databendlabs / openraft / 19390656609

88%

main: 88%

Build Type

push

github

Committed by drmingdrmer

Commit Message

fix: prevent heartbeat-induced log reversion panic

Use per-follower matching log id in heartbeat `prev_log_id` instead of
committed log id to prevent false reversion detection when followers
fall behind.

Problem:

When using committed log id `c` as `prev_log_id` in heartbeat
AppendEntries, the follower may not have `c` yet and returns conflict.
If this conflict response arrives after the follower successfully
replicates `c`, the leader updates its matching cursor to `c` and
incorrectly interprets the delayed conflict as log reversion.

The root cause is the leader using a non-existent log id as
`prev_log_id`.

Solution:

Use the known matching log id (last confirmed replicated) as
`prev_log_id` in heartbeats, eliminating false conflicts.

Changes:
- Add matching field to HeartbeatEvent
- Change heartbeat workers to per-follower channels
- Include matching log id from progress in heartbeat events
- Use matching log id as `prev_log_id` in heartbeat append-entries
- Add test reproducing issue #1500

- Fix: #1500

Run Details

44 of 46 new or added lines in 4 files covered. (95.65%)

2 existing lines in 2 files now uncovered.

12893 of 13994 relevant lines covered (92.13%)

155043.24 hits per line