pgrac is built in public and still early — testing, feedback, and patches all help. You don't need to be a Postgres hacker to be useful here.
- Build it and run the Quick start (see the README): build with
--enable-cluster, bootstrap a node withpgrac-init, and checkpg_cluster_nodes/pg_cluster_ic_peers. File anything that breaks. - Poke holes in the architecture (
docs/architecture/overview.md). If you've operated Oracle RAC or any shared-storage cluster, tell us where the design is wrong — correctness under Cache Fusion, recovery, and fencing are the areas we most want scrutiny on. - Pick a
good first issue. These are scoped so they don't require deep cluster internals.
The cluster substrate is live (TCP interconnect, LMON heartbeat,
SCN/ITL/undo block format, multi-node bootstrap). Cache Fusion / GES /
cross-node recovery are scaffolded and return FEATURE_NOT_SUPPORTED — patches
there are best coordinated via an issue first, since the protocols are still in
flux.
- Don't regress the non-cluster path. The
--disable-clusterbuild must stay binary-equivalent to upstream PostgreSQL 16.13 and pass the full 219-test regression suite. - Assertions follow
AGENTS.md— a required runtime check must have a real production branch, not live only insideAssert(). - Match upstream PostgreSQL C style (
.clang-formatand.editorconfigare in the tree).
Open an issue at https://github.com/sqlrush/pgrac/issues. For anything that looks like a data-corruption or memory-safety bug, please flag it in the title.