S3-compatible object store for small self-hosted geo-distributed deployments https://garagehq.deuxfleurs.fr/
Go to file
Alex db6f1f35a8 Rename epidemic_factor to epidemic_fanout (that's what it is); complete conf example in readme 2020-06-30 15:03:34 +02:00
src Rename epidemic_factor to epidemic_fanout (that's what it is); complete conf example in readme 2020-06-30 15:03:34 +02:00
.gitignore Work on API 2020-04-28 10:18:14 +00:00
Cargo.lock Implement HTTP ranges in get 2020-05-04 13:09:23 +00:00
Cargo.toml Split code for modular compilation 2020-04-24 10:10:01 +00:00
LICENSE First commit: skeleton for something great 2020-04-05 23:33:42 +02:00
Makefile Implement multipart uploads 2020-04-26 20:39:32 +00:00
README.md Rename epidemic_factor to epidemic_fanout (that's what it is); complete conf example in readme 2020-06-30 15:03:34 +02:00
TODO update todo 2020-04-26 20:50:23 +00:00
genkeys.sh Cleanup 2020-04-12 19:18:31 +02:00
rustfmt.toml Fix the Sync issue. Details: 2020-04-10 22:01:48 +02:00
test_delete.sh Do not insert deletion marker if there is no object to delete 2020-04-21 14:33:12 +00:00
test_read.sh More aggressive sync timings & improve other stuff 2020-04-19 17:59:59 +00:00
test_write.sh Better concurrency: 2020-04-22 16:51:52 +00:00



Garage is a lightweight S3-compatible distributed object store, with the following goals:

  • As self-contained as possible
  • Easy to set up
  • Highly resilient to network failures, network latency, disk failures, sysadmin failures
  • Relatively simple
  • Made for multi-datacenter deployments

Non-goals include:

  • Extremely high performance
  • Complete implementation of the S3 API
  • Erasure coding (our replication model is simply to copy the data as is on several nodes)

Our main use case is to provide a distributed storage layer for small-scale self hosted services such as Deuxfleurs.

Setting up Garage

Use the genkeys.sh script to generate TLS keys for encrypting communications between Garage nodes. The script takes no arguments and will generate keys in pki/. This script creates a certificate authority garage-ca which signs certificates for individual Garage nodes. Garage nodes from a same cluster authenticate themselves by verifying that they have certificates signed by the same certificate authority.

Garage requires two locations to store its data: a metadata directory, and a data directory. The metadata directory is used to store metadata such as object lists, and should ideally be located on an SSD drive. The data directory is used to store the chunks of data of the objects stored in Garage. In a typical deployment the data directory is stored on a standard HDD.

Garage does not handle TLS for its S3 API endpoint. This should be handled by adding a reverse proxy.

Create a configuration file with the following structure:

block_size = 1048576			# objects are split in blocks of maximum this number of bytes

metadata_dir = "/path/to/ssd/metadata/directory"
data_dir = "/path/to/hdd/data/directory"

rpc_bind_addr = "[::]:3901"		# the port other Garage nodes will use to talk to this node

bootstrap_peers = [
	# Ideally this list should contain the IP addresses of all other Garage nodes of the cluster.
	# Use Ansible or any kind of configuration templating to generate this automatically.

max_concurrent_rpc_requests = 12
data_replication_factor = 3
meta_replication_factor = 3
meta_epidemic_fanout = 3

# NOT RECOMMENDED: you can skip this section if you don't want to encrypt intra-cluster traffic
# Thanks to genkeys.sh, generating the keys and certificates is easy, so there is NO REASON NOT TO DO IT.
ca_cert = "/path/to/garage/pki/garage-ca.crt"
node_cert = "/path/to/garage/pki/garage.crt"
node_key = "/path/to/garage/pki/garage.key"

api_bind_addr = "[::1]:3900"	# the S3 API port, HTTP without TLS. Add a reverse proxy for the TLS part.
region = "garage"				# set this to anything. S3 API calls will fail if they are not made against the region set here.

Build Garage using cargo build --release. Then, run it using either ./target/release/garage server -c path/to/config_file.toml or cargo run --release -- server -c path/to/config_file.toml.

Set the RUST_LOG environment to garage=debug to dump some debug information. Set it to garage=trace to dump even more debug information. Set it to garage=warn to show nothing except warnings and errors.

Setting up cluster nodes

Once all your garage nodes are running, you will need to:

  1. check that they are correctly talking to one another;
  2. configure them with their physical location (in the case of a multi-dc deployment) and a number of "ring tokens" proportionnal to the storage space available on each node;
  3. create some S3 API keys and buckets;
  4. ???;
  5. profit!

To run these administrative tasks, you will need to use the garage command line tool and it to connect to any of the cluster's nodes on the RPC port. The garage CLI also needs TLS keys and certificates of its own to authenticate and be authenticated in the cluster. A typicall invocation will be as follows:

./target/release/garage --ca-cert=pki/garage-ca.crt --client-cert=pki/garage-client.crt --client-key=pki/garage-client.key <...>

Notes to self

What to repair

  • tables: to do a full sync of metadata, should not be necessary because it is done every hour by the system
  • versions and block_refs: very time consuming, usefull if deletions have not been propagated, improves garbage collection
  • blocks: very usefull to resync/rebalance blocks betweeen nodes