-
Story
-
Resolution: Unresolved
-
Major
-
None
-
None
-
2
-
False
-
None
-
False
-
Pipelines Sprint 227, Pipelines Sprint 228, Pipelines Sprint 229, Pipelines Sprint 230, Pipelines Sprint 231
According to Tekton upstream documentation:
To get a "Guaranteed" QoS, a TaskRun pod must have compute resources set for all of its containers, including init containers which are injected by Tekton, and all containers must have their requests equal to their limits. This can be achieved by using LimitRanges to apply default requests and limits.
This is not what seems to happen when it has been tested.
Find below the LimitRange created:
$ oc get limitranges -n sre-ci-test project-limits -oyaml
apiVersion: v1
kind: LimitRange
metadata:
annotations:
app.kubernetes.io/created-by: argocd
app.kubernetes.io/managed-by: helm
kubectl.kubernetes.io/last-applied-configuration: |
{"apiVersion":"v1","kind":"LimitRange","metadata":{"annotations":{"app.kubernetes.io/created-by":"argocd","app.kubernetes.io/managed-by":"helm"},"labels":{"app.kubernetes.io/instance":"sre-ci-test"},"name":"project-limits","namespace":"sre-ci-test"},"spec":{"limits":[{"max":{"cpu":"14","memory":"14Gi"},"type":"Pod"},{"default":{"cpu":"100m","ephemeral-storage":"20Gi","memory":"64Mi"},"defaultRequest":{"cpu":"100m","ephemeral-storage":"1Gi","memory":"64Mi"},"max":{"cpu":"14","memory":"14Gi"},"type":"Container"}]}}
creationTimestamp: "2021-12-14T12:18:46Z"
labels:
app.kubernetes.io/instance: sre-ci-test
managedFields:
- apiVersion: v1
fieldsType: FieldsV1
fieldsV1:
f:metadata:
f:annotations:
.: {}
f:app.kubernetes.io/created-by: {}
f:app.kubernetes.io/managed-by: {}
f:kubectl.kubernetes.io/last-applied-configuration: {}
f:labels:
.: {}
f:app.kubernetes.io/instance: {}
f:spec:
f:limits: {}
manager: argocd-application-controller
operation: Update
time: "2021-12-20T21:27:54Z"
name: project-limits
namespace: sre-ci-test
resourceVersion: "1466633760"
uid: 117ae94b-9228-41fb-9204-1adb4d3c1a9a
spec:
limits:
- max:
cpu: "14"
memory: 14Gi
type: Pod
- default:
cpu: 100m
ephemeral-storage: 20Gi
memory: 64Mi
defaultRequest:
cpu: 100m
ephemeral-storage: 1Gi
memory: 64Mi
max:
cpu: "14"
memory: 14Gi
type: Container
Default requests and limits for CPU and memory are the same. However, many containers are not set like that and that is triggering the QoS of the pods to Burstable. Example:
limits: cpu: 100m ephemeral-storage: 20Gi memory: 1Gi requests: cpu: 50m ephemeral-storage: 512Mi memory: 1Gi [...] qosClass: Burstable
Pod name used for the example: sre-casa-0fd04ea-lane-0-82v65-pod-dzbxr, which will be attached privately.
We would like to set the QoS of all the task pods to Guaranteed, but that cannot be achieved by setting LimitRanges. Is there any workaround?
Edit: this has been moved to "Feature" type because it is a feature available for Tekton 0.38 and later versions.
- is documented by
-
-
- Open
-
- is related to
-
RFE-3181 To enable a configuration whereby all the task pods stay in "Pending" status when not enough hardware resources are available in the cluster
-
- Accepted
-
- links to
