<@ULVA73B9P> I use to have logs like this appear in the dashboard, where the jobs name with a unique...
m
@Marvin I use to have logs like this appear in the dashboard, where the jobs name with a unique suffix would show:
Kubernetes job 'my-workflow-test-2v7s6' created in namespace 'research-prefect3'
but now I don't see this
m
thought for 207.7 seconds
That message is worker-emitted flow-run logging, and there’s a good reason you may have stopped seeing it. What I found: The message source - The log line
Copy code
Kubernetes job 'my-workflow-test-2v7s6' created in namespace 'research-prefect3'
is emitted by the Kubernetes worker, not by your flow code. - It is sent as a flow run log so it can appear in the run’s dashboard logs. Why it may no longer appear There are two execution paths in the Kubernetes worker now: -
run()
does include the
"Kubernetes job ... created"
log -
_initiate_run()
does not include that log So if your runs are now going through the
_initiate_run()
path instead of
run()
, you would stop seeing that message even though the job is still being created. This is the most likely explanation if: - nothing else about logging changed - runs still execute successfully on Kubernetes - you specifically miss only that “job created with unique suffix” line Other reasons it might disappear: -
PREFECT_LOGGING_TO_API_ENABLED=false
- log level raised above
INFO
- worker/API connectivity issues preventing log delivery But if other run logs still show up, then the changed code path is the stronger explanation. Important detail This is not about general worker log forwarding to the UI in the usual sense. That specific message is logged as a flow run log, so if flow run logs are working overall, the issue is more likely that the message simply is not emitted on the path your runs are taking. Relevant context This appears to be due to a relatively recent implementation difference in Prefect 3.x: -
_initiate_run()
was introduced for a non-blocking submission path - the
"Kubernetes job created"
message was added in
run()
but not in
_initiate_run()
So the behavior can look inconsistent depending on how the run is launched. Next steps If you want, I can help you narrow down which path your runs are taking and whether this is a regression in your current Prefect /
prefect-kubernetes
version. Useful links: - Prefect repo: PrefectHQ/prefect - Kubernetes integration source area: prefect-kubernetes source If you send me: - your Prefect version - your
prefect-kubernetes
version - how the run is launched (
deployment
,
serve
, ad-hoc submission, etc.) …I can tell you more precisely why that log disappeared.