Bug 1329994

Created attachment 1150316 [details]
deployer logs attached

I think this is the same issue with using non-secured registries (i.e. without a cert signed by a trusted CA) as was discussed and worked around in https://github.com/openshift/origin-aggregated-logging/issues/107#issuecomment-208964239 - can you confirm? What does the output of `oc get is` look like? For the deployment to work, tags need to be imported successfully - can you import them manually as in that link?

It looks like the triggers on the DCs aren't firing when you import the tags. I have seen this take some minutes, as if a controller were doing the rounds periodically looking for new tags to trigger deployments.

When you did manual deployments, with no trigger overriding the image, you get the dummy image definition:

Error syncing pod, skipping: failed to "StartContainer" for "elasticsearch" with ErrImagePull: "Error: image library/logging-elasticsearch:latest not found"

Two avenues to follow up:
1) why didn't the tag import trigger a deployment - the real problem.
2) put the real image in the DC definition as a fallback.

@lmeyer  Thanks for the instructions. I now understand it is the authenticated registry which caused the original problem. Unfortunately the work around in comment #8 doesn't work for me, EFK pods is still in Error status with ErrImagePull: "Error: image library/logging-elasticsearch:latest not found" in events after doing all the steps in comment #8.

However, I have worked around the EFK issue by specifying EFK and the logging-auth-proxy images directly in EFK dc, so this should not be a release blocker now. Set this to verified since Kibana UI is working fine after I get EFK pods running.