It takes as paramters the replication factor and the zone redundancy, computes the
largest partition size reachable with these constraints, and among the possible
assignation with this partition size, it computes the one that moves the least number
of partitions compared to the previous assignation.
This computation uses graph algorithms defined in graph_algo.rs
Unfortunately, rusqlite uses the opposite logic for enabling/disabling
bundled libraries to others (libsodium-sys, zstd-sys). Cargo features
are very limited and doesn't allow to enable feature A in a dependency
iff feature B is disabled.
Note, lmdb-rkv-sys doesn't need any special treatment because it
automatically links against system liblmdb if found via pkgconf.
Linux distros should build garage with
`--no-default-features --features system-libs` to disable bundled-libs
and enable system-libs.
If this feature is enabled, libsodium-sys and zstd-sys will link
dynamically against system-provided libraries instead of building
and linking statically the bundled (possibly outdated and vulnerable)
copies of them. This feature is intended mainly for linux package
maintainers.
By default, structopt reports the value provided by
the env var CARGO_PKG_VERSION, feeded by Cargo when reading
Cargo.toml. However for Garage we use a versioning based on git,
so we often report a version that is behind the real version.
In this commit, we create garage_util::version::garage() that
reports the right version and configure all structopt subcommands
to call this function instead of using the env var.
- [x] New background worker trait
- [x] Adapt all current workers to use new API
- [x] Command to list currently running workers, and whether they are active, idle, or dead
- [x] Error reporting
- Optimizations
- [x] Merkle updater: several items per iteration
- [ ] Use `tokio::task::spawn_blocking` where appropriate so that CPU-intensive tasks don't block other things going on
- scrub:
- [x] have only one worker with a channel to start/pause/cancel
- [x] automatic scrub
- [x] ability to view and change tranquility from CLI
- [x] persistence of a few info
- [ ] Testing
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#332
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#337
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#335
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
- [x] Refactoring of internal counting API
- [x] Repair procedure for counters (it's an offline procedure!!!)
- [x] New counter for objects in buckets
- [x] Add quotas to buckets struct
- [x] Add CLI to manage bucket quotas
- [x] Add admin API to manage bucket quotas
- [x] Apply quotas by adding checks on put operations
- [x] Proof-read
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#326
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
- [x] Design interface
- [x] Implement Sled backend
- [x] Re-implement the SledCountedTree hack ~~on Sled backend~~ on all backends (i.e. over the abstraction)
- [x] Convert Garage code to use generic interface
- [x] Proof-read converted Garage code
- [ ] Test everything well
- [x] Implement sqlite backend
- [x] Implement LMDB backend
- [ ] (Implement Persy backend?)
- [ ] (Implement other backends? (like RocksDB, ...))
- [x] Implement backend choice in config file and garage server module
- [x] Add CLI for converting between DB formats
- Exploit the new interface to put more things in transactions
- [x] `.updated()` trigger on Garage tables
Fix#284
**Bugs**
- [x] When exporting sqlite, trees iterate empty??
- [x] LMDB doesn't work
**Known issues for various back-ends**
- Sled:
- Eats all my RAM and also all my disk space
- `.len()` has to traverse the whole table
- Is actually quite slow on some operations
- And is actually pretty bad code...
- Sqlite:
- Requires a lock to be taken on all operations. The lock is also taken when iterating on a table with `.iter()`, and the lock isn't released until the iterator is dropped. This means that we must be VERY carefull to not do anything else inside a `.iter()` loop or else we will have a deadlock! Most such cases have been eliminated from the Garage codebase, but there might still be some that remain. If your Garage-over-Sqlite seems to hang/freeze, this is the reason.
- (adapter uses a bunch of unsafe code)
- Heed (LMDB):
- Not suited for 32-bit machines as it has to map the whole DB in memory.
- (adpater uses a tiny bit of unsafe code)
**My recommendation:** avoid 32-bit machines and use LMDB as much as possible.
**Converting databases** is actually quite easy. For example from Sled to LMDB:
```bash
cd src/db
cargo run --features cli --bin convert -- -i path/to/garage/meta/db -a sled -o path/to/garage/meta/db.lmdb -b lmdb
```
Then, just add this to your `config.toml`:
```toml
db_engine = "lmdb"
```
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#322
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
- [x] Better distinguish error types
- [x] Parse error messages received from server
- [x] Remove `src/` folder layer, we don't have that for other crates
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#307
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
**Spec:**
- [x] Start writing
- [x] Specify all layout endpoints
- [x] Specify all endpoints for operations on keys
- [x] Specify all endpoints for operations on key/bucket permissions
- [x] Specify all endpoints for operations on buckets
- [x] Specify all endpoints for operations on bucket aliases
View rendered spec at <https://git.deuxfleurs.fr/Deuxfleurs/garage/src/branch/admin-api/doc/drafts/admin-api.md>
**Code:**
- [x] Refactor code for admin api to use common api code that was created for K2V
**General endpoints:**
- [x] Metrics
- [x] GetClusterStatus
- [x] ConnectClusterNodes
- [x] GetClusterLayout
- [x] UpdateClusterLayout
- [x] ApplyClusterLayout
- [x] RevertClusterLayout
**Key-related endpoints:**
- [x] ListKeys
- [x] CreateKey
- [x] ImportKey
- [x] GetKeyInfo
- [x] UpdateKey
- [x] DeleteKey
**Bucket-related endpoints:**
- [x] ListBuckets
- [x] CreateBucket
- [x] GetBucketInfo
- [x] DeleteBucket
- [x] PutBucketWebsite
- [x] DeleteBucketWebsite
**Operations on key/bucket permissions:**
- [x] BucketAllowKey
- [x] BucketDenyKey
**Operations on bucket aliases:**
- [x] GlobalAliasBucket
- [x] GlobalUnaliasBucket
- [x] LocalAliasBucket
- [x] LocalUnaliasBucket
**And also:**
- [x] Separate error type for the admin API (this PR includes a quite big refactoring of error handling)
- [x] Add management of website access
- [ ] Check that nothing is missing wrt what can be done using the CLI
- [ ] Improve formatting of the spec
- [x] Make sure everyone is cool with the API design
Fix#231Fix#295
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#298
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
lib.rs could use getting split in modules, but I'm not sure how exactly
Co-authored-by: trinity-1686a <trinity@deuxfleurs.fr>
Reviewed-on: Deuxfleurs/garage#303
Co-authored-by: trinity-1686a <trinity.pointard@gmail.com>
Co-committed-by: trinity-1686a <trinity.pointard@gmail.com>
**Specification:**
View spec at [this URL](https://git.deuxfleurs.fr/Deuxfleurs/garage/src/branch/k2v/doc/drafts/k2v-spec.md)
- [x] Specify the structure of K2V triples
- [x] Specify the DVVS format used for causality detection
- [x] Specify the K2V index (just a counter of number of values per partition key)
- [x] Specify single-item endpoints: ReadItem, InsertItem, DeleteItem
- [x] Specify index endpoint: ReadIndex
- [x] Specify multi-item endpoints: InsertBatch, ReadBatch, DeleteBatch
- [x] Move to JSON objects instead of tuples
- [x] Specify endpoints for polling for updates on single values (PollItem)
**Implementation:**
- [x] Table for K2V items, causal contexts
- [x] Indexing mechanism and table for K2V index
- [x] Make API handlers a bit more generic
- [x] K2V API endpoint
- [x] K2V API router
- [x] ReadItem
- [x] InsertItem
- [x] DeleteItem
- [x] PollItem
- [x] ReadIndex
- [x] InsertBatch
- [x] ReadBatch
- [x] DeleteBatch
**Testing:**
- [x] Just a simple Python script that does some requests to check visually that things are going right (does not contain parsing of results or assertions on returned values)
- [x] Actual tests:
- [x] Adapt testing framework
- [x] Simple test with InsertItem + ReadItem
- [x] Test with several Insert/Read/DeleteItem + ReadIndex
- [x] Test all combinations of return formats for ReadItem
- [x] Test with ReadBatch, InsertBatch, DeleteBatch
- [x] Test with PollItem
- [x] Test error codes
- [ ] Fix most broken stuff
- [x] test PollItem broken randomly
- [x] when invalid causality tokens are given, errors should be 4xx not 5xx
**Improvements:**
- [x] Descending range queries
- [x] Specify
- [x] Implement
- [x] Add test
- [x] Batch updates to index counter
- [x] Put K2V behind `k2v` feature flag
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#293
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
fixes#295, partially
Co-authored-by: Alex Auvolat <alex@adnab.me>
Reviewed-on: Deuxfleurs/garage#297
Co-authored-by: Alex <alex@adnab.me>
Co-committed-by: Alex <alex@adnab.me>
The function now computes an optimal assignation (with respect to partition size) that minimizes the distance to the former assignation, using flow algorithms.
This commit was written by Mendes Oulamara <mendes.oulamara@pm.me>
This removes the >1mb s3_copy restriction.
This restriction doesn't seem to be documented anywhere (I could be wrong). It also causes some software to fail (such as #248).
Co-authored-by: Rob Landers <landers.robert@gmail.com>
Reviewed-on: Deuxfleurs/garage#280
Co-authored-by: withinboredom <landers.robert@gmail.com>
Co-committed-by: withinboredom <landers.robert@gmail.com>
This change helps ensure that nodes for each partition are spread
over all datacenters, a property that wasn't ensured previously
when going from a 2 DC deployment to a 3 DC deployment
- Global dependencies updated in Cargo.lock
- New module created in src/admin to host:
- the (future) admin REST API
- the metric collection
- add configuration block
No metrics implemented yet
This commit adds support to discover garage instances running in
kubernetes.
Once enabled by setting `kubernetes_namespace` and
`kubernetes_service_name` garage will create a Custom Resources
`garagenodes.deuxfleurs.fr` with nodes public key as the resource name.
and IP and Port information as spec in the namespace configured by
`kubernetes_namespace`.
For discovering nodes the resources are filtered with the optionally set
`kubernetes_service_name` which sets a label
`garage.deuxfleurs.fr/service` on the resources.
This allows to separate multiple garage deployments in a single
namespace.
the `kubernetes_skip_crd` variable allows to disable the creation of the
CRD by garage itself. The user must deploy this manually.
Nodes would stabilize on different encoding formats for the values,
some having the pre-migration format and some having the post-migration
format. This would be reflected in the Merkle trees never converging
and thus having an infinite resync loop.
Implement ListMultipartUploads, also refactor ListObjects and ListObjectsV2.
It took me some times as I wanted to propose the following things:
- Using an iterator instead of the loop+goto pattern. I find it easier to read and it should enable some optimizations. For example, when consuming keys of a common prefix, we do many [redundant checks](https://git.deuxfleurs.fr/Deuxfleurs/garage/src/branch/main/src/api/s3_list.rs#L125-L156) while the only thing to do is to [check if the following key is still part of the common prefix](https://git.deuxfleurs.fr/Deuxfleurs/garage/src/branch/feature/s3-multipart-compat/src/api/s3_list.rs#L476).
- Try to name things (see ExtractionResult and RangeBegin enums) and to separate concerns (see ListQuery and Accumulator)
- An IO closure to make unit tests possibles.
- Unit tests, to track regressions and document how to interact with the code
- Integration tests with `s3api`. In the future, I would like to move them in Rust with the aws rust SDK.
Merging of the logic of ListMultipartUploads and ListObjects was not a goal but a consequence of the previous modifications.
Some points that we might want to discuss:
- ListObjectsV1, when using pagination and delimiters, has a weird behavior (it lists multiple times the same prefix) with `aws s3api` due to the fact that it can not use our optimization to skip the whole prefix. It is independant from my refactor and can be tested with the commented `s3api` tests in `test-smoke.sh`. It probably has the same weird behavior on the official AWS S3 implementation.
- Considering ListMultipartUploads, I had to "abuse" upload id marker to support prefix skipping. I send an `upload-id-marker` with the hardcoded value `include` to emulate your "including" token.
- Some ways to test ListMultipartUploads with existing software (my tests are limited to s3api for now).
Co-authored-by: Quentin Dufour <quentin@deuxfleurs.fr>
Reviewed-on: Deuxfleurs/garage#171
Co-authored-by: Quentin <quentin@dufour.io>
Co-committed-by: Quentin <quentin@dufour.io>
- Fix bucket delete
- fix merge of bucket creation date
- Replace deletable with option in aliases
Rationale: if two aliases point to conflicting bucket, resolving
by making an arbitrary choice risks making data accessible when it
shouldn't be. We'd rather resolve to deleting the alias until
someone puts it back.
- ensure bucket names are correct aws s3 names
- when making aliases, ensure timestamps of links in both ways are the
same
- fix small remarks by trinity
- don't have a separate website_access field
fix#77
this does not store anything but a on/off switch for website, and does not implement GetBucketWebsite as it would require storing more. GetBucketWebsite should be pretty easy to implement once data is stored though.
Co-authored-by: Trinity Pointard <trinity.pointard@gmail.com>
Reviewed-on: Deuxfleurs/garage#174
Co-authored-by: trinity-1686a <trinity.pointard@gmail.com>
Co-committed-by: trinity-1686a <trinity.pointard@gmail.com>
fix#161
Current request router was organically grown, and is getting messier and messier with each addition.
This router cover exaustively existing API endpoints (with exceptions listed in [#161(comment)](Deuxfleurs/garage#161 (comment)) either because new and old api endpoint can't feasabily be differentied, or it's more lambda than s3).
Co-authored-by: Trinity Pointard <trinity.pointard@gmail.com>
Reviewed-on: Deuxfleurs/garage#163
Reviewed-by: Alex <alex@adnab.me>
Co-authored-by: trinity-1686a <trinity.pointard@gmail.com>
Co-committed-by: trinity-1686a <trinity.pointard@gmail.com>
- change the terminology: the network configuration becomes the role
table, the configuration of a nodes becomes a node's role
- the modification of the role table takes place in two steps: first,
changes are staged in a CRDT data structure. Then, once the user is
happy with the changes, they can commit them all at once (or revert
them).
- update documentation
- fix tests
- implement smarter partition assignation algorithm
This patch breaks the format of the network configuration: when
migrating, the cluster will be in a state where no roles are assigned.
All roles must be re-assigned and commited at once. This migration
should not pose an issue.
- Explicit "replication_mode" configuration parameters that takes
either "none", "2" or "3" as values, instead of letting user configure
replication factor themselves. These are presets whose corresponding
replication/quorum values can be found in replication/mode.rs
- Explicit support for single-node and two-node deployments
(number of nodes must be at least "replication_mode", with "none"
we can have only one node)
- Ring is now stored much more compactly with 256*8 + n*32 bytes,
instead of 256*32 bytes
- Support for gateway-only nodes that do not store data
(these nodes still need a metadata_directory to store the list
of bucket and keys since those are stored on all nodes; it also
technically needs a data_directory to start but it will stay
empty unless we have bugs)
- Use quick_xml and serde for all XML response returned by the S3 API.
- Include tests for all structs used to generate XML
- Remove old manual XML escaping function which was unsafe
- return XML errors
- implement AuthorizationHeaderMalformed error to redirect clients to
correct location (used by minio client)
- implement GetBucketLocation
- fix DeleteObjects XML parsing and response
With the previous behaviour, repairing could see some data as absent
and decide that the object or version was deleted,
thus going on to delete the version and blocks.
In the case where read_quorum + write_quorum <= replication_factor
however, entries may not yet be returned by the get, thus data would
have been deleted that should hot have been. The new behavior is more
cautious and just skips the entry when the warning is emitted.
droping the slot later (after reading the request response)
means that we aren't freeing our quota slot,
so the maximum number of simultaneous requests now also counts the
response reading phase
TODO next: quotas per rpc destination node, or maybe per datacenter (?)
- ListBucket does not require any of the parameters (delimiter,
prefix, max-keys, etc)
- URLs are properly percent_decoded
- PutObject and DeleteObject calls now answer correctly
(empty body, version id in the x-amz-version-id: header)
Use Notify instead of stupid sleep in background worker
Use Semaphore to limit concurrent requests in rpc_client
Make more background tasks cancellable
Trying to separate:
1. Stuff for handling the swarm of nodes and generic table data replication
2. Stuff for the object store core application: metadata tables and block management
3. Stuff for the S3 API
