https://github.com/super-flat/parti
🪩 parti is cluster sharding via raft over gRPC
https://github.com/super-flat/parti
grpc partitioning raft sharding
Last synced: 2 months ago
JSON representation
🪩 parti is cluster sharding via raft over gRPC
- Host: GitHub
- URL: https://github.com/super-flat/parti
- Owner: super-flat
- Created: 2021-09-18T23:04:31.000Z (over 4 years ago)
- Default Branch: main
- Last Pushed: 2025-12-18T02:57:52.000Z (3 months ago)
- Last Synced: 2025-12-21T09:08:43.146Z (3 months ago)
- Topics: grpc, partitioning, raft, sharding
- Language: Go
- Homepage:
- Size: 366 KB
- Stars: 6
- Watchers: 3
- Forks: 1
- Open Issues: 12
-
Metadata Files:
- Readme: readme.md
Awesome Lists containing this project
README
# parti
[](https://github.com/super-flat/parti/actions/workflows/main.yml)
Parti provides a simple, partitioned messaging framework across a cluster of nodes. Send a message for a given partition to any node in the cluster, and the message will be forwarded to the node that owns that partition. The cluster will automatically rebalance as nodes are added and removed. Under the hood, Parti uses [hashicorp/raft](https://github.com/hashicorp/raft) to distribute work and manage membership.
### SAMPLE
```sh
# run node 1
make run1
# run node 2 (in another terminal)
make run2
# run node 3 (in another terminal)
make run3
# observe the partitions by asking node 1
go run ./client stats --addr 0.0.0.0:50101
# ping a partition from a given node
go run ./client ping --addr 0.0.0.0:50101 --partition 0
go run ./client ping --addr 0.0.0.0:50101 --partition 9
# send a message to various partitions and observe it forward
# curl 'localhost:50001/send?partition=9&message=msg2'
go run ./example send --partition 0 --addr 0.0.0.0:50001 hello world
# curl 'localhost:50001/send?partition=1&message=msg1'
go run ./example send --partition 9 --addr 0.0.0.0:50001 hello world
```
### Proposed rebalance procedure
1. tell all nodes the pause the partition
2. confirm with prior owner that they have shut down (or confirm they are dead and not in the cluster)
3. inform all nodes of the new owner and that it's started