Fix automation and change name of the datasource (#385)

* Fix automation and change name of the datasource Signed-off-by: Jakub Stejskal <[email protected]> * Revert branch for deployment-hub Signed-off-by: Jakub Stejskal <[email protected]> --------- Signed-off-by: Jakub Stejskal <[email protected]>
skodjob · Jan 13, 2025 · 134053a · 134053a
1 parent 856b2fd
commit 134053a
Show file tree

Hide file tree

Showing 27 changed files with 167 additions and 33 deletions.
diff --git a/.github/workflows/skodjob-ee.yaml b/.github/workflows/skodjob-ee.yaml
@@ -29,7 +29,7 @@ jobs:
         uses: docker/build-push-action@v6
         with:
           context: ./skodjob-ee/context
-          file: ./skodjob-ee/context/Dockerfile
+          file: ./skodjob-ee/context/Containerfile
           push: true
           # TODO - Change org when we will create a new one
           tags: quay.io/tealc/skodjob-ee:latest
diff --git a/image-update/strimzi-image-update.sh b/image-update/strimzi-image-update.sh
@@ -212,6 +212,9 @@ do
         LATEST_DIGEST=$(skopeo inspect --override-arch amd64 --override-os linux docker://"$TARGET_ORG_REPO"/"$IMAGE":latest-kafka-"$PREFIX"  --format "{{ .Digest }}")
     elif [[ $ELEMENT == *"kafka@"* ]]; then
         continue
+    # Do not care about bridge version
+    elif [[ $ELEMENT == *"bridge@"* ]]; then
+        continue
     else
         LATEST_DIGEST=$(skopeo inspect  --override-arch amd64 --override-os linux docker://"$TARGET_ORG_REPO"/"$IMAGE" --format "{{ .Digest }}")
     fi

diff --git a/...tion-hub/tasks/infra-setup/install/advanced-cluster-management/install_observability.yaml b/...tion-hub/tasks/infra-setup/install/advanced-cluster-management/install_observability.yaml
@@ -18,6 +18,16 @@
     template: templates/acm/06-thanos-storage-secret.yaml.j2
     verify_ssl: no
 
+# This is needed if the clusters are created via our ACM.
+# MultiClusterObservability is installed only on managed clusters with this label
+- name: Label each cluster with vendor=OpenShift
+  ansible.builtin.shell: "oc label managedcluster {{ cluster.name }} vendor=OpenShift --overwrite"
+  environment:
+    KUBECONFIG: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+  loop: "{{ clusters_dict.values() }}"
+  loop_control:
+    loop_var: cluster
+
 - name: "Create multicluster observability on Infra cluster"
   kubernetes.core.k8s:
     kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"

diff --git a/install/roles/automation-hub/tasks/infra-setup/install/install_argo_workflow.yaml b/install/roles/automation-hub/tasks/infra-setup/install/install_argo_workflow.yaml
@@ -12,6 +12,7 @@
         labels:
           project: "skodjob"
           secret: "cert-manager"
+          strimzi.io/sync-secrets: "true"
 
 - name: Clean helm releases for workflows
   shell: "oc delete secret -l owner=helm -n {{ infra_workflow_namespace }} || true"

diff --git a/install/roles/automation-hub/tasks/infra-setup/install/install_strimzi.yaml b/install/roles/automation-hub/tasks/infra-setup/install/install_strimzi.yaml
@@ -1,5 +1,5 @@
 ---
-- name: Create Strimzi namespaces on Infra cluster
+- name: Create Strimzi Operator namespaces on Infra cluster
   kubernetes.core.k8s:
     kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
     verify_ssl: no
@@ -11,12 +11,28 @@
         name: "{{ item }}"
         labels:
           project: "skodjob"
-          strimzi.io/sync-secrets: "true"
+          strimzi.io/type: operators
   loop:
     - "{{ strimzi_operator_namespace }}"
-    - "{{ strimzi_kafka_namespace }}"
     - "{{ strimzi_drain_cleaner_namespace }}"
 
+- name: Create Strimzi Event Bus namespaces on Infra cluster
+  kubernetes.core.k8s:
+    kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+    verify_ssl: no
+    state: present
+    definition:
+      apiVersion: v1
+      kind: Namespace
+      metadata:
+        name: "{{ item }}"
+        labels:
+          project: "skodjob"
+          strimzi.io/sync-secrets: "true"
+          strimzi.io/type: operands
+  loop:
+    - "{{ strimzi_kafka_namespace }}"
+
 - name: Clean helm releases for workflows
   shell: "oc delete secret -l owner=helm -n {{ strimzi_operator_namespace }} || true"
   environment:

diff --git a/install/roles/automation-hub/tasks/monitoring/deploy-data-source.yaml b/install/roles/automation-hub/tasks/monitoring/deploy-data-source.yaml
@@ -1,12 +1,22 @@
 ---
 # Get Grafana access token and update clusters data
-- name: "Export access token {{ cluster.name }}"
-  #  shell: "oc serviceaccounts get-token grafana-serviceaccount -n {{ target.sa_namespace }}"
-  shell: "oc get  $(oc get secret -o name -n {{ cluster.monitoring_namespace }} | grep {{ sa_name }}-token) -n {{ cluster.monitoring_namespace }} -o=jsonpath='{.data.token}' | base64 -d"
+- name: "Export access token for {{ cluster.name }}"
+  shell: "oc get secret {{ sa_name }} -n {{ cluster.monitoring_namespace }} -o=jsonpath='{.data.token}' | base64 -d"
   environment:
     KUBECONFIG: "{{ kubeconfig_path }}/{{ cluster.name }}"
   register: "access_token_out"
 
+- debug:
+    msg: "{{ access_token_out }}"
+
+- name: "Set facts grafana access token - {{ cluster.name }}"
+  set_fact: access_token="{{ access_token_out.stdout }}"
+
+- name: Check that token exists
+  fail:
+    msg: "Token is not properly stored!\n{{ access_token }}"
+  when: access_token | length == 0
+
 - name: "Update facts for {{ cluster.name }}"
   ansible.utils.update_fact:
     updates:
@@ -23,7 +33,7 @@
     clusters_dict: "{{ updated_data.clusters_dict }}"
   when: is_worker is defined
 
-- name: "Create Grafana data-source {{ cluster.name }}"
+- name: "Create Grafana data-source {{ datasource_name }}"
   kubernetes.core.k8s:
     kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
     namespace: "{{ cluster.monitoring_namespace }}"

diff --git a/install/roles/automation-hub/tasks/monitoring/install_monitoring.yaml b/install/roles/automation-hub/tasks/monitoring/install_monitoring.yaml
@@ -134,18 +134,21 @@
   vars:
     is_worker: True
     sa_name: "{{ cluster.name }}-grafana"
+    datasource_name: "{{ cluster.name }}"
     url: "https://thanos-querier-openshift-monitoring.apps.{{ cluster.name }}.{{ cluster_domain }}:443"
 
 - name: Create data-sources on Infra cluster for Infra data
   include_tasks: deploy-data-source.yaml
   loop:
     - {name: "{{ infra_context_name }}",
+       datasource_name: "{{ infra_context_name }}",
        sa_namespace: "{{ infra_user_namespace }}",
        monitoring_namespace: "{{ infra_monitoring_namespace }}",
        target_cluster_name: "{{ infra_context_name }}",
        context: "{{ infra_context_name }}",
        url: "https://thanos-querier.openshift-monitoring.svc.cluster.local:9091"}
-    - {name: "skodjob-thanos",
+    - {name: "{{ infra_context_name }}",
+       datasource_name: "skodjob-thanos",
        sa_namespace: "{{ infra_user_namespace }}",
        monitoring_namespace: "{{ infra_monitoring_namespace }}",
        target_cluster_name: "{{ infra_context_name }}",
@@ -154,7 +157,8 @@
   loop_control:
     loop_var: cluster
   vars:
-    sa_name: "skodjob-sa"
+    sa_name: "{{ infra_context_name }}-grafana"
+    datasource_name: "{{ cluster.datasource_name }}"
 
 - name: Configure Alertmanager to for sending alerts on Infra cluster from all clusters
   kubernetes.core.k8s:

diff --git a/install/roles/automation-hub/tasks/scenario-deployment/workflows/infra-argo-workflows.yaml b/install/roles/automation-hub/tasks/scenario-deployment/workflows/infra-argo-workflows.yaml
@@ -20,3 +20,79 @@
     verify_ssl: no
     apply: true
   with_fileglob: 'templates/argo-workflows/eventsource/*.j2'
+
+- name: Wait for EventSources - Deployed
+  kubernetes.core.k8s_info:
+    kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+    namespace: "{{ infra_workflow_namespace }}"
+    kind: EventSource
+    api_version: argoproj.io/v1alpha1
+    name: "{{ item }}"
+    wait: true
+    verify_ssl: no
+    wait_condition:
+      type: Deployed
+      status: True
+  loop:
+    - automation-hub
+  register: eventsource_info
+
+# TODO - Fail if single sensor is not found, this could be improved to properly check the status
+- name: Fail if EventSource is not found
+  fail:
+    msg: "EventSource automation-hub not found."
+  loop: "{{ eventsource_info.results }}"
+  when: item.resources | length == 0
+
+- name: Wait for EventSources - SourcesProvided
+  kubernetes.core.k8s_info:
+    kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+    namespace: "{{ infra_workflow_namespace }}"
+    kind: EventSource
+    api_version: argoproj.io/v1alpha1
+    name: "{{ item }}"
+    wait: true
+    verify_ssl: no
+    wait_condition:
+      type: SourcesProvided
+      status: True
+  loop:
+    - automation-hub
+
+- name: Wait for Sensors - Deployed
+  kubernetes.core.k8s_info:
+    kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+    namespace: "{{ infra_workflow_namespace }}"
+    kind: Sensor
+    api_version: argoproj.io/v1alpha1
+    name: "{{ item }}"
+    wait: true
+    verify_ssl: no
+    wait_condition:
+      type: Deployed
+      status: True
+  loop:
+    - automation-hub
+  register: sensor_info
+
+# TODO - Fail if single sensor is not found, this could be improved to properly check the status
+- name: Fail if Sensor is not found
+  fail:
+    msg: "Sensor automation-hub not found."
+  loop: "{{ sensor_info.results }}"
+  when: item.resources | length == 0
+
+- name: Wait for Sensors - TriggersProvided
+  kubernetes.core.k8s_info:
+    kubeconfig: "{{ kubeconfig_path }}/{{ infra_context_name }}"
+    namespace: "{{ infra_workflow_namespace }}"
+    kind: Sensor
+    api_version: argoproj.io/v1alpha1
+    name: "{{ item }}"
+    wait: true
+    verify_ssl: no
+    wait_condition:
+      type: TriggersProvided
+      status: True
+  loop:
+    - automation-hub
diff --git a/install/roles/automation-hub/templates/acm/07-multicluster-observability.yaml.j2 b/install/roles/automation-hub/templates/acm/07-multicluster-observability.yaml.j2
@@ -24,8 +24,9 @@ spec:
     receiveStorageSize: 100Gi
     ruleStorageSize: 10Gi
     storageClass: standard-csi
+    # This might need increase in the future
     storeStorageSize: 10Gi
-  # For more info see https://access.redhat.com/documentation/en-us/red_hat_advanced_cluster_management_for_kubernetes/2.7/html-single/apis/index
+  # For more info see https://access.redhat.com/documentation/en-us/red_hat_advanced_cluster_management_for_kubernetes/2.12/html-single/apis/index
   advanced:
     receive:
       replicas: 3

diff --git a/install/roles/automation-hub/templates/argo/applications/amq-streams-infra.yaml b/install/roles/automation-hub/templates/argo/applications/amq-streams-infra.yaml
@@ -3,6 +3,8 @@ kind: ApplicationSet
 metadata:
   name: amq-streams-infra
 spec:
+  syncPolicy:
+    preserveResourcesOnDeletion: true
   generators:
     - matrix:
         generators:
@@ -36,7 +38,6 @@ spec:
           selfHeal: true
         syncOptions:
           - CreateNamespace=true
-        preserveResourcesOnDeletion: true
         managedNamespaceMetadata:
           labels:
             project: skodjob

diff --git a/install/roles/automation-hub/templates/argo/applications/debezium.yaml b/install/roles/automation-hub/templates/argo/applications/debezium.yaml
@@ -3,6 +3,8 @@ kind: ApplicationSet
 metadata:
   name: debezium
 spec:
+  syncPolicy:
+    preserveResourcesOnDeletion: true
   generators:
     - matrix:
         generators:
@@ -36,7 +38,6 @@ spec:
           selfHeal: true
         syncOptions:
           - CreateNamespace=true
-        preserveResourcesOnDeletion: true
         managedNamespaceMetadata:
           labels:
             project: skodjob

diff --git a/...es/automation-hub/templates/argo/applications/monitoring/alerts/infra/argo-alerts.yaml.j2 b/...es/automation-hub/templates/argo/applications/monitoring/alerts/infra/argo-alerts.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/argo
     directory:
       recurse: true

diff --git a/...go/applications/monitoring/alerts/worker-01/strimzi-alerts-cluster-operator-kraft.yaml.j2 b/...go/applications/monitoring/alerts/worker-01/strimzi-alerts-cluster-operator-kraft.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/strimzi-operator
     directory:
       recurse: true

diff --git a/...tes/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-cluster-operator.yaml.j2 b/...tes/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-cluster-operator.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/strimzi-operator
     directory:
       recurse: true

diff --git a/...plates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-drain-cleaner.yaml.j2 b/...plates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-drain-cleaner.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/strimzi-drain-cleaner
     directory:
       recurse: true

diff --git a/...plates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-connect.yaml.j2 b/...plates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-connect.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka-connect
     directory:
       recurse: true

diff --git a/...emplates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-kraft.yaml.j2 b/...emplates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-kraft.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka-kraft
     directory:
       recurse: true

diff --git a/...ates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-mm2-kraft.yaml.j2 b/...ates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-mm2-kraft.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka-mirror-maker
     directory:
       recurse: true

diff --git a/.../templates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-mm2.yaml.j2 b/.../templates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka-mm2.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka-mirror-maker
     directory:
       recurse: true

diff --git a/...-hub/templates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka.yaml.j2 b/...-hub/templates/argo/applications/monitoring/alerts/worker-01/strimzi-alerts-kafka.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka
     directory:
       recurse: true

diff --git a/...tes/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-cluster-operator.yaml.j2 b/...tes/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-cluster-operator.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/strimzi-operator
     directory:
       recurse: true

diff --git a/...plates/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-drain-cleaner.yaml.j2 b/...plates/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-drain-cleaner.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/strimzi-drain-cleaner
     directory:
       recurse: true

diff --git a/...-hub/templates/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-kafka.yaml.j2 b/...-hub/templates/argo/applications/monitoring/alerts/worker-02/strimzi-alerts-kafka.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/alerts/kafka
     directory:
       recurse: true

diff --git a/...tomation-hub/templates/argo/applications/monitoring/dashboards/grafana-dashboards.yaml.j2 b/...tomation-hub/templates/argo/applications/monitoring/dashboards/grafana-dashboards.yaml.j2
@@ -9,7 +9,7 @@ spec:
   project: monitoring
   source:
     repoURL: https://github.com/skodjob/deployment-hub
-    targetRevision: HEAD
+    targetRevision: {{ deployment_hub_branch }}
     path: metrics/grafana/dashboards
     directory:
       recurse: true