Queue latency with Celery integration interpreted as milliseconds, whereas it's computed in seconds

[ElstnerM]

How do you use Sentry?

Sentry Saas (sentry.io)

Version

2.339

Steps to Reproduce

1. Integrate a Celery queue via the Python SDK's Celery Integration
2. Schedule tasks ending up in this queue
3. Go to Insights -> Backend -> Queues
4. Open average duration graph
5. Look at average time in qeueue
6. Observe values

Expected Result

Latencies should match the actual time that tasks spent in the queue before being picked up by a worker

Actual Result

Graph shows 1/1000 of the actual values (e.g. instead of displaying seconds, it's displaying milliseconds)

Digging into the JSON raw data of relevant spans, the value messaging.message.receive.latency appears twice in the JSON (see attached screenshot), once without the milli unit and once with it. My hunch is that the unit is inserted erroneously downstream from the Celery integration, and picked up by the graph to show 'ms' on the y axis. The Celery integration definitely computes and writes the value in seconds (see second screenshot).

[linear]

PY-1790 Queue latency with Celery integration interpreted as milliseconds, whereas it's computed in seconds

[sentrivana]

Hey @ElstnerM, thanks for reporting and investigating. That indeed looks wrong, we need to figure out where the millisecond is coming from and if we should be sending the latency in seconds in the first place.

[sentrivana]

We should indeed be sending the value in milliseconds rather than seconds. Thanks for catching this! Will roll a fix.