amadan
Flaky data race: TestHeldCheckpointVacuumRestoreAndTransfer (litestream writeLTXFromDB vs DB.Close) open
go test ./internal/lease/ -run TestHeldCheckpointVacuumRestoreAndTransfer -race -count=8 reports a data race about once in eight runs (one in three full ./internal/lease/... race runs), on origin/main 8d8e4bae as well as on feat/hostclaim-storage, which only adds files to internal/lease.
The race is in the pinned Litestream v0.5.14: DB.writeLTXFromDB (db.go:2099) against DB.Close (db.go:802), reached from the test at checkpoint_test.go:242. CI does not run -race (Makefile:18), so it never shows there.
It is not a Plan 2 regression. It matters for the claim work because §6.4a's clean-close certificate depends on the replicator's close being well-ordered with its last LTX write: if Close can race a write in flight, a "closed" replicator may not have finished its last upload. Worth a look before Plan 6 wires certificates to Replicator.Close.
Found by: platform-1e, Plan 2 review (claim spec rev 6.14).