Jetlag CI health check (#59579)

* Jetlag CI health check Signed-off-by: Raul Sevilla <[email protected]> * Assign to build06 Signed-off-by: Raul Sevilla <[email protected]> * Metadata Signed-off-by: Raul Sevilla <[email protected]> --------- Signed-off-by: Raul Sevilla <[email protected]>
openshift · Dec 5, 2024 · aeba08e · aeba08e
1 parent 6b521e3
commit aeba08e
Show file tree

Hide file tree

Showing 8 changed files with 116 additions and 12 deletions.
diff --git a/ci-operator/config/redhat-performance/jetlag/redhat-performance-jetlag-main.yaml b/ci-operator/config/redhat-performance/jetlag/redhat-performance-jetlag-main.yaml
@@ -19,7 +19,7 @@ resources:
 tests:
 - always_run: false
   as: deploy-5nodes
-  cluster: build05
+  cluster: build06
   restrict_network_access: false
   steps:
     cluster_profile: metal-perfscale-jetlag
@@ -34,9 +34,10 @@ tests:
     test:
     - chain: openshift-qe-installer-bm-ping
     - chain: openshift-qe-installer-bm-deploy
+    - chain: openshift-qe-cluster-health
 - always_run: false
   as: deploy-5nodes-dev
-  cluster: build05
+  cluster: build06
   restrict_network_access: false
   steps:
     cluster_profile: metal-perfscale-jetlag
@@ -52,9 +53,10 @@ tests:
     test:
     - chain: openshift-qe-installer-bm-ping
     - chain: openshift-qe-installer-bm-deploy
+    - chain: openshift-qe-cluster-health
 - always_run: false
   as: deploy-sno
-  cluster: build05
+  cluster: build06
   restrict_network_access: false
   steps:
     cluster_profile: metal-perfscale-jetlag
@@ -66,9 +68,10 @@ tests:
     test:
     - chain: openshift-qe-installer-bm-ping
     - chain: openshift-qe-installer-bm-deploy
+    - chain: openshift-qe-cluster-health
 - always_run: false
   as: deploy-sno-dev
-  cluster: build05
+  cluster: build06
   restrict_network_access: false
   steps:
     cluster_profile: metal-perfscale-jetlag
@@ -81,6 +84,7 @@ tests:
     test:
     - chain: openshift-qe-installer-bm-ping
     - chain: openshift-qe-installer-bm-deploy
+    - chain: openshift-qe-cluster-health
 zz_generated_metadata:
   branch: main
   org: redhat-performance

diff --git a/ci-operator/jobs/redhat-performance/jetlag/redhat-performance-jetlag-main-presubmits.yaml b/ci-operator/jobs/redhat-performance/jetlag/redhat-performance-jetlag-main-presubmits.yaml
@@ -5,15 +5,15 @@ presubmits:
     branches:
     - ^main$
     - ^main-
-    cluster: build05
+    cluster: build06
     context: ci/prow/deploy-5nodes
     decorate: true
     decoration_config:
       skip_cloning: true
     labels:
       ci-operator.openshift.io/cloud: metal-perfscale-jetlag
       ci-operator.openshift.io/cloud-cluster-profile: metal-perfscale-jetlag
-      ci-operator.openshift.io/cluster: build05
+      ci-operator.openshift.io/cluster: build06
       ci.openshift.io/generator: prowgen
       job-release: "4.17"
       pj-rehearse.openshift.io/can-be-rehearsed: "true"
@@ -81,15 +81,15 @@ presubmits:
     branches:
     - ^main$
     - ^main-
-    cluster: build05
+    cluster: build06
     context: ci/prow/deploy-5nodes-dev
     decorate: true
     decoration_config:
       skip_cloning: true
     labels:
       ci-operator.openshift.io/cloud: metal-perfscale-jetlag
       ci-operator.openshift.io/cloud-cluster-profile: metal-perfscale-jetlag
-      ci-operator.openshift.io/cluster: build05
+      ci-operator.openshift.io/cluster: build06
       ci.openshift.io/generator: prowgen
       job-release: "4.17"
       pj-rehearse.openshift.io/can-be-rehearsed: "true"
@@ -157,15 +157,15 @@ presubmits:
     branches:
     - ^main$
     - ^main-
-    cluster: build05
+    cluster: build06
     context: ci/prow/deploy-sno
     decorate: true
     decoration_config:
       skip_cloning: true
     labels:
       ci-operator.openshift.io/cloud: metal-perfscale-jetlag
       ci-operator.openshift.io/cloud-cluster-profile: metal-perfscale-jetlag
-      ci-operator.openshift.io/cluster: build05
+      ci-operator.openshift.io/cluster: build06
       ci.openshift.io/generator: prowgen
       job-release: "4.17"
       pj-rehearse.openshift.io/can-be-rehearsed: "true"
@@ -233,15 +233,15 @@ presubmits:
     branches:
     - ^main$
     - ^main-
-    cluster: build05
+    cluster: build06
     context: ci/prow/deploy-sno-dev
     decorate: true
     decoration_config:
       skip_cloning: true
     labels:
       ci-operator.openshift.io/cloud: metal-perfscale-jetlag
       ci-operator.openshift.io/cloud-cluster-profile: metal-perfscale-jetlag
-      ci-operator.openshift.io/cluster: build05
+      ci-operator.openshift.io/cluster: build06
       ci.openshift.io/generator: prowgen
       job-release: "4.17"
       pj-rehearse.openshift.io/can-be-rehearsed: "true"

diff --git a/ci-operator/step-registry/openshift-qe/cluster-health/OWNERS b/ci-operator/step-registry/openshift-qe/cluster-health/OWNERS
@@ -0,0 +1 @@
+../OWNERS
diff --git a/...step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-chain.metadata.json b/...step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-chain.metadata.json
@@ -0,0 +1,33 @@
+{
+	"path": "openshift-qe/cluster-health/openshift-qe-cluster-health-chain.yaml",
+	"owners": {
+		"approvers": [
+			"afcollins",
+			"josecastillolema",
+			"jtaleric",
+			"mffiedler",
+			"paigerube14",
+			"qiliredhat",
+			"rsevilla87",
+			"svetsa-rh"
+		],
+		"reviewers": [
+			"jogeo",
+			"memodi",
+			"mffiedler",
+			"paigerube14",
+			"rpattath",
+			"skordas",
+			"jtaleric",
+			"svetsa-rh",
+			"shahsahil264",
+			"liqcui",
+			"krishvoor",
+			"afcollins",
+			"rsevilla87",
+			"vishnuchalla",
+			"sachinninganure",
+			"chaitanyaenr"
+		]
+	}
+}
diff --git a/ci-operator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-chain.yaml b/ci-operator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-chain.yaml
@@ -0,0 +1,6 @@
+chain:
+  as: openshift-qe-cluster-health
+  steps:
+  - ref: openshift-qe-cluster-health
+  documentation: |-
+    This chain uses oc adm wait-for-stable-cluster to run a cluster-health check
diff --git a/...perator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-commands.sh b/...perator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-commands.sh
@@ -0,0 +1,4 @@
+#!/bin/bash
+set -eu
+
+oc adm wait-for-stable-cluster --minimum-stable-period=${MINIMUM_STABLE_PERIOD} --timeout=${TIMEOUT}
diff --git a/...r/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-ref.metadata.json b/...r/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-ref.metadata.json
@@ -0,0 +1,33 @@
+{
+	"path": "openshift-qe/cluster-health/openshift-qe-cluster-health-ref.yaml",
+	"owners": {
+		"approvers": [
+			"afcollins",
+			"josecastillolema",
+			"jtaleric",
+			"mffiedler",
+			"paigerube14",
+			"qiliredhat",
+			"rsevilla87",
+			"svetsa-rh"
+		],
+		"reviewers": [
+			"jogeo",
+			"memodi",
+			"mffiedler",
+			"paigerube14",
+			"rpattath",
+			"skordas",
+			"jtaleric",
+			"svetsa-rh",
+			"shahsahil264",
+			"liqcui",
+			"krishvoor",
+			"afcollins",
+			"rsevilla87",
+			"vishnuchalla",
+			"sachinninganure",
+			"chaitanyaenr"
+		]
+	}
+}
diff --git a/ci-operator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-ref.yaml b/ci-operator/step-registry/openshift-qe/cluster-health/openshift-qe-cluster-health-ref.yaml
@@ -0,0 +1,23 @@
+ref:
+  as: openshift-qe-cluster-health
+  from_image:
+    namespace: ci
+    name: ocp-qe-perfscale-ci
+    tag: latest
+  cli: latest
+  commands: openshift-qe-cluster-health-commands.sh
+  env:
+  - name: MINIMUM_STABLE_PERIOD
+    default: "2m"
+    documentation: |-
+      Minimum duration to consider a cluster stable
+  - name: TIMEOUT
+    default: "5m"
+    documentation: |-
+      Duration before the command times out
+  resources:
+    requests:
+      cpu: 100m
+      memory: 100Mi
+  documentation: >-
+    This chain uses oc adm wait-for-stable-cluster to run a cluster-health check