mirror of
https://github.com/n8n-io/n8n.git
synced 2026-10-22 04:21:32 +02:00
4031d7b2a3
- Keep the all-quarantined grace timer referenced: with every lane dead, queued acquire() promises hold no live handles, so an unref'd deadline let the process exit 0 mid-run instead of aborting loudly. - Stash transport-evicted builds for the end-of-run cleanup drain — their restored workflows, data tables, and thread/sandbox previously leaked once the cache entry was evicted for the retry. - Bound run-debug capture with a 60s deadline so a stalled debug endpoint cannot block per-case cleanup (and with it the run) indefinitely. - Widen the baseline completion-marker probe window to 25 candidates so a burst of failed captures cannot hide an older completed baseline. - Move the builder-cache prune out of the shared load-n8n-docker action into a pre-eval disk-runway check: pressure develops during the eval, not at job start, and the shared action's prune would cold-start other workflows' builds on sticky-disk runners. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| build-n8n-docker | ||
| ci-filter | ||
| docker-registry-login | ||
| load-n8n-docker | ||
| setup-nodejs | ||