Hide Forgot
Description of problem: when use descheduler operator to create a descheduler, it not use down stream image(quay.io/openshift-release-dev/ocp-v4.0-art-dev:XXX) Version-Release number of selected component (if applicable): [core@ip-10-0-26-190 ~]$ oc get clusterversion NAME VERSION AVAILABLE PROGRESSING SINCE STATUS version 4.0.0-0.1 True False 6h Cluster version is 4.0.0-0.1 [core@ip-10-0-26-190 ~]$ oc version oc v4.0.0-0.125.0 kubernetes v1.11.0+406fc897d8 features: Basic-Auth GSSAPI Kerberos SPNEGO Server https://qe-jialiu3-api.qe.devcluster.openshift.com:6443 kubernetes v1.11.0+c69f926354 How reproducible: always Steps to Reproduce: 1.create a descheduler-operator and descheduler by deploy (refer to: https://polarion.engineering.redhat.com/polarion/#/project/OSE/workitem?id=OCP-21205) Pay attention to： modify operator.yaml, replace the content of field "image" with "quay.io/openshift-release-dev/ocp-v4.0-art-dev:v4.0.0-0.139.0.0-ose-descheduler-operator" 2.check descheduler job pod's image [core@ip-10-0-26-190 ~]$ oc get pod example-descheduler-1-1547630580-zdxs6 -o yaml | grep -i image image: registry.svc.ci.openshift.org/openshift/origin-v4.0:descheduler imageID: registry.svc.ci.openshift.org/openshift/origin-v4.0@sha256:27e026b56615259eda5903f398dc3f2472a926242cc2c7bc84923066ed6da545 Actual results: Expected results: in step2, the Pods should use downstream image built from RHEL: quay.io/openshift-release-dev/ocp-v4.0-art-dev:v4.0.0-0.139.0.0-ose-descheduler Additional info:
Hi MinLi, In the first step, you're modifying the container image for descheduler-operator where the job is created for descheduler, so there will be a mismatch. Having said that, descheduler operator needs to provide a way for descheduler which it is managing. I have created a PR for it, https://github.com/openshift/descheduler-operator/pull/41/files With this, we need to update the CR, with the specified image instead of updating deployment, if you update deployment, it will just change the descheduler-operator's image location. This is how CR would look like: apiVersion: descheduler.io/v1alpha1 kind: Descheduler metadata: name: descheduler-cr spec: schedule: "*/1 * * * ?" image: "quay.io/openshift/origin-descheduler:latest" #Please note that this field needs to be updated for descheduler, not descheduler-operator strategies: - name: "lownodeutilization" params: - name: "cputhreshold" value: "10" - name: "memorythreshold" value: "20" - name: "memorytargetthreshold" value: "30" Once the PR gets lgtm'ed and merged, we shouldn't see any problem with it.
@ravig, I understand what you mean, thx~ The reason why I modify the descheduler-operator image in step 1 is that: the descheduler-operator image should also use down stream image according to "Beta Images for OCP 4.0" (https://docs.google.com/spreadsheets/d/1n6lEgEPjs7rtSK9WEGXb1Z4oVse7MM2Q8jrGpsR4AFY/edit#gid=1869224058). FYI~
MinLi, The PR got merged. Can you please verify the fix suggested above for updating images of both descheduler-operator and descheduler?
@rgudimet Is there a guideline or readme which image is proper to use? Should the image name be ose-descheduler? For images ready to be released to customers, they are built from RHEL. Most origin-* images are built from CentOS.
@ravig , after update descheduler image as you mentioned, the problem still reproduced. env info: [core@ip-10-0-15-55 ~]$ oc version oc v4.0.0-0.147.0 kubernetes v1.11.0+dde478551e features: Basic-Auth GSSAPI Kerberos SPNEGO Server https://qe-jialiu2-api.qe.devcluster.openshift.com:6443 kubernetes v1.11.0+8868a98a7b [core@ip-10-0-15-55 ~]$ oc get clusterversion NAME VERSION AVAILABLE PROGRESSING SINCE STATUS version 4.0.0-0.2 True False 1d Cluster version is 4.0.0-0.2
@MinLi, I am not sure why are you looking at clusterversion? Any specific reason for this? Shouldn't we check for descheduler deployment? > Is there a guideline or readme which image is proper to use? Should the image name be ose-descheduler? I can update the readme in repo but I am not sure about the image. It is configurable as such, we can choose whatever the image we want it to be(as long as we have access to registry).
MinLi, Looking at the test in Polarion, you need to specify spec.image in the `Kind: Descheduler` you create to override the default descheduler image of `quay.io/openshift/origin-descheduler:latest`
(In reply to ravig from comment #7) > @MinLi, I am not sure why are you looking at clusterversion? Any specific > reason for this? Shouldn't we check for descheduler deployment? > @ravig, QE need attach env info when verify any bug, it's a must requirements. > I specify spec.image to "quay.io/openshift/origin-descheduler:latest" in "kind: Descheduler" CR file, but when check descheduler job pod's image, it is as the same as before: $ oc get pod example-descheduler-1-XXX -o yaml | grep -i image image: registry.svc.ci.openshift.org/openshift/origin-v4.0:descheduler imageID: registry.svc.ci.openshift.org/openshift/origin-v4.0@sha256:27e026b56615259eda5903f398dc3f2472a926242cc2c7bc84923066ed6da545 > > Is there a guideline or readme which image is proper to use? Should the image name be ose-descheduler? > > I can update the readme in repo but I am not sure about the image. It is > configurable as such, we can choose whatever the image we want it to be(as > long as we have access to registry). >@ravig, By my understanding, user can configure image of descheduler and descheduler-operator according to release image list? If so, you need update the image of readme in repo any time we release a OCP version? Is this what you mean?
(In reply to Seth Jennings from comment #8) > MinLi, > > Looking at the test in Polarion, you need to specify spec.image in the > `Kind: Descheduler` you create to override the default descheduler image of > `quay.io/openshift/origin-descheduler:latest` Seth Jennings, I do specify spec.image to "quay.io/openshift/origin-descheduler:latest" in "kind: Descheduler" CR file, but descheduler job pod's image is as the same as before, pls refer to Comment 9.
(In reply to MinLi from comment #10) > (In reply to Seth Jennings from comment #8) > > MinLi, > > > > Looking at the test in Polarion, you need to specify spec.image in the > > `Kind: Descheduler` you create to override the default descheduler image of > > `quay.io/openshift/origin-descheduler:latest` > > Seth Jennings, I do specify spec.image to > "quay.io/openshift/origin-descheduler:latest" in "kind: Descheduler" CR > file, but descheduler job pod's image is as the same as before, pls refer to > Comment 9. Sorry, Pls see the "UP-TO-DATE reply" in Comment 11.
verified! image: ose-descheduler-operator: quay.io/openshift-release-dev/ocp-v4.0-art-dev:v4.0.0-0.171.0.0-ose-descheduler-operator ose-descheduler: quay.io/openshift-release-dev/ocp-v4.0-art-dev:v4.0.0-0.171.0.0-ose-descheduler version info: [core@ip-10-0-19-185 ~]$ oc version oc v4.0.0-0.170.0 kubernetes v1.12.4+45dbe929fa features: Basic-Auth GSSAPI Kerberos SPNEGO [root@localhost lyman]# oc get clusterversion NAME VERSION AVAILABLE PROGRESSING SINCE STATUS version 4.0.0-0.nightly-2019-02-13-204401 True False 4h54m Error while reconciling 4.0.0-0.nightly-2019-02-13-204401: the cluster operator monitoring is failing
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2019:0758