Confluence 8.2 indexing cancellation policy: best-effort to force-stop transition

Answer to the Question

For a workspace that mainly performs small edits but occasionally runs large page imports, the best‑effort policy is the default that usually offers the best balance between search quality and repair effort. It allows short indexing jobs to finish normally and only interrupts jobs that exceed the configured indexing.timeout.ms. Because most small edits finish well under the timeout, the index rarely becomes partially updated and manual repair is infrequent. If large imports regularly exceed the timeout, switching to force‑stop can prevent repeated search relevance hiccups, but it will increase the need for Integrity Checker runs or occasional full reindexes.

Why Best‑Effort Is Generally Preferable

• Small edits (< 10 KB) finish in a few seconds, far below typical timeouts.
• Jobs that exceed the timeout are merely interrupted; the index remains usable, though possibly with a temporary drop in relevance.
• No orphaned segments are left behind, so the Integrity Checker rarely reports issues.

When to Consider Force‑Stop

• Large imports that consistently exceed the timeout (e.g., > 30 min) and cause frequent relevance drops.
• You prefer to avoid repeated search degradation even if it means a higher repair cadence.

Verification Steps After a Timeout‑Induced Cancellation

1. Check Confluence logs for entries such as InterruptedException or Indexing task timed out to confirm the cancellation was triggered.
   • Command line: grep -i "timed out" $CONFLUENCE_HOME/logs/atlassian-confluence.log
2. Run the Integrity Checker (or the built‑in Lucene health check) to detect orphaned or incomplete segments.
   • Navigate to Admin > System > Indexing > Integrity Checker and start a scan.
   • Review the report for any "orphaned" or "corrupted" entries.
3. Verify search relevance by querying a set of representative pages, especially those affected by the large import.
   • Compare results against a recent successful indexing snapshot or a known good baseline.
4. Inspect index segment statistics in the index directory.
   • A sudden drop in segment count after a force‑stop may indicate incomplete segment creation.
5. **If any inconsistencies are found**, schedule a full reindex and re‑run the Integrity Checker to ensure all segments are healthy.

Missing Diagnostic Detail

To refine the recommendation, we need to know how often large imports exceed the configured timeout. Could you provide the average duration of your largest import jobs relative to the indexing.timeout.ms setting?