prisma-risk / tsoracle / 26379037445

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix(driver-paxos): rebase snapshot baseline at recovery to avoid spurious post-restart snapshot (#375)

SnapshotPolicy::every(N) starts last_snapshot_at = 0 and is reconstructed
fresh on every restart. A node reopening its log at a decided index far
past N satisfies `decided_idx >= 0 + N` on its first post-recovery check,
firing an unintended snapshot.

ApplyEngine::recover already knows the recovered decided index (the cursor
after folding the decided suffix), so it now rebases the policy baseline to
it. The next snapshot then fires N entries past the restart point rather
than immediately. This extends the existing "recovery skips compaction"
intent to the first decision after recovery, without persisting any new
state or touching the storage codec.

Adds SnapshotPolicy::rebase, unit coverage for the rebased cadence, and an
integration test that reproduces the spurious snapshot via the
tsoracle.paxos.snapshot.total metric.

Closes #258

Coverage Stats

57 of 60 new or added lines in 3 files covered. (95.0%)

2 existing lines in 1 file now uncovered.

12178 of 12847 relevant lines covered (94.79%)

425172.22 hits per line