prisma-risk / tsoracle / 26350727505

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix(openraft-toolkit): fsync truncate_after and purge writes (#260) (#279)

truncate_after and purge committed their WriteBatch via the non-synced
self.db.write(batch), whereas save_vote and append use
write_opt(batch, write_sync_opts()) with set_sync(true). Each op is a
single atomic batch, so a crash loses the whole op cleanly — the hazard
is durability: an acknowledged truncate/purge could be lost on power
loss while a synced save_vote survives, so recovery could present a
durable new-term vote alongside a non-durable, un-truncated log tail —
a combination openraft expects to be durably consistent.

Both now use write_opt with write_sync_opts(), matching save_vote and
append.

Add failpoint regression guards around the write path in both methods
(truncate::{before_write_batch,after_write_before_sync} and the purge::
equivalents), mirroring append's failpoints, plus four tests asserting
that a panic before the write loses nothing and a return after the
write keeps the mutation durable. These guard the write path's
position; the sync flag itself has no in-process observable effect and
is not unit-testable without OS-level power-loss injection.

Coverage Stats

20 of 20 new or added lines in 1 file covered. (100.0%)

4 existing lines in 1 file now uncovered.

10620 of 11204 relevant lines covered (94.79%)

487665.7 hits per line