Data & Analytics
An Airbyte Salesforce sync can keep running after data stops moving
Detect a Salesforce source request that stalls while the Airbyte job stays running.
November 10, 2026
A job can report running long after useful work has stopped. An Airbyte Salesforce issue describes source version 2.10.2 staying alive with no new log lines for many hours. The reporter traces a possible gap between timeouts added for token refresh and describe requests and unbounded stream data requests. This is an open issue, not a verified universal failure mode.
Monitor progress, not just process life
Track the timestamp of the last source record, state update and log line alongside job duration. Alert when those stop advancing beyond a stream-specific threshold. Airbyte's heartbeat guide distinguishes a live container from useful sync progress; the connector can remain responsive enough to heartbeat while an upstream read waits.
Check which Salesforce stream and API path was active, whether data volume or rate limits changed, and whether the socket produced an exception. Capture sanitized logs and connector version before canceling a stuck run. A cancellation is an operational recovery step, not proof of root cause.
Protect freshness commitments
Set a maximum tolerated data age for each downstream use. If the connection is locked by a long-running attempt, mark dependent datasets stale and stop claiming the next schedule completed. Compare the next successful sync's cursor and record counts with Salesforce before releasing dashboards.
Airbyte's Salesforce connector documentation lists separate authentication and API limitations. Rule those out with evidence rather than assuming every long run is this reported timeout path.
Adovo’s view
Availability is not freshness. A process that is alive but no longer moving data is an outage for the people using the result. Measure forward progress, give it a deadline and make stale data visible to its consumers.
