fix: ClusterBackup watchdog + heartbeat; backup-restore doc + power-loss drill#57
Merged
Merged
Conversation
…oss drill (match#36, admin-gateway#9) Root cause of the silent stall (live-diagnosed): the PM launched ClusterBackupApp with no environment, so CLUSTER_ADDRESSES defaulted to "localhost" = ONE consensus endpoint (node0's member-facing port). With SourceType.LEADER, the moment the leader was any node but 0, the agent excluded its only endpoint and wedged in PublicationGroup.nextCursor()'s do-while forever: no duty cycle -> no client keepalives -> its own driver timed the client out and unbound the consensus-response subscription; no errors, no stall events, mark-file activity frozen, one core burned at ~90% for days while the process looked "running". (Companion env fix in admin-gateway; the single-endpoint+exclusion infinite loop is an Aeron 1.51 edge case worth reporting upstream.) Separately, the local archive conductor died under storm load (ConductorServiceTimeoutException, service interval 25s > 10s) with the same silent outcome. Defenses added here: - backup-watchdog thread: tracks RESPONSE-side listener callbacks only (backup response, recording-log update, live-log advance — query sends don't count, a wedged agent still sends). No progress for BACKUP_STALL_EXIT_SEC (default 300s, env-overridable) -> loud FATAL + Runtime.halt(3) so the PM restarts the whole stack fresh, bounded by the PM crash-loop cap. halt, not exit: shutdown hooks can block on the same wedged agents. - heartbeat backup-progress.json written atomically every 5s next to the backup data (pid, last query/response/live-log ms, live-log position, snapshots retrieved, stall count, OK/STALLED) — consumed by the admin gateway's backup-info/status freshness reporting. - listener now logs backup responses + recording-log updates (~1/min). - docs/backup-restore.md: architecture, RPO/RTO, freshness monitoring, single-node + full-cluster restore procedures. - tools/durability/backup_restore_drill.sh: scripted power-loss drill (kill -9 all, wipe /dev/shm, restore every node from disk, verify consensus); gated behind an explicit acknowledgement flag. Live-verified with the admin-gateway env fix: first successful backup cycle since Jul 2 — response from leader (member 1), 2 snapshots retrieved, recording.log populated, live-log position advancing, heartbeat state OK, backup-info fresh:true. Closes #36 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #36 (P1.3, v0.3.0-beta). Companion PR in admin-gateway (env fix + freshness surfacing) — merge both; the admin-gateway one carries the actual unwedge.
See the commit message for the full root-cause chain. Highlights:
nextCursor()loop (agent wedged mid-call, ~90% of a core, zero errors/logs, mark-file activity frozen, consensus subscription unbound by client timeout). Aeron 1.51 single-endpoint edge case; our config was the trigger.BACKUP_STALL_EXIT_SEC(300s) →halt(3)→ PM restart under the crash-loop cap. Heartbeat JSON every 5s for admin freshness reporting.docs/backup-restore.md(architecture, RPO ≈ live-log lag bounded by 60s query interval, restore procedures) +tools/durability/backup_restore_drill.sh(build(deps-dev): bump org.apache.maven.plugins:maven-surefire-plugin from 3.2.5 to 3.5.6 #9 exit criterion; destructive, ack-flag-gated, NOT yet run).backup-infofresh:true.🤖 Generated with Claude Code