n8n's observability and reliability week while Trigger.dev fixes delayed starts

n8n's reliability and tracing push

n8n's updates this week were mostly about keeping jobs and telemetry moving when infrastructure gets flaky. OpenTelemetry export survives a restart now even with Sentry enabled, and project span attributes flow through to node spans so traces stay attributable deeper into a run. Public transport got unwedged: pubsub subscriptions restore themselves after a Redis reconnect, and half-open subscriber connections are now detected rather than silently swallowing events. When a published workflow references a node type the instance can't load, the triggers are torn down cleanly instead of lingering half-registered.

Smaller n8n fixes, front and back

Instance reporting no longer fails for runs crossing the UTC midnight boundary, and the public API's credential PATCH endpoint returns something other than HTTP 405. Declarative routing is now constrained during base URL ownership checks, and listing credentials no longer pays to load every project member. Agents ship enabled by default, a notable flip for anyone who previously had to turn them on by hand. A handful of parallel patches to older n8n branches bump adm-zip and keep Qdrant payload handling in line.

Trigger.dev fixes delayed warm starts

Trigger.dev's bug this week was a nasty one for anyone on npm with Zod 3 pinned: a bad Zod import could make warm starts fail and push runs out by roughly five minutes, with cold starts untouched. The fix restores warm starts and requeues affected runs; nothing was lost, but you'll want to redeploy with the current CLI before trusting start latency again.

Inngest scopes sandbox secrets

Inngest's JS SDK added exact-name secrets to sandbox create options, letting you pass values like OPENAI_API_KEY and have them saved under the environment variable name your application already reads — with duplicate and literal-env collisions rejected up front.