databendlabs / openraft / 26814810383

87%

Build Type

push

github

Committed by drmingdrmer

Commit Message

fix: drain stranded client responders on PurgeLog

# Summary

A former leader's pending client write could be superseded by a snapshot
install on a lagging follower, leaving its responder in the queue. The next
apply then panicked the RaftCore task on a purged log index, permanently
freezing that Raft group. PurgeLog now drains and fails such responders,
mirroring TruncateLog.

# Details

Root cause: when a follower lags past the snapshot's last index,
install_full_snapshot skips the conflict-truncate (the only step that drains
responders) because the local log does not reach that index, and removes the
superseded entries via PurgeLog. PurgeLog did not touch client_responders, so a
responder remained below the new purge boundary; the next apply called
get_log_id(index).unwrap() on a purged index and panicked.

Fix: PurgeLog drains responders up to the purged index and completes each with
ForwardToLeader. In normal operation purge follows apply, so the drained range
is empty and behavior is unchanged; it only fires on the snapshot-strand path.

A regression test reproduces the path: leader accepts a write, loses leadership
with the responder pending, lags, installs a snapshot past its local log, then
applies once.

- Fix: #1761

Coverage Stats

8 of 8 new or added lines in 1 file covered. (100.0%)

5 existing lines in 2 files now uncovered.

16383 of 18931 relevant lines covered (86.54%)

145291.83 hits per line