Do not check logs in serverless tests #2267
Conversation
These apps do not have them
|
run tests |
| @@ -19,4 +20,9 @@ public class ServerlessExtensionOpenShiftHttpMinimumReactiveIT extends HttpMinim | |||
| protected RequestSpecification givenSpec() { | |||
| return HTTPS_CLIENT_SPEC; | |||
| } | |||
|
|
|||
| @Override | |||
| @Disabled("Serverless deployments do not have logs which we can check") | |||
There was a problem hiding this comment.
wow, this is surprising
Can you share why is that?
app.getLogs() can't retrieve them or even with oc tool there are no logs?
There was a problem hiding this comment.
If you run oc status, there would be no pods to read logs from (which is, I guess, expected). Calling app.getLogs returns empty list.
There was a problem hiding this comment.
ok, I guess we would need to actively collect logs in FW for any serverless deployment before it gets scaled to zero.
There was a problem hiding this comment.
BTW, here how such deployment looks like:
Warning: apps.openshift.io/v1 DeploymentConfig is deprecated in v4.14+, unavailable in v4.10000+
In project fd-demonstration on server https://api.ocp4-17<...>
svc/app-00001 - <...> ports 80->8012, 443->8112
svc/app - kourier-internal.knative-serving-ingress.svc.cluster.local
svc/app-00001-private - <...> ports 80->8012, 443->8112, 9090->http-autometric, 9091->http-usermetric, 8022, 8012
deployment/app-00001-deployment deploys image-registry.openshift-image-registry.svc:5000/fd-demonstration/app@sha256:<...>,registry.redhat.io/openshift-serverless-1/serving-queue-rhel8@sha256:<...>
deployment #1 running for 22 minutes
bc/app source builds uploaded code on istag/openjdk-17:latest
-> istag/app:1.0.0-SNAPSHOT
build #1 succeeded 22 minutes ago
1 info identified, use 'oc status --suggest' to see details.
mvn clean verify -Dit.test=ServerlessExtensionOpenShiftHttpMinimumReactiveIT -Popenshift -Dts.openshift.delete.project.after.all=false -Dts.openshift.ephemeral.namespaces.enabled=false -Doc.reruns=0
from http/http-minimum-reactive
There was a problem hiding this comment.
When I tried this, with @fedinskiy 's command, the pod were present in the openshift while tests were searching for logs and remained there for quite a while even after the test failed. Logs from the pod were manually readable no problem and contained all the lines tests are searching for.
IDK why tests were not able to read those. But I agree that for now it is best to disable it investigate the issue with logs reading.
There was a problem hiding this comment.
@mocenas was it "build pod" or "app pod"? They can be distinguished by "Completed" status (or lack thereof). I have only seen the former.
There was a problem hiding this comment.
@mocenas was it "build pod" or "app pod"? They can be distinguished by "Completed" status (or lack thereof). I have only seen the former.
I was app pod. It was in state Running. (also build pod had "build" in name).
There was a problem hiding this comment.
Hmm, now I see temporary pod, which lives for roughly 80 seconds, but the last line in its logs contains "11:26:56,497 Installed features: [cdi, kubernetes, rest, rest-jackson, smallrye-context-propagation, vertx]", so no notification about interception there.
Maybe it is a bug in Quarkus interception, but for now we cann't check it automatically anyway
There was a problem hiding this comment.
Let's consider that you really can't inspect Knative serving function logs (which makes it basically unusable in production, I suppose they just export it automatically to grafana or something like that we don't have) - you still have running application that returns response. Can you set quarkus.log.file.enable and quarkus.log.file.path and read it from the file system inside Quarkus application resource method? Here you have operator resource that you call https://github.com/quarkus-qe/quarkus-test-suite/blob/main/http/http-minimum-reactive/src/test/java/io/quarkus/ts/http/minimum/reactive/HttpMinimumReactiveIT.java#L99, is there something that stops you from adding a new method that returns logs for certain category (class - so that it is short)?
Interestingly, this can be implemented as a FW feature as well when at least one extension is based on Vert.x HTTP (so 99 of 100 deployments), but I suspect that there is proper solution possible and it just needs a time that we don't have.
Summary
Follow-up to #2266
Please select the relevant options.
run testsphrase in comment)Checklist: