netsa v1.14.2: upgrades and restarts no longer lose logs being processed
netsa v1.14.2 is a reliability release: we reviewed the read checkpoints, warm standby, incremental snapshots, AI guardrails and shadow mode added in 1.12 to 1.14, and fixed several issues that could affect data integrity during upgrades, restarts or failover.
Logs in the processing queue are no longer lost on upgrade or restart
Previously the processing workers stopped immediately on exit while the read checkpoint had already moved forward when logs entered the queue, so during catch-up or traffic peaks, unprocessed logs in the queue could be lost for good. Now:
- The read checkpoint only moves past logs that have been processed, even when workers finish batches in a different order;
- On exit, logs already in the queue are processed first (for up to 15 seconds), and whatever times out is read again at the next start;
- In the same scenario, all 601,000 entries were stored after a restart, with no duplicates and nothing missing.
Other improvements
- The read checkpoint no longer moves past an unfinished half line at the end of a file;
- After taking over, a warm standby no longer reads its local logs from the start as new content, avoiding floods of old alerts;
- Incremental snapshots in a backup are read before the full snapshot, so restores always line up;
- In AI triage, targets in past decisions and similar cases sent to the model are wrapped in delimiters too;
- Shadow mode hit counts are reset when a rule's match conditions change.
How to upgrade
docker compose pull && docker compose up -d
The full change log is in the project's release notes. For everything netsa can do, visit the product page.