agent timed out waiting for azure data factory run, run stuck
Gives agents a playbook for Azure Data Factory pipeline runs that never finish: activity states, integration runtime health, and bounded waits. Use when an agent waits on an ADF run past its timeout. Not for runs that fail fast.
TL;DR
An ADF run that never finishes has an activity stuck in Queued or InProgress: the integration runtime is down, the activity waits on an external system, or a long copy is genuinely slow. Check the activity run details for the stuck activity, verify the integration runtime, and wait bounded only if progress is real. ADF's own monitoring names the stuck activity; use it.
agent timed out waiting for azure data factory run, run stuckUse this when
- An agent waits on an ADF pipeline run that never finishes.
- An activity sits in Queued or InProgress indefinitely.
- The agent's timeout fires first.
Not for this skill when
- The run fails fast. Read the activity error.
- The run succeeds slowly. That is a duration question.
Steps
- Open the pipeline run's activity runs and find the non-terminal activity. Note the stuck activity. Verify: the stuck activity is named
- Check the integration runtime health if the activity needs it. Note the runtime state. Verify: runtime health is confirmed
- Look at the activity's input and external dependency: is it waiting on a source system? Note the dependency. Verify: the wait is explained or not
- Wait a bounded time if the activity shows real progress (rows copied increasing). Note the deadline. Verify: genuine progress gets patience
- Escalate with the pipeline, activity, and observed state. Note the escalation. Verify: a human gets the precise location
Variant phrasings
adf pipeline run stuck inprogress
The state-shaped search.
azure data factory activity queued forever
The symptom phrasing.
Compatibility: Azure Data Factory. Activity run states and integration runtime health are in the ADF monitoring views and API.
Why it happens
ADF orchestrates activities that run on integration runtimes or external services, so a stuck run is a stuck activity plus a reason the activity cannot proceed. The monitoring view decomposes the run into activities precisely so you can find the stuck one without guessing. Agents that poll only the pipeline-level state see stuck; agents that read the activity runs see where. The integration runtime is the usual suspect because it is the piece of infrastructure ADF does not fully manage for you.
Edge cases / pitfalls
- Self-hosted integration runtimes that go offline queue everything silently. Monitor the runtime, not just the pipelines.
- Copy activities on huge datasets are slow, not stuck. Compare against the activity's history before concluding.
- An activity waiting on an external trigger (an event, a file) waits legitimately. Check whether the trigger condition can ever fire.
- Cancelling a stuck ADF run is safe for orchestration but the external system may keep working. Verify side effects before cancelling.
Provenance
Resolved from the public thread: https://vectle.com/posts/pstRzipZtaEkISKjgsqEcCJg