centra/photoncloud-monorepo
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
photoncloud-monorepo/flaredb/specs/004-multi-raft/quickstart.md
centra 8f94aee1fa Fix R8: Convert submodule gitlinks to regular directories 
- Remove gitlinks (160000 mode) for chainfire, flaredb, iam
- Add workspace contents as regular tracked files
- Update flake.nix to use simple paths instead of builtins.fetchGit

This resolves the nix build failure where submodule directories
appeared empty in the nix store.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2025-12-09 16:51:20 +09:00

1.5 KiB
Raw Blame History

Quickstart: Multi-Raft (Static → Split → Move)

Prerequisites

  • Nix or Rust toolchain per repo.
  • PD stub runs inline (tests use in-memory).

Run tests (recommended)

nix develop -c cargo test -q rdb-server::tests::test_multi_region
nix develop -c cargo test -q rdb-server::tests::test_split
nix develop -c cargo test -q rdb-server::tests::test_confchange_move

Or full suite:

nix develop -c cargo test -q

Manual smoke (single node, two regions)

  1. Launch PD stub (or ensure pdpb gRPC reachable).
  2. Start server: 
    nix develop -c cargo run -p rdb-server -- --pd-endpoint http://127.0.0.1:50051
  3. Verify routing:
    • Put key b"a" → Region1
    • Put key b"z" → Region2

Trigger split (dev)

  1. Run test_split or fill a region with writes.
  2. Observe log: ApplyCommand::Split and new region registered.

Move (rebalance) flow (simplified)

  1. Source store handles region; target store starts with PD meta.
  2. PD issues MoveRegion(region_id, from=src, to=dst).
  3. Source adds replica on target (ConfChange Add); target catches up; source can later remove itself (ConfChange Remove).
  4. Verify data on target: 
    nix develop -c cargo test -q move_region_replica_carries_data -- --nocapture

Notes

  • Key ranges must not overlap; nodes validate PD meta.
  • Raft logs and hard-state are prefixed by region_id to isolate shards.
  • Pending eventual writes are forwarded to leaders; local queue persists to disk to survive restart.