self.decode_entry(&old_bytes).map_err(db::TxError::Abort)?;
let new_entry = f(Some(old_entry.clone()));
@ -194,23 +209,24 @@ where
None => (None, None, f(None)),
};
// Scenario 1: the value changed, so of course there is a change
let value_changed = Some(&new_entry) != old_entry.as_ref();
// Changed can be true in two scenarios
// Scenario 1: the actual represented value changed,
// so of course the messagepack encoding changed as well
// Scenario 2: the value didn't change but due to a migration in the
// data format, the messagepack encoding changed. In this case
// we have to write the migrated value in the table and update
// the associated Merkle tree entry.
let new_bytes = rmp_to_vec_all_named(&new_entry)
// data format, the messagepack encoding changed. In this case,
// we also have to write the migrated value in the table and update
// the associated Merkle tree entry.
let new_bytes = new_entry
.encode()
.map_err(Error::RmpEncode)
.map_err(db::TxError::Abort)?;
let encoding_changed = Some(&new_bytes[..]) != old_bytes.as_ref().map(|x| &x[..]);
let changed = Some(&new_bytes[..]) != old_bytes.as_deref();
drop(old_bytes);
if value_changed || encoding_changed {
let new_bytes_hash = blake2sum(&new_bytes[..]);
tx.insert(&self.merkle_todo, tree_key, new_bytes_hash.as_slice())?;
tx.insert(&self.store, tree_key, new_bytes)?;
if changed {
let new_bytes_hash = blake2sum(&new_bytes);
tx.insert(&self.merkle_todo, &tree_key, new_bytes_hash.as_slice())?;
tx.insert(&self.store, &tree_key, new_bytes)?;
self.instance
.updated(&mut tx, old_entry.as_ref(), Some(&new_entry))?;
@ -236,7 +252,7 @@ where
let pk_hash = Hash::try_from(&tree_key[..32]).unwrap();
let nodes = self.replication.write_nodes(&pk_hash);
if nodes.first() == Some(&self.system.id) {
GcTodoEntry::new(tree_key.to_vec(), new_bytes_hash).save(&self.gc_todo)?;
GcTodoEntry::new(tree_key, new_bytes_hash).save(&self.gc_todo)?;
}
}
@ -252,10 +268,11 @@ where
.db()
.transaction(|mut tx| match tx.get(&self.store, k)? {
Some(cur_v) if cur_v == v => {
let old_entry = self.decode_entry(v).map_err(db::TxError::Abort)?;
tx.remove(&self.store, k)?;
tx.insert(&self.merkle_todo, k, vec![])?;
let old_entry = self.decode_entry(v).map_err(db::TxError::Abort)?;
self.instance.updated(&mut tx, Some(&old_entry), None)?;
Ok(true)
}
@ -279,10 +296,11 @@ where
.db()
.transaction(|mut tx| match tx.get(&self.store, k)? {
Some(cur_v) if blake2sum(&cur_v[..]) == vhash => {
let old_entry = self.decode_entry(&cur_v[..]).map_err(db::TxError::Abort)?;
tx.remove(&self.store, k)?;
tx.insert(&self.merkle_todo, k, vec![])?;
let old_entry = self.decode_entry(&cur_v[..]).map_err(db::TxError::Abort)?;
self.instance.updated(&mut tx, Some(&old_entry), None)?;
Ok(true)
}
@ -296,6 +314,32 @@ where
Ok(removed)
}
// ---- Insert queue functions ----
pub(crate) fn queue_insert(
&self,
tx: &mut db::Transaction,
ins: &F::E,
) -> db::TxResult<(), Error> {
let tree_key = self.tree_key(ins.partition_key(), ins.sort_key());
let new_entry = match tx.get(&self.insert_queue, &tree_key)? {
Some(old_v) => {
let mut entry = self.decode_entry(&old_v).map_err(db::TxError::Abort)?;
entry.merge(ins);
entry.encode()
}
None => ins.encode(),
};
let new_entry = new_entry
.map_err(Error::RmpEncode)
.map_err(db::TxError::Abort)?;
tx.insert(&self.insert_queue, &tree_key, new_entry)?;
self.insert_queue_notify.notify_one();
Ok(())
}
// ---- Utility functions ----
pub fn tree_key(&self, p: &F::P, s: &F::S) -> Vec<u8> {
@ -305,18 +349,18 @@ where
}
pub fn decode_entry(&self, bytes: &[u8]) -> Result<F::E, Error> {
match rmp_serde::decode::from_read_ref::<_, F::E>(bytes) {
Ok(x) => Ok(x),
Err(e) => match F::try_migrate(bytes) {
Some(x) => Ok(x),
None => {
warn!("Unable to decode entry of {}: {}", F::TABLE_NAME, e);
for line in hexdump::hexdump_iter(bytes) {
debug!("{}", line);
}
Err(e.into())
match F::E::decode(bytes) {
Some(x) => Ok(x),
None => {
error!("Unable to decode entry of {}", F::TABLE_NAME);
for line in hexdump::hexdump_iter(bytes) {
debug!("{}", line);
}
},
Err(Error::Message(format!(
"Unable to decode entry of {}",
F::TABLE_NAME
)))
}
}
}

View file

@ -31,7 +31,7 @@ const TABLE_GC_BATCH_SIZE: usize = 1024;
// and the moment the garbage collection actually happens)
const TABLE_GC_DELAY: Duration = Duration::from_secs(24 * 3600);
pub(crate) struct TableGc<F: TableSchema + 'static, R: TableReplication + 'static> {
pub(crate) struct TableGc<F: TableSchema, R: TableReplication> {
system: Arc<System>,
data: Arc<TableData<F, R>>,
@ -49,29 +49,26 @@ impl Rpc for GcRpc {
type Response = Result<GcRpc, Error>;
}
impl<F, R> TableGc<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
pub(crate) fn launch(system: Arc<System>, data: Arc<TableData<F, R>>) -> Arc<Self> {
impl<F: TableSchema, R: TableReplication> TableGc<F, R> {
pub(crate) fn new(system: Arc<System>, data: Arc<TableData<F, R>>) -> Arc<Self> {
let endpoint = system
.netapp
.endpoint(format!("garage_table/gc.rs/Rpc:{}", F::TABLE_NAME));
let gc = Arc::new(Self {
system: system.clone(),
system,
data,
endpoint,
});
gc.endpoint.set_handler(gc.clone());
system.background.spawn_worker(GcWorker::new(gc.clone()));
gc
}
pub(crate) fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) {
bg.spawn_worker(GcWorker::new(self.clone()));
}
async fn gc_loop_iter(&self) -> Result<Option<Duration>, Error> {
let now = now_msec();
@ -276,11 +273,7 @@ where
}
#[async_trait]
impl<F, R> EndpointHandler<GcRpc> for TableGc<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> EndpointHandler<GcRpc> for TableGc<F, R> {
async fn handle(self: &Arc<Self>, message: &GcRpc, _from: NodeID) -> Result<GcRpc, Error> {
match message {
GcRpc::Update(items) => {
@ -298,20 +291,12 @@ where
}
}
struct GcWorker<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
struct GcWorker<F: TableSchema, R: TableReplication> {
gc: Arc<TableGc<F, R>>,
wait_delay: Duration,
}
impl<F, R> GcWorker<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> GcWorker<F, R> {
fn new(gc: Arc<TableGc<F, R>>) -> Self {
Self {
gc,
@ -321,21 +306,15 @@ where
}
#[async_trait]
impl<F, R> Worker for GcWorker<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> Worker for GcWorker<F, R> {
fn name(&self) -> String {
format!("{} GC", F::TABLE_NAME)
}
fn info(&self) -> Option<String> {
let l = self.gc.data.gc_todo_len().unwrap_or(0);
if l > 0 {
Some(format!("{} items in queue", l))
} else {
None
fn status(&self) -> WorkerStatus {
WorkerStatus {
queue_length: Some(self.gc.data.gc_todo_len().unwrap_or(0) as u64),
..Default::default()
}
}
@ -349,10 +328,7 @@ where
}
}
async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState {
if *must_exit.borrow() {
return WorkerState::Done;
}
async fn wait_for_work(&mut self) -> WorkerState {
tokio::time::sleep(self.wait_delay).await;
WorkerState::Busy
}

View file

@ -4,16 +4,18 @@
#[macro_use]
extern crate tracing;
mod metrics;
pub mod schema;
pub mod util;
pub mod data;
pub mod replication;
pub mod table;
mod gc;
mod merkle;
pub mod replication;
mod metrics;
mod queue;
mod sync;
pub mod table;
pub use schema::*;
pub use table::*;

View file

@ -3,12 +3,14 @@ use std::time::Duration;
use async_trait::async_trait;
use serde::{Deserialize, Serialize};
use tokio::select;
use tokio::sync::watch;
use garage_db as db;
use garage_util::background::*;
use garage_util::data::*;
use garage_util::encode::{nonversioned_decode, nonversioned_encode};
use garage_util::error::Error;
use garage_rpc::ring::*;
@ -64,22 +66,18 @@ pub enum MerkleNode {
Leaf(Vec<u8>, Hash),
}
impl<F, R> MerkleUpdater<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
pub(crate) fn launch(background: &BackgroundRunner, data: Arc<TableData<F, R>>) -> Arc<Self> {
let empty_node_hash = blake2sum(&rmp_to_vec_all_named(&MerkleNode::Empty).unwrap()[..]);
impl<F: TableSchema, R: TableReplication> MerkleUpdater<F, R> {
pub(crate) fn new(data: Arc<TableData<F, R>>) -> Arc<Self> {
let empty_node_hash = blake2sum(&nonversioned_encode(&MerkleNode::Empty).unwrap()[..]);
let ret = Arc::new(Self {
Arc::new(Self {
data,
empty_node_hash,
});
})
}
background.spawn_worker(MerkleWorker(ret.clone()));
ret
pub(crate) fn spawn_workers(self: &Arc<Self>, background: &BackgroundRunner) {
background.spawn_worker(MerkleWorker(self.clone()));
}
fn updater_loop_iter(&self) -> Result<WorkerState, Error> {
@ -276,7 +274,7 @@ where
tx.remove(&self.data.merkle_tree, k.encode())?;
Ok(self.empty_node_hash)
} else {
let vby = rmp_to_vec_all_named(v).map_err(|e| db::TxError::Abort(e.into()))?;
let vby = nonversioned_encode(v).map_err(|e| db::TxError::Abort(e.into()))?;
let rethash = blake2sum(&vby[..]);
tx.insert(&self.data.merkle_tree, k.encode(), vby)?;
Ok(rethash)
@ -293,32 +291,27 @@ where
Ok(self.data.merkle_tree.len()?)
}
pub fn merkle_tree_fast_len(&self) -> Result<Option<usize>, Error> {
Ok(self.data.merkle_tree.fast_len()?)
}
pub fn todo_len(&self) -> Result<usize, Error> {
Ok(self.data.merkle_todo.len()?)
}
}
struct MerkleWorker<F, R>(Arc<MerkleUpdater<F, R>>)
where
F: TableSchema + 'static,
R: TableReplication + 'static;
struct MerkleWorker<F: TableSchema, R: TableReplication>(Arc<MerkleUpdater<F, R>>);
#[async_trait]
impl<F, R> Worker for MerkleWorker<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> Worker for MerkleWorker<F, R> {
fn name(&self) -> String {
format!("{} Merkle tree updater", F::TABLE_NAME)
format!("{} Merkle", F::TABLE_NAME)
}
fn info(&self) -> Option<String> {
let l = self.0.todo_len().unwrap_or(0);
if l > 0 {
Some(format!("{} items in queue", l))
} else {
None
fn status(&self) -> WorkerStatus {
WorkerStatus {
queue_length: Some(self.0.todo_len().unwrap_or(0) as u64),
..Default::default()
}
}
@ -337,11 +330,11 @@ where
.unwrap()
}
async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState {
if *must_exit.borrow() {
return WorkerState::Done;
async fn wait_for_work(&mut self) -> WorkerState {
select! {
_ = tokio::time::sleep(Duration::from_secs(60)) => (),
_ = self.0.data.merkle_todo_notify.notified() => (),
}
tokio::time::sleep(Duration::from_secs(10)).await;
WorkerState::Busy
}
}
@ -372,7 +365,7 @@ impl MerkleNode {
fn decode_opt(ent: &Option<db::Value>) -> Result<Self, Error> {
match ent {
None => Ok(MerkleNode::Empty),
Some(v) => Ok(rmp_serde::decode::from_read_ref::<_, MerkleNode>(&v[..])?),
Some(v) => Ok(nonversioned_decode::<MerkleNode>(&v[..])?),
}
}

View file

@ -5,6 +5,8 @@ use garage_db::counted_tree_hack::CountedTree;
/// TableMetrics reference all counter used for metrics
pub struct TableMetrics {
pub(crate) _table_size: ValueObserver<u64>,
pub(crate) _merkle_tree_size: ValueObserver<u64>,
pub(crate) _merkle_todo_len: ValueObserver<u64>,
pub(crate) _gc_todo_len: ValueObserver<u64>,
@ -20,9 +22,43 @@ pub struct TableMetrics {
pub(crate) sync_items_received: Counter<u64>,
}
impl TableMetrics {
pub fn new(table_name: &'static str, merkle_todo: db::Tree, gc_todo: CountedTree) -> Self {
pub fn new(
table_name: &'static str,
store: db::Tree,
merkle_tree: db::Tree,
merkle_todo: db::Tree,
gc_todo: CountedTree,
) -> Self {
let meter = global::meter(table_name);
TableMetrics {
_table_size: meter
.u64_value_observer(
"table.size",
move |observer| {
if let Ok(Some(v)) = store.fast_len() {
observer.observe(
v as u64,
&[KeyValue::new("table_name", table_name)],
);
}
},
)
.with_description("Number of items in table")
.init(),
_merkle_tree_size: meter
.u64_value_observer(
"table.merkle_tree_size",
move |observer| {
if let Ok(Some(v)) = merkle_tree.fast_len() {
observer.observe(
v as u64,
&[KeyValue::new("table_name", table_name)],
);
}
},
)
.with_description("Number of nodes in table's Merkle tree")
.init(),
_merkle_todo_len: meter
.u64_value_observer(
"table.merkle_updater_todo_queue_length",

77
src/table/queue.rs Normal file
View file

@ -0,0 +1,77 @@
use std::sync::Arc;
use std::time::Duration;
use async_trait::async_trait;
use tokio::select;
use tokio::sync::watch;
use garage_util::background::*;
use garage_util::error::Error;
use crate::replication::*;
use crate::schema::*;
use crate::table::*;
const BATCH_SIZE: usize = 100;
pub(crate) struct InsertQueueWorker<F, R>(pub(crate) Arc<Table<F, R>>)
where
F: TableSchema,
R: TableReplication;
#[async_trait]
impl<F: TableSchema, R: TableReplication> Worker for InsertQueueWorker<F, R> {
fn name(&self) -> String {
format!("{} queue", F::TABLE_NAME)
}
fn status(&self) -> WorkerStatus {
WorkerStatus {
queue_length: Some(self.0.data.insert_queue.len().unwrap_or(0) as u64),
..Default::default()
}
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
let mut kv_pairs = vec![];
let mut values = vec![];
for entry_kv in self.0.data.insert_queue.iter()? {
let (k, v) = entry_kv?;
values.push(self.0.data.decode_entry(&v)?);
kv_pairs.push((k, v));
if kv_pairs.len() > BATCH_SIZE {
break;
}
}
if kv_pairs.is_empty() {
return Ok(WorkerState::Idle);
}
self.0.insert_many(values).await?;
self.0.data.insert_queue.db().transaction(|mut tx| {
for (k, v) in kv_pairs.iter() {
if let Some(v2) = tx.get(&self.0.data.insert_queue, k)? {
if &v2 == v {
tx.remove(&self.0.data.insert_queue, k)?;
}
}
}
Ok(())
})?;
Ok(WorkerState::Busy)
}
async fn wait_for_work(&mut self) -> WorkerState {
select! {
_ = tokio::time::sleep(Duration::from_secs(600)) => (),
_ = self.0.data.insert_queue_notify.notified() => (),
}
WorkerState::Busy
}
}

View file

@ -2,7 +2,7 @@ use garage_rpc::ring::*;
use garage_util::data::*;
/// Trait to describe how a table shall be replicated
pub trait TableReplication: Send + Sync {
pub trait TableReplication: Send + Sync + 'static {
// See examples in table_sharded.rs and table_fullcopy.rs
// To understand various replication methods

View file

@ -2,11 +2,14 @@ use serde::{Deserialize, Serialize};
use garage_db as db;
use garage_util::data::*;
use garage_util::migrate::Migrate;
use crate::crdt::Crdt;
/// Trait for field used to partition data
pub trait PartitionKey {
pub trait PartitionKey:
Clone + PartialEq + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static
{
/// Get the key used to partition
fn hash(&self) -> Hash;
}
@ -27,7 +30,7 @@ impl PartitionKey for FixedBytes32 {
}
/// Trait for field used to sort data
pub trait SortKey {
pub trait SortKey: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static {
/// Get the key used to sort
fn sort_key(&self) -> &[u8];
}
@ -46,7 +49,7 @@ impl SortKey for FixedBytes32 {
/// Trait for an entry in a table. It must be sortable and partitionnable.
pub trait Entry<P: PartitionKey, S: SortKey>:
Crdt + PartialEq + Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync
Crdt + PartialEq + Clone + Migrate + Send + Sync + 'static
{
/// Get the key used to partition
fn partition_key(&self) -> &P;
@ -65,23 +68,16 @@ pub trait TableSchema: Send + Sync + 'static {
const TABLE_NAME: &'static str;
/// The partition key used in that table
type P: PartitionKey + Clone + PartialEq + Serialize + for<'de> Deserialize<'de> + Send + Sync;
type P: PartitionKey;
/// The sort key used int that table
type S: SortKey + Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync;
type S: SortKey;
/// They type for an entry in that table
type E: Entry<Self::P, Self::S>;
/// The type for a filter that can be applied to select entries
/// (e.g. filter out deleted entries)
type Filter: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync;
// Action to take if not able to decode current version:
// try loading from an older version
/// Try migrating an entry from an older version
fn try_migrate(_bytes: &[u8]) -> Option<Self::E> {
None
}
type Filter: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static;
/// Actions triggered by data changing in a table. If such actions
/// include updates to the local database that should be applied

View file

@ -2,6 +2,7 @@ use std::collections::VecDeque;
use std::sync::Arc;
use std::time::{Duration, Instant};
use arc_swap::ArcSwapOption;
use async_trait::async_trait;
use futures_util::stream::*;
use opentelemetry::KeyValue;
@ -13,7 +14,8 @@ use tokio::sync::{mpsc, watch};
use garage_util::background::*;
use garage_util::data::*;
use garage_util::error::Error;
use garage_util::encode::{debug_serialize, nonversioned_encode};
use garage_util::error::{Error, OkOrMessage};
use garage_rpc::ring::*;
use garage_rpc::system::System;
@ -27,12 +29,12 @@ use crate::*;
// Do anti-entropy every 10 minutes
const ANTI_ENTROPY_INTERVAL: Duration = Duration::from_secs(10 * 60);
pub struct TableSyncer<F: TableSchema + 'static, R: TableReplication + 'static> {
pub struct TableSyncer<F: TableSchema, R: TableReplication> {
system: Arc<System>,
data: Arc<TableData<F, R>>,
merkle: Arc<MerkleUpdater<F, R>>,
add_full_sync_tx: mpsc::UnboundedSender<()>,
add_full_sync_tx: ArcSwapOption<mpsc::UnboundedSender<()>>,
endpoint: Arc<Endpoint<SyncRpc, Self>>,
}
@ -60,12 +62,8 @@ struct TodoPartition {
retain: bool,
}
impl<F, R> TableSyncer<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
pub(crate) fn launch(
impl<F: TableSchema, R: TableReplication> TableSyncer<F, R> {
pub(crate) fn new(
system: Arc<System>,
data: Arc<TableData<F, R>>,
merkle: Arc<MerkleUpdater<F, R>>,
@ -74,34 +72,40 @@ where
.netapp
.endpoint(format!("garage_table/sync.rs/Rpc:{}", F::TABLE_NAME));
let (add_full_sync_tx, add_full_sync_rx) = mpsc::unbounded_channel();
let syncer = Arc::new(Self {
system: system.clone(),
system,
data,
merkle,
add_full_sync_tx,
add_full_sync_tx: ArcSwapOption::new(None),
endpoint,
});
syncer.endpoint.set_handler(syncer.clone());
system.background.spawn_worker(SyncWorker {
syncer: syncer.clone(),
ring_recv: system.ring.clone(),
ring: system.ring.borrow().clone(),
add_full_sync_rx,
todo: vec![],
next_full_sync: Instant::now() + Duration::from_secs(20),
});
syncer
}
pub fn add_full_sync(&self) {
if self.add_full_sync_tx.send(()).is_err() {
error!("({}) Could not add full sync", F::TABLE_NAME);
}
pub(crate) fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) {
let (add_full_sync_tx, add_full_sync_rx) = mpsc::unbounded_channel();
self.add_full_sync_tx
.store(Some(Arc::new(add_full_sync_tx)));
bg.spawn_worker(SyncWorker {
syncer: self.clone(),
ring_recv: self.system.ring.clone(),
ring: self.system.ring.borrow().clone(),
add_full_sync_rx,
todo: vec![],
next_full_sync: Instant::now() + Duration::from_secs(20),
});
}
pub fn add_full_sync(&self) -> Result<(), Error> {
let tx = self.add_full_sync_tx.load();
let tx = tx
.as_ref()
.ok_or_message("table sync worker is not running")?;
tx.send(()).ok_or_message("send error")?;
Ok(())
}
// ----
@ -295,7 +299,7 @@ where
);
return Ok(());
}
let root_ck_hash = hash_of::<MerkleNode>(&root_ck)?;
let root_ck_hash = hash_of_merkle_node(&root_ck)?;
// Check if they have the same root checksum
// If so, do nothing.
@ -452,16 +456,12 @@ where
// ======= SYNCHRONIZATION PROCEDURE -- RECEIVER SIDE ======
#[async_trait]
impl<F, R> EndpointHandler<SyncRpc> for TableSyncer<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> EndpointHandler<SyncRpc> for TableSyncer<F, R> {
async fn handle(self: &Arc<Self>, message: &SyncRpc, from: NodeID) -> Result<SyncRpc, Error> {
match message {
SyncRpc::RootCkHash(range, h) => {
let (_root_ck_key, root_ck) = self.get_root_ck(*range)?;
let hash = hash_of::<MerkleNode>(&root_ck)?;
let hash = hash_of_merkle_node(&root_ck)?;
Ok(SyncRpc::RootCkDifferent(hash != *h))
}
SyncRpc::GetNode(k) => {
@ -490,7 +490,7 @@ where
// -------- Sync Worker ---------
struct SyncWorker<F: TableSchema + 'static, R: TableReplication + 'static> {
struct SyncWorker<F: TableSchema, R: TableReplication> {
syncer: Arc<TableSyncer<F, R>>,
ring_recv: watch::Receiver<Arc<Ring>>,
ring: Arc<Ring>,
@ -499,7 +499,7 @@ struct SyncWorker<F: TableSchema + 'static, R: TableReplication + 'static> {
next_full_sync: Instant,
}
impl<F: TableSchema + 'static, R: TableReplication + 'static> SyncWorker<F, R> {
impl<F: TableSchema, R: TableReplication> SyncWorker<F, R> {
fn add_full_sync(&mut self) {
let system = &self.syncer.system;
let data = &self.syncer.data;
@ -565,17 +565,15 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> SyncWorker<F, R> {
}
#[async_trait]
impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWorker<F, R> {
impl<F: TableSchema, R: TableReplication> Worker for SyncWorker<F, R> {
fn name(&self) -> String {
format!("{} sync", F::TABLE_NAME)
}
fn info(&self) -> Option<String> {
let l = self.todo.len();
if l > 0 {
Some(format!("{} partitions remaining", l))
} else {
None
fn status(&self) -> WorkerStatus {
WorkerStatus {
queue_length: Some(self.todo.len() as u64),
..Default::default()
}
}
@ -588,10 +586,7 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWor
}
}
async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState {
if *must_exit.borrow() {
return WorkerState::Done;
}
async fn wait_for_work(&mut self) -> WorkerState {
select! {
s = self.add_full_sync_rx.recv() => {
if let Some(()) = s {
@ -620,8 +615,8 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWor
// ---- UTIL ----
fn hash_of<T: Serialize>(x: &T) -> Result<Hash, Error> {
Ok(blake2sum(&rmp_to_vec_all_named(x)?[..]))
fn hash_of_merkle_node(x: &MerkleNode) -> Result<Hash, Error> {
Ok(blake2sum(&nonversioned_encode(x)?[..]))
}
fn join_ordered<'a, K: Ord + Eq, V1, V2>(

View file

@ -14,9 +14,11 @@ use opentelemetry::{
use garage_db as db;
use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_util::error::Error;
use garage_util::metrics::RecordDuration;
use garage_util::migrate::Migrate;
use garage_rpc::system::System;
use garage_rpc::*;
@ -25,16 +27,18 @@ use crate::crdt::Crdt;
use crate::data::*;
use crate::gc::*;
use crate::merkle::*;
use crate::queue::InsertQueueWorker;
use crate::replication::*;
use crate::schema::*;
use crate::sync::*;
use crate::util::*;
pub struct Table<F: TableSchema + 'static, R: TableReplication + 'static> {
pub struct Table<F: TableSchema, R: TableReplication> {
pub system: Arc<System>,
pub data: Arc<TableData<F, R>>,
pub merkle_updater: Arc<MerkleUpdater<F, R>>,
pub syncer: Arc<TableSyncer<F, R>>,
gc: Arc<TableGc<F, R>>,
endpoint: Arc<Endpoint<TableRpc<F>, Self>>,
}
@ -61,11 +65,7 @@ impl<F: TableSchema> Rpc for TableRpc<F> {
type Response = Result<TableRpc<F>, Error>;
}
impl<F, R> Table<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> Table<F, R> {
// =============== PUBLIC INTERFACE FUNCTIONS (new, insert, get, etc) ===============
pub fn new(instance: F, replication: R, system: Arc<System>, db: &db::Db) -> Arc<Self> {
@ -75,15 +75,16 @@ where
let data = TableData::new(system.clone(), instance, replication, db);
let merkle_updater = MerkleUpdater::launch(&system.background, data.clone());
let merkle_updater = MerkleUpdater::new(data.clone());
let syncer = TableSyncer::launch(system.clone(), data.clone(), merkle_updater.clone());
TableGc::launch(system.clone(), data.clone());
let syncer = TableSyncer::new(system.clone(), data.clone(), merkle_updater.clone());
let gc = TableGc::new(system.clone(), data.clone());
let table = Arc::new(Self {
system,
data,
merkle_updater,
gc,
syncer,
endpoint,
});
@ -93,6 +94,13 @@ where
table
}
pub fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) {
self.merkle_updater.spawn_workers(bg);
self.syncer.spawn_workers(bg);
self.gc.spawn_workers(bg);
bg.spawn_worker(InsertQueueWorker(self.clone()));
}
pub async fn insert(&self, e: &F::E) -> Result<(), Error> {
let tracer = opentelemetry::global::tracer("garage_table");
let span = tracer.start(format!("{} insert", F::TABLE_NAME));
@ -111,7 +119,7 @@ where
let hash = e.partition_key().hash();
let who = self.data.replication.write_nodes(&hash);
let e_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(e)?));
let e_enc = Arc::new(ByteBuf::from(e.encode()?));
let rpc = TableRpc::<F>::Update(vec![e_enc]);
self.system
@ -128,6 +136,11 @@ where
Ok(())
}
/// Insert item locally
pub fn queue_insert(&self, tx: &mut db::Transaction, e: &F::E) -> db::TxResult<(), Error> {
self.data.queue_insert(tx, e)
}
pub async fn insert_many<I, IE>(&self, entries: I) -> Result<(), Error>
where
I: IntoIterator<Item = IE> + Send + Sync,
@ -157,7 +170,7 @@ where
let entry = entry.borrow();
let hash = entry.partition_key().hash();
let who = self.data.replication.write_nodes(&hash);
let e_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(entry)?));
let e_enc = Arc::new(ByteBuf::from(entry.encode()?));
for node in who {
call_list.entry(node).or_default().push(e_enc.clone());
}
@ -259,9 +272,11 @@ where
if not_all_same {
let self2 = self.clone();
let ent2 = ret_entry.clone();
self.system
.background
.spawn_cancellable(async move { self2.repair_on_read(&who[..], ent2).await });
tokio::spawn(async move {
if let Err(e) = self2.repair_on_read(&who[..], ent2).await {
warn!("Error doing repair on read: {}", e);
}
});
}
}
@ -358,11 +373,12 @@ where
.into_iter()
.map(|k| ret.get(&k).unwrap().clone())
.collect::<Vec<_>>();
self.system.background.spawn_cancellable(async move {
tokio::spawn(async move {
for v in to_repair {
self2.repair_on_read(&who[..], v).await?;
if let Err(e) = self2.repair_on_read(&who[..], v).await {
warn!("Error doing repair on read: {}", e);
}
}
Ok(())
});
}
@ -393,7 +409,7 @@ where
// =============== UTILITY FUNCTION FOR CLIENT OPERATIONS ===============
async fn repair_on_read(&self, who: &[Uuid], what: F::E) -> Result<(), Error> {
let what_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(&what)?));
let what_enc = Arc::new(ByteBuf::from(what.encode()?));
self.system
.rpc
.try_call_many(
@ -408,11 +424,7 @@ where
}
#[async_trait]
impl<F, R> EndpointHandler<TableRpc<F>> for Table<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
impl<F: TableSchema, R: TableReplication> EndpointHandler<TableRpc<F>> for Table<F, R> {
async fn handle(
self: &Arc<Self>,
msg: &TableRpc<F>,

View file

@ -49,3 +49,9 @@ impl EnumerationOrder {
}
}
}
impl Default for EnumerationOrder {
fn default() -> Self {
EnumerationOrder::Forward
}
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_util"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,7 +14,7 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_db = { version = "0.8.0", path = "../db" }
garage_db = { version = "0.8.1", path = "../db" }
arc-swap = "1.0"
async-trait = "0.1"
@ -23,6 +23,7 @@ bytes = "1.0"
digest = "0.10"
err-derive = "0.3"
git-version = "0.3.4"
hexdump = "0.1"
xxhash-rust = { version = "0.8", default-features = false, features = ["xxh3"] }
hex = "0.4"
lazy_static = "1.4"
@ -46,6 +47,8 @@ hyper = "0.14"
opentelemetry = { version = "0.17", features = [ "rt-tokio", "metrics", "trace" ] }
[dev-dependencies]
mktemp = "0.4"
[features]
k2v = []

View file

@ -1,48 +0,0 @@
//! Job worker: a generic worker that just processes incoming
//! jobs one by one
use std::sync::Arc;
use async_trait::async_trait;
use tokio::sync::{mpsc, Mutex};
use crate::background::worker::*;
use crate::background::*;
pub(crate) struct JobWorker {
pub(crate) index: usize,
pub(crate) job_chan: Arc<Mutex<mpsc::UnboundedReceiver<(Job, bool)>>>,
pub(crate) next_job: Option<Job>,
}
#[async_trait]
impl Worker for JobWorker {
fn name(&self) -> String {
format!("Job worker #{}", self.index)
}
async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
match self.next_job.take() {
None => return Ok(WorkerState::Idle),
Some(job) => {
job.await?;
Ok(WorkerState::Busy)
}
}
}
async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState {
loop {
match self.job_chan.lock().await.recv().await {
Some((job, cancellable)) => {
if cancellable && *must_exit.borrow() {
continue;
}
self.next_job = Some(job);
return WorkerState::Busy;
}
None => return WorkerState::Done,
}
}
}
}

View file

@ -1,27 +1,19 @@
//! Job runner for futures and async functions
pub mod job_worker;
pub mod vars;
pub mod worker;
use core::future::Future;
use std::collections::HashMap;
use std::pin::Pin;
use std::sync::Arc;
use serde::{Deserialize, Serialize};
use tokio::sync::{mpsc, watch, Mutex};
use tokio::sync::{mpsc, watch};
use crate::error::Error;
use worker::WorkerProcessor;
pub use worker::{Worker, WorkerState};
pub(crate) type JobOutput = Result<(), Error>;
pub(crate) type Job = Pin<Box<dyn Future<Output = JobOutput> + Send>>;
/// Job runner for futures and async functions
pub struct BackgroundRunner {
send_job: mpsc::UnboundedSender<(Job, bool)>,
send_worker: mpsc::UnboundedSender<Box<dyn Worker>>,
worker_info: Arc<std::sync::Mutex<HashMap<usize, WorkerInfo>>>,
}
@ -29,19 +21,27 @@ pub struct BackgroundRunner {
#[derive(Clone, Serialize, Deserialize, Debug)]
pub struct WorkerInfo {
pub name: String,
pub info: Option<String>,
pub status: WorkerStatus,
pub state: WorkerState,
pub errors: usize,
pub consecutive_errors: usize,
pub last_error: Option<(String, u64)>,
}
/// WorkerStatus is a struct returned by the worker with a bunch of canonical
/// fields to indicate their status to CLI users. All fields are optional.
#[derive(Clone, Serialize, Deserialize, Debug, Default)]
pub struct WorkerStatus {
pub tranquility: Option<u32>,
pub progress: Option<String>,
pub queue_length: Option<u64>,
pub persistent_errors: Option<u64>,
pub freeform: Vec<String>,
}
impl BackgroundRunner {
/// Create a new BackgroundRunner
pub fn new(
n_runners: usize,
stop_signal: watch::Receiver<bool>,
) -> (Arc<Self>, tokio::task::JoinHandle<()>) {
pub fn new(stop_signal: watch::Receiver<bool>) -> (Arc<Self>, tokio::task::JoinHandle<()>) {
let (send_worker, worker_out) = mpsc::unbounded_channel::<Box<dyn Worker>>();
let worker_info = Arc::new(std::sync::Mutex::new(HashMap::new()));
@ -52,24 +52,7 @@ impl BackgroundRunner {
worker_processor.run().await;
});
let (send_job, queue_out) = mpsc::unbounded_channel();
let queue_out = Arc::new(Mutex::new(queue_out));
for i in 0..n_runners {
let queue_out = queue_out.clone();
send_worker
.send(Box::new(job_worker::JobWorker {
index: i,
job_chan: queue_out.clone(),
next_job: None,
}))
.ok()
.unwrap();
}
let bgrunner = Arc::new(Self {
send_job,
send_worker,
worker_info,
});
@ -80,31 +63,6 @@ impl BackgroundRunner {
self.worker_info.lock().unwrap().clone()
}
/// Spawn a task to be run in background
pub fn spawn<T>(&self, job: T)
where
T: Future<Output = JobOutput> + Send + 'static,
{
let boxed: Job = Box::pin(job);
self.send_job
.send((boxed, false))
.ok()
.expect("Could not put job in queue");
}
/// Spawn a task to be run in background. It may get discarded before running if spawned while
/// the runner is stopping
pub fn spawn_cancellable<T>(&self, job: T)
where
T: Future<Output = JobOutput> + Send + 'static,
{
let boxed: Job = Box::pin(job);
self.send_job
.send((boxed, true))
.ok()
.expect("Could not put job in queue");
}
pub fn spawn_worker<W>(&self, worker: W)
where
W: Worker + 'static,

113
src/util/background/vars.rs Normal file
View file

@ -0,0 +1,113 @@
use std::collections::HashMap;
use std::str::FromStr;
use crate::error::{Error, OkOrMessage};
use crate::migrate::Migrate;
use crate::persister::PersisterShared;
pub struct BgVars {
vars: HashMap<&'static str, Box<dyn BgVarTrait>>,
}
impl BgVars {
pub fn new() -> Self {
Self {
vars: HashMap::new(),
}
}
pub fn register_rw<V, T, GF, SF>(
&mut self,
p: &PersisterShared<V>,
name: &'static str,
get_fn: GF,
set_fn: SF,
) where
V: Migrate + Default + Send + Sync,
T: FromStr + ToString + Send + Sync + 'static,
GF: Fn(&PersisterShared<V>) -> T + Send + Sync + 'static,
SF: Fn(&PersisterShared<V>, T) -> Result<(), Error> + Send + Sync + 'static,
{
let p1 = p.clone();
let get_fn = move || get_fn(&p1);
let p2 = p.clone();
let set_fn = move |v| set_fn(&p2, v);
self.vars.insert(name, Box::new(BgVar { get_fn, set_fn }));
}
pub fn register_ro<V, T, GF>(&mut self, p: &PersisterShared<V>, name: &'static str, get_fn: GF)
where
V: Migrate + Default + Send + Sync,
T: FromStr + ToString + Send + Sync + 'static,
GF: Fn(&PersisterShared<V>) -> T + Send + Sync + 'static,
{
let p1 = p.clone();
let get_fn = move || get_fn(&p1);
let set_fn = move |_| Err(Error::Message(format!("Cannot set value of {}", name)));
self.vars.insert(name, Box::new(BgVar { get_fn, set_fn }));
}
pub fn get(&self, var: &str) -> Result<String, Error> {
Ok(self
.vars
.get(var)
.ok_or_message("variable does not exist")?
.get())
}
pub fn get_all(&self) -> Vec<(&'static str, String)> {
self.vars.iter().map(|(k, v)| (*k, v.get())).collect()
}
pub fn set(&self, var: &str, val: &str) -> Result<(), Error> {
self.vars
.get(var)
.ok_or_message("variable does not exist")?
.set(val)
}
}
impl Default for BgVars {
fn default() -> Self {
Self::new()
}
}
// ----
trait BgVarTrait: Send + Sync + 'static {
fn get(&self) -> String;
fn set(&self, v: &str) -> Result<(), Error>;
}
struct BgVar<T, GF, SF>
where
T: FromStr + ToString + Send + Sync + 'static,
GF: Fn() -> T + Send + Sync + 'static,
SF: Fn(T) -> Result<(), Error> + Sync + Send + 'static,
{
get_fn: GF,
set_fn: SF,
}
impl<T, GF, SF> BgVarTrait for BgVar<T, GF, SF>
where
T: FromStr + ToString + Sync + Send + 'static,
GF: Fn() -> T + Sync + Send + 'static,
SF: Fn(T) -> Result<(), Error> + Sync + Send + 'static,
{
fn get(&self) -> String {
(self.get_fn)().to_string()
}
fn set(&self, vstr: &str) -> Result<(), Error> {
let value = vstr
.parse()
.map_err(|_| Error::Message(format!("invalid value: {}", vstr)))?;
(self.set_fn)(value)
}
}

View file

@ -1,6 +1,6 @@
use std::collections::HashMap;
use std::sync::Arc;
use std::time::{Duration, Instant};
use std::time::Duration;
use async_trait::async_trait;
use futures::future::*;
@ -10,10 +10,14 @@ use serde::{Deserialize, Serialize};
use tokio::select;
use tokio::sync::{mpsc, watch};
use crate::background::WorkerInfo;
use crate::background::{WorkerInfo, WorkerStatus};
use crate::error::Error;
use crate::time::now_msec;
// All workers that haven't exited for this time after an exit signal was recieved
// will be interrupted in the middle of whatever they are doing.
const EXIT_DEADLINE: Duration = Duration::from_secs(8);
#[derive(PartialEq, Copy, Clone, Serialize, Deserialize, Debug)]
pub enum WorkerState {
Busy,
@ -26,7 +30,7 @@ impl std::fmt::Display for WorkerState {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
WorkerState::Busy => write!(f, "Busy"),
WorkerState::Throttled(t) => write!(f, "Thr:{:.3}", t),
WorkerState::Throttled(_) => write!(f, "Busy*"),
WorkerState::Idle => write!(f, "Idle"),
WorkerState::Done => write!(f, "Done"),
}
@ -37,8 +41,8 @@ impl std::fmt::Display for WorkerState {
pub trait Worker: Send {
fn name(&self) -> String;
fn info(&self) -> Option<String> {
None
fn status(&self) -> WorkerStatus {
Default::default()
}
/// Work: do a basic unit of work, if one is available (otherwise, should return
@ -50,10 +54,8 @@ pub trait Worker: Send {
async fn work(&mut self, must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error>;
/// Wait for work: await for some task to become available. This future can be interrupted in
/// the middle for any reason. This future doesn't have to await on must_exit.changed(), we
/// are doing it for you. Therefore it only receives a read refernce to must_exit which allows
/// it to check if we are exiting.
async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState;
/// the middle for any reason, for example if an interrupt signal was recieved.
async fn wait_for_work(&mut self) -> WorkerState;
}
pub(crate) struct WorkerProcessor {
@ -93,11 +95,9 @@ impl WorkerProcessor {
let task_id = next_task_id;
next_task_id += 1;
let stop_signal = self.stop_signal.clone();
let stop_signal_worker = self.stop_signal.clone();
let mut worker = WorkerHandler {
task_id,
stop_signal,
stop_signal_worker,
worker: new_worker,
state: WorkerState::Busy,
errors: 0,
@ -119,7 +119,7 @@ impl WorkerProcessor {
match wi.get_mut(&worker.task_id) {
Some(i) => {
i.state = worker.state;
i.info = worker.worker.info();
i.status = worker.worker.status();
i.errors = worker.errors;
i.consecutive_errors = worker.consecutive_errors;
if worker.last_error.is_some() {
@ -130,7 +130,7 @@ impl WorkerProcessor {
wi.insert(worker.task_id, WorkerInfo {
name: worker.worker.name(),
state: worker.state,
info: worker.worker.info(),
status: worker.worker.status(),
errors: worker.errors,
consecutive_errors: worker.consecutive_errors,
last_error: worker.last_error.take(),
@ -153,26 +153,14 @@ impl WorkerProcessor {
}
// We are exiting, drain everything
let drain_half_time = Instant::now() + Duration::from_secs(5);
let drain_everything = async move {
while let Some(mut worker) = workers.next().await {
if worker.state == WorkerState::Done {
info!(
"Worker {} (TID {}) exited",
worker.worker.name(),
worker.task_id
);
} else if Instant::now() > drain_half_time {
warn!("Worker {} (TID {}) interrupted between two iterations in state {:?} (this should be fine)", worker.worker.name(), worker.task_id, worker.state);
} else {
workers.push(
async move {
worker.step().await;
worker
}
.boxed(),
);
}
while let Some(worker) = workers.next().await {
info!(
"Worker {} (TID {}) exited (last state: {:?})",
worker.worker.name(),
worker.task_id,
worker.state
);
}
};
@ -180,7 +168,7 @@ impl WorkerProcessor {
_ = drain_everything => {
info!("All workers exited peacefully \\o/");
}
_ = tokio::time::sleep(Duration::from_secs(9)) => {
_ = tokio::time::sleep(EXIT_DEADLINE) => {
error!("Some workers could not exit in time, we are cancelling some things in the middle");
}
}
@ -190,7 +178,6 @@ impl WorkerProcessor {
struct WorkerHandler {
task_id: usize,
stop_signal: watch::Receiver<bool>,
stop_signal_worker: watch::Receiver<bool>,
worker: Box<dyn Worker>,
state: WorkerState,
errors: usize,
@ -225,33 +212,19 @@ impl WorkerHandler {
},
WorkerState::Throttled(delay) => {
// Sleep for given delay and go back to busy state
if !*self.stop_signal.borrow() {
select! {
_ = tokio::time::sleep(Duration::from_secs_f32(delay)) => (),
_ = self.stop_signal.changed() => (),
select! {
_ = tokio::time::sleep(Duration::from_secs_f32(delay)) => {
self.state = WorkerState::Busy;
}
_ = self.stop_signal.changed() => (),
}
self.state = WorkerState::Busy;
}
WorkerState::Idle => {
if *self.stop_signal.borrow() {
select! {
new_st = self.worker.wait_for_work(&self.stop_signal_worker) => {
self.state = new_st;
}
_ = tokio::time::sleep(Duration::from_secs(1)) => {
// stay in Idle state
}
}
} else {
select! {
new_st = self.worker.wait_for_work(&self.stop_signal_worker) => {
self.state = new_st;
}
_ = self.stop_signal.changed() => {
// stay in Idle state
}
select! {
new_st = self.worker.wait_for_work() => {
self.state = new_st;
}
_ = self.stop_signal.changed() => (),
}
}
WorkerState::Done => unreachable!(),

View file

@ -34,7 +34,11 @@ pub struct Config {
pub compression_level: Option<i32>,
/// RPC secret key: 32 bytes hex encoded
pub rpc_secret: String,
/// Note: When using `read_config` this should never be `None`
pub rpc_secret: Option<String>,
/// Optional file where RPC secret key is read from
pub rpc_secret_file: Option<String>,
/// Address to bind for RPC
pub rpc_bind_addr: SocketAddr,
@ -177,7 +181,31 @@ pub fn read_config(config_file: PathBuf) -> Result<Config, Error> {
let mut config = String::new();
file.read_to_string(&mut config)?;
Ok(toml::from_str(&config)?)
let mut parsed_config: Config = toml::from_str(&config)?;
match (&parsed_config.rpc_secret, &parsed_config.rpc_secret_file) {
(Some(_), None) => {
// no-op
}
(Some(_), Some(_)) => {
return Err("only one of `rpc_secret` and `rpc_secret_file` can be set".into())
}
(None, Some(rpc_secret_file_path_string)) => {
let mut rpc_secret_file = std::fs::OpenOptions::new()
.read(true)
.open(rpc_secret_file_path_string)?;
let mut rpc_secret_from_file = String::new();
rpc_secret_file.read_to_string(&mut rpc_secret_from_file)?;
// trim_end: allows for use case such as `echo "$(openssl rand -hex 32)" > somefile`.
// also editors sometimes add a trailing newline
parsed_config.rpc_secret = Some(String::from(rpc_secret_from_file.trim_end()));
}
(None, None) => {
return Err("either `rpc_secret` or `rpc_secret_file` needs to be set".into())
}
};
Ok(parsed_config)
}
fn default_compression() -> Option<i32> {
@ -233,3 +261,123 @@ where
deserializer.deserialize_any(OptionVisitor)
}
#[cfg(test)]
mod tests {
use crate::error::Error;
use std::fs::File;
use std::io::Write;
#[test]
fn test_rpc_secret_is_required() -> Result<(), Error> {
let path1 = mktemp::Temp::new_file()?;
let mut file1 = File::create(path1.as_path())?;
writeln!(
file1,
r#"
metadata_dir = "/tmp/garage/meta"
data_dir = "/tmp/garage/data"
replication_mode = "3"
rpc_bind_addr = "[::]:3901"
[s3_api]
s3_region = "garage"
api_bind_addr = "[::]:3900"
"#
)?;
assert_eq!(
"either `rpc_secret` or `rpc_secret_file` needs to be set",
super::read_config(path1.to_path_buf())
.unwrap_err()
.to_string()
);
drop(path1);
drop(file1);
let path2 = mktemp::Temp::new_file()?;
let mut file2 = File::create(path2.as_path())?;
writeln!(
file2,
r#"
metadata_dir = "/tmp/garage/meta"
data_dir = "/tmp/garage/data"
replication_mode = "3"
rpc_bind_addr = "[::]:3901"
rpc_secret = "foo"
[s3_api]
s3_region = "garage"
api_bind_addr = "[::]:3900"
"#
)?;
let config = super::read_config(path2.to_path_buf())?;
assert_eq!("foo", config.rpc_secret.unwrap());
drop(path2);
drop(file2);
Ok(())
}
#[test]
fn test_rpc_secret_file_works() -> Result<(), Error> {
let path_secret = mktemp::Temp::new_file()?;
let mut file_secret = File::create(path_secret.as_path())?;
writeln!(file_secret, "foo")?;
drop(file_secret);
let path_config = mktemp::Temp::new_file()?;
let mut file_config = File::create(path_config.as_path())?;
let path_secret_path = path_secret.as_path().display();
writeln!(
file_config,
r#"
metadata_dir = "/tmp/garage/meta"
data_dir = "/tmp/garage/data"
replication_mode = "3"
rpc_bind_addr = "[::]:3901"
rpc_secret_file = "{path_secret_path}"
[s3_api]
s3_region = "garage"
api_bind_addr = "[::]:3900"
"#
)?;
let config = super::read_config(path_config.to_path_buf())?;
assert_eq!("foo", config.rpc_secret.unwrap());
drop(path_config);
drop(path_secret);
drop(file_config);
Ok(())
}
#[test]
fn test_rcp_secret_and_rpc_secret_file_cannot_be_set_both() -> Result<(), Error> {
let path_config = mktemp::Temp::new_file()?;
let mut file_config = File::create(path_config.as_path())?;
writeln!(
file_config,
r#"
metadata_dir = "/tmp/garage/meta"
data_dir = "/tmp/garage/data"
replication_mode = "3"
rpc_bind_addr = "[::]:3901"
rpc_secret= "dummy"
rpc_secret_file = "dummy"
[s3_api]
s3_region = "garage"
api_bind_addr = "[::]:3900"
"#
)?;
assert_eq!(
"only one of `rpc_secret` and `rpc_secret_file` can be set",
super::read_config(path_config.to_path_buf())
.unwrap_err()
.to_string()
);
drop(path_config);
drop(file_config);
Ok(())
}
}

View file

@ -140,34 +140,3 @@ pub fn fasthash(data: &[u8]) -> FastHash {
pub fn gen_uuid() -> Uuid {
rand::thread_rng().gen::<[u8; 32]>().into()
}
// RMP serialization with names of fields and variants
/// Serialize to MessagePack
pub fn rmp_to_vec_all_named<T>(val: &T) -> Result<Vec<u8>, rmp_serde::encode::Error>
where
T: Serialize + ?Sized,
{
let mut wr = Vec::with_capacity(128);
let mut se = rmp_serde::Serializer::new(&mut wr)
.with_struct_map()
.with_string_variants();
val.serialize(&mut se)?;
Ok(wr)
}
/// Serialize to JSON, truncating long result
pub fn debug_serialize<T: Serialize>(x: T) -> String {
match serde_json::to_string(&x) {
Ok(ss) => {
if ss.len() > 100 {
// TODO this can panic if 100 is not a codepoint boundary, but inside a 2 Bytes
// (or more) codepoint
ss[..100].to_string()
} else {
ss
}
}
Err(e) => format!("<JSON serialization error: {}>", e),
}
}

42
src/util/encode.rs Normal file
View file

@ -0,0 +1,42 @@
use serde::{Deserialize, Serialize};
/// Serialize to MessagePacki, without versionning
/// (see garage_util::migrate for functions that manage versionned
/// data formats)
pub fn nonversioned_encode<T>(val: &T) -> Result<Vec<u8>, rmp_serde::encode::Error>
where
T: Serialize + ?Sized,
{
let mut wr = Vec::with_capacity(128);
let mut se = rmp_serde::Serializer::new(&mut wr)
.with_struct_map()
.with_string_variants();
val.serialize(&mut se)?;
Ok(wr)
}
/// Deserialize from MessagePacki, without versionning
/// (see garage_util::migrate for functions that manage versionned
/// data formats)
pub fn nonversioned_decode<T>(bytes: &[u8]) -> Result<T, rmp_serde::decode::Error>
where
T: for<'de> Deserialize<'de> + ?Sized,
{
rmp_serde::decode::from_read_ref::<_, T>(bytes)
}
/// Serialize to JSON, truncating long result
pub fn debug_serialize<T: Serialize>(x: T) -> String {
match serde_json::to_string(&x) {
Ok(ss) => {
if ss.len() > 100 {
// TODO this can panic if 100 is not a codepoint boundary, but inside a 2 Bytes
// (or more) codepoint
ss[..100].to_string()
} else {
ss
}
}
Err(e) => format!("<JSON serialization error: {}>", e),
}
}

View file

@ -7,6 +7,7 @@ use err_derive::Error;
use serde::{de::Visitor, Deserialize, Deserializer, Serialize, Serializer};
use crate::data::*;
use crate::encode::debug_serialize;
/// Regroup all Garage errors
#[derive(Debug, Error)]

View file

@ -1,4 +1,4 @@
pub fn format_table(data: Vec<String>) {
pub fn format_table_to_string(data: Vec<String>) -> String {
let data = data
.iter()
.map(|s| s.split('\t').collect::<Vec<_>>())
@ -24,5 +24,9 @@ pub fn format_table(data: Vec<String>) {
out.push('\n');
}
print!("{}", out);
out
}
pub fn format_table(data: Vec<String>) {
print!("{}", format_table_to_string(data));
}

View file

@ -8,11 +8,12 @@ pub mod background;
pub mod config;
pub mod crdt;
pub mod data;
pub mod encode;
pub mod error;
pub mod formater;
pub mod metrics;
pub mod migrate;
pub mod persister;
pub mod time;
pub mod token_bucket;
pub mod tranquilizer;
pub mod version;

159
src/util/migrate.rs Normal file
View file

@ -0,0 +1,159 @@
use serde::{Deserialize, Serialize};
/// Indicates that this type has an encoding that can be migrated from
/// a previous version upon upgrades of Garage.
pub trait Migrate: Serialize + for<'de> Deserialize<'de> + 'static {
/// A sequence of bytes to add at the beginning of the serialized
/// string, to identify that the data is of this version.
const VERSION_MARKER: &'static [u8] = b"";
/// The previous version of this data type, from which items of this version
/// can be migrated.
type Previous: Migrate;
/// The migration function that transforms a value decoded in the old format
/// to an up-to-date value.
fn migrate(previous: Self::Previous) -> Self;
/// Decode an encoded version of this type, going through a migration if necessary.
fn decode(bytes: &[u8]) -> Option<Self> {
let marker_len = Self::VERSION_MARKER.len();
if bytes.get(..marker_len) == Some(Self::VERSION_MARKER) {
if let Ok(value) = rmp_serde::decode::from_read_ref::<_, Self>(&bytes[marker_len..]) {
return Some(value);
}
}
Self::Previous::decode(bytes).map(Self::migrate)
}
/// Encode this type with optionnal version marker
fn encode(&self) -> Result<Vec<u8>, rmp_serde::encode::Error> {
let mut wr = Vec::with_capacity(128);
wr.extend_from_slice(Self::VERSION_MARKER);
let mut se = rmp_serde::Serializer::new(&mut wr)
.with_struct_map()
.with_string_variants();
self.serialize(&mut se)?;
Ok(wr)
}
}
/// Indicates that this type has no previous encoding version to be migrated from.
pub trait InitialFormat: Serialize + for<'de> Deserialize<'de> + 'static {
/// A sequence of bytes to add at the beginning of the serialized
/// string, to identify that the data is of this version.
const VERSION_MARKER: &'static [u8] = b"";
}
impl<T: InitialFormat> Migrate for T {
const VERSION_MARKER: &'static [u8] = <T as InitialFormat>::VERSION_MARKER;
type Previous = NoPrevious;
fn migrate(_previous: Self::Previous) -> Self {
unreachable!();
}
}
/// Internal type used by InitialFormat, not meant for general use.
#[derive(Serialize, Deserialize)]
pub enum NoPrevious {}
impl Migrate for NoPrevious {
type Previous = NoPrevious;
fn migrate(_previous: Self::Previous) -> Self {
unreachable!();
}
fn decode(_bytes: &[u8]) -> Option<Self> {
None
}
fn encode(&self) -> Result<Vec<u8>, rmp_serde::encode::Error> {
unreachable!()
}
}
#[cfg(test)]
mod test {
use super::*;
#[derive(Serialize, Deserialize, PartialEq, Eq, Debug)]
struct V1 {
a: usize,
b: String,
}
impl InitialFormat for V1 {}
#[derive(Serialize, Deserialize, PartialEq, Eq, Debug)]
struct V2 {
a: usize,
b: Vec<String>,
c: String,
}
impl Migrate for V2 {
const VERSION_MARKER: &'static [u8] = b"GtestV2";
type Previous = V1;
fn migrate(prev: V1) -> V2 {
V2 {
a: prev.a,
b: vec![prev.b],
c: String::new(),
}
}
}
#[test]
fn test_v1() {
let x = V1 {
a: 12,
b: "hello".into(),
};
let x_enc = x.encode().unwrap();
let y = V1::decode(&x_enc).unwrap();
assert_eq!(x, y);
}
#[test]
fn test_v2() {
let x = V2 {
a: 12,
b: vec!["hello".into(), "world".into()],
c: "plop".into(),
};
let x_enc = x.encode().unwrap();
assert_eq!(&x_enc[..V2::VERSION_MARKER.len()], V2::VERSION_MARKER);
let y = V2::decode(&x_enc).unwrap();
assert_eq!(x, y);
}
#[test]
fn test_migrate() {
let x = V1 {
a: 12,
b: "hello".into(),
};
let x_enc = x.encode().unwrap();
let xx = V1::decode(&x_enc).unwrap();
assert_eq!(x, xx);
let y = V2::decode(&x_enc).unwrap();
assert_eq!(
y,
V2 {
a: 12,
b: vec!["hello".into()],
c: "".into(),
}
);
let y_enc = y.encode().unwrap();
assert_eq!(&y_enc[..V2::VERSION_MARKER.len()], V2::VERSION_MARKER);
let z = V2::decode(&y_enc).unwrap();
assert_eq!(y, z);
}
}

View file

@ -1,23 +1,19 @@
use std::io::{Read, Write};
use std::path::{Path, PathBuf};
use std::sync::{Arc, RwLock};
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use serde::{Deserialize, Serialize};
use crate::data::*;
use crate::error::Error;
use crate::migrate::Migrate;
pub struct Persister<T: Serialize + for<'de> Deserialize<'de>> {
pub struct Persister<T: Migrate> {
path: PathBuf,
_marker: std::marker::PhantomData<T>,
}
impl<T> Persister<T>
where
T: Serialize + for<'de> Deserialize<'de>,
{
impl<T: Migrate> Persister<T> {
pub fn new(base_dir: &Path, file_name: &str) -> Self {
let mut path = base_dir.to_path_buf();
path.push(file_name);
@ -27,18 +23,37 @@ where
}
}
fn decode(&self, bytes: &[u8]) -> Result<T, Error> {
match T::decode(bytes) {
Some(v) => Ok(v),
None => {
error!(
"Unable to decode persisted data file {}",
self.path.display()
);
for line in hexdump::hexdump_iter(bytes) {
debug!("{}", line);
}
Err(Error::Message(format!(
"Unable to decode persisted data file {}",
self.path.display()
)))
}
}
}
pub fn load(&self) -> Result<T, Error> {
let mut file = std::fs::OpenOptions::new().read(true).open(&self.path)?;
let mut bytes = vec![];
file.read_to_end(&mut bytes)?;
let value = rmp_serde::decode::from_read_ref(&bytes[..])?;
let value = self.decode(&bytes[..])?;
Ok(value)
}
pub fn save(&self, t: &T) -> Result<(), Error> {
let bytes = rmp_to_vec_all_named(t)?;
let bytes = t.encode()?;
let mut file = std::fs::OpenOptions::new()
.write(true)
@ -57,12 +72,12 @@ where
let mut bytes = vec![];
file.read_to_end(&mut bytes).await?;
let value = rmp_serde::decode::from_read_ref(&bytes[..])?;
let value = self.decode(&bytes[..])?;
Ok(value)
}
pub async fn save_async(&self, t: &T) -> Result<(), Error> {
let bytes = rmp_to_vec_all_named(t)?;
let bytes = t.encode()?;
let mut file = tokio::fs::File::create(&self.path).await?;
file.write_all(&bytes[..]).await?;
@ -70,3 +85,36 @@ where
Ok(())
}
}
pub struct PersisterShared<V: Migrate + Default>(Arc<(Persister<V>, RwLock<V>)>);
impl<V: Migrate + Default> Clone for PersisterShared<V> {
fn clone(&self) -> PersisterShared<V> {
PersisterShared(self.0.clone())
}
}
impl<V: Migrate + Default> PersisterShared<V> {
pub fn new(base_dir: &Path, file_name: &str) -> Self {
let persister = Persister::new(base_dir, file_name);
let value = persister.load().unwrap_or_default();
Self(Arc::new((persister, RwLock::new(value))))
}
pub fn get_with<F, R>(&self, f: F) -> R
where
F: FnOnce(&V) -> R,
{
let value = self.0 .1.read().unwrap();
f(&value)
}
pub fn set_with<F>(&self, f: F) -> Result<(), Error>
where
F: FnOnce(&mut V),
{
let mut value = self.0 .1.write().unwrap();
f(&mut value);
self.0 .0.save(&value)
}
}

View file

@ -1,40 +0,0 @@
use std::time::{Duration, Instant};
use tokio::time::sleep;
pub struct TokenBucket {
// Replenish rate: number of tokens per second
replenish_rate: u64,
// Current number of tokens
tokens: u64,
// Last replenish time
last_replenish: Instant,
}
impl TokenBucket {
pub fn new(replenish_rate: u64) -> Self {
Self {
replenish_rate,
tokens: 0,
last_replenish: Instant::now(),
}
}
pub async fn take(&mut self, tokens: u64) {
while self.tokens < tokens {
let needed = tokens - self.tokens;
let delay = (needed as f64) / (self.replenish_rate as f64);
sleep(Duration::from_secs_f64(delay)).await;
self.replenish();
}
self.tokens -= tokens;
}
pub fn replenish(&mut self) {
let now = Instant::now();
let new_tokens =
((now - self.last_replenish).as_secs_f64() * (self.replenish_rate as f64)) as u64;
self.tokens += new_tokens;
self.last_replenish = now;
}
}

View file

@ -1,6 +1,6 @@
[package]
name = "garage_web"
version = "0.8.0"
version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>", "Quentin Dufour <quentin@dufour.io>"]
edition = "2018"
license = "AGPL-3.0"
@ -14,10 +14,10 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
garage_api = { version = "0.8.0", path = "../api" }
garage_model = { version = "0.8.0", path = "../model" }
garage_util = { version = "0.8.0", path = "../util" }
garage_table = { version = "0.8.0", path = "../table" }
garage_api = { version = "0.8.1", path = "../api" }
garage_model = { version = "0.8.1", path = "../model" }
garage_util = { version = "0.8.1", path = "../util" }
garage_table = { version = "0.8.1", path = "../table" }
err-derive = "0.3"
tracing = "0.1.30"