Refactor restore scripts to make them more robust and create events

Signed-off-by: Nicolas Bigler <[email protected]>
vshn · Jul 26, 2023 · 7aa39eb · 7aa39eb
1 parent 1de69dc
commit 7aa39eb
Show file tree

Hide file tree

Showing 3 changed files with 61 additions and 8 deletions.
diff --git a/pkg/comp-functions/functions/vshnredis/restore.go b/pkg/comp-functions/functions/vshnredis/restore.go
@@ -105,7 +105,7 @@ func addPrepareRestoreJob(ctx context.Context, comp *vshnv1.VSHNRedis, iof *runt
 									Value: comp.ObjectMeta.Labels[claimNamespaceLabel],
 								},
 								{
-									Name:  "CLAIM_NAME",
+									Name:  "SOURCE_CLAIM_NAME",
 									Value: comp.Spec.Parameters.Restore.ClaimName,
 								},
 								{
@@ -247,6 +247,8 @@ func addCleanUpJob(ctx context.Context, comp *vshnv1.VSHNRedis, iof *runtime.Run
 	cleanupRestoreJobName := truncateObjectName(comp.Name + "-" + comp.Spec.Parameters.Restore.BackupName + "-cleanup-job")
 	restoreJobName := truncateObjectName(comp.Name + "-" + comp.Spec.Parameters.Restore.BackupName + "-restore-job")
 	restoreSecret := "statefulset-replicas-" + comp.Spec.Parameters.Restore.ClaimName + "-" + comp.Spec.Parameters.Restore.BackupName
+	claimNamespaceLabel := "crossplane.io/claim-namespace"
+	claimNameLabel := "crossplane.io/claim-name"
 
 	prepJob := &batchv1.Job{
 		ObjectMeta: metav1.ObjectMeta{
@@ -269,9 +271,17 @@ func addCleanUpJob(ctx context.Context, comp *vshnv1.VSHNRedis, iof *runtime.Run
 							Args: []string{cleanupRestoreScript},
 							Env: []corev1.EnvVar{
 								{
-									Name:  "CLAIM_NAME",
+									Name:  "CLAIM_NAMESPACE",
+									Value: comp.ObjectMeta.Labels[claimNamespaceLabel],
+								},
+								{
+									Name:  "SOURCE_CLAIM_NAME",
 									Value: comp.Spec.Parameters.Restore.ClaimName,
 								},
+								{
+									Name:  "DEST_CLAIM_NAME",
+									Value: comp.ObjectMeta.Labels[claimNameLabel],
+								},
 								{
 									Name:  "RESTORE_JOB_NAME",
 									Value: restoreJobName,

diff --git a/pkg/comp-functions/functions/vshnredis/script/cleanupRestore.sh b/pkg/comp-functions/functions/vshnredis/script/cleanupRestore.sh
@@ -4,18 +4,61 @@ set -euo pipefail
 
 echo "Wait for restore to complete"
 
-counter=0
-until [ $counter -eq 300 ] || [[ $(kubectl -n "${TARGET_NAMESPACE}" get job "${RESTORE_JOB_NAME}" -o jsonpath='{.status.succeeded}' 2> /dev/null) -eq 1 ]];
+until [[ $(kubectl -n "${TARGET_NAMESPACE}" get job "${RESTORE_JOB_NAME}" -o jsonpath='{.status.succeeded}' 2> /dev/null) -eq 1 ]] || [[ $(kubectl -n "${TARGET_NAMESPACE}" get job "${RESTORE_JOB_NAME}" -o jsonpath='{.status.failed}' 2> /dev/null) -eq 1 ]];
 do
-  (( counter+=1 ))
   sleep 1
 done
 
+if [[ $(kubectl -n "${TARGET_NAMESPACE}" get job "${RESTORE_JOB_NAME}" -o jsonpath='{.status.failed}' 2> /dev/null) -eq 1 ]]; then
+  kubectl apply -f - <<EOF
+  apiVersion: v1
+  kind: Event
+  metadata:
+    name: "${DEST_CLAIM_NAME}-restore-failed"
+    namespace: "${CLAIM_NAMESPACE}"
+  type: Warning
+  firstTimestamp: $(date --utc +%FT%TZ)
+  lastTimestamp: $(date --utc +%FT%TZ)
+  message: "Restore of ${DEST_CLAIM_NAME} failed"
+  involvedObject:
+    apiVersion: vshn.appcat.vshn.io/v1
+    kind: VSHNRedis
+    name: "${DEST_CLAIM_NAME}"
+    namespace: "${CLAIM_NAMESPACE}"
+    uid: "$(kubectl -n ${CLAIM_NAMESPACE} get vshnredis.vshn.appcat.vshn.io ${DEST_CLAIM_NAME} -o jsonpath='{.metadata.uid}')"
+  reason: RestoreFailed
+  source:
+    component: "${TARGET_NAMESPACE}/${RESTORE_JOB_NAME}/job.batch/v1"
+EOF
+fi
+
+if [[ $(kubectl -n "${TARGET_NAMESPACE}" get job "${RESTORE_JOB_NAME}" -o jsonpath='{.status.succeeded}' 2> /dev/null) -eq 1 ]]; then
+  kubectl apply -f - <<EOF
+  apiVersion: v1
+  kind: Event
+  metadata:
+    name: "${DEST_CLAIM_NAME}-restore-completed"
+    namespace: "${CLAIM_NAMESPACE}"
+  type: Normal
+  firstTimestamp: $(date --utc +%FT%TZ)
+  lastTimestamp: $(date --utc +%FT%TZ)
+  message: "Restore of ${DEST_CLAIM_NAME} succeeded"
+  involvedObject:
+    apiVersion: vshn.appcat.vshn.io/v1
+    kind: VSHNRedis
+    name: "${DEST_CLAIM_NAME}"
+    namespace: "${CLAIM_NAMESPACE}"
+    uid: "$(kubectl -n ${CLAIM_NAMESPACE} get vshnredis.vshn.appcat.vshn.io ${DEST_CLAIM_NAME} -o jsonpath='{.metadata.uid}')"
+  reason: RestoreSucceeded
+  source:
+    component: "${TARGET_NAMESPACE}/${RESTORE_JOB_NAME}/job.batch/v1"
+EOF
+fi
 echo "scaling up redis"
 
 kubectl -n "${TARGET_NAMESPACE}" scale statefulset redis-master --replicas "${NUM_REPLICAS}"
 
 echo "cleanup secret"
 
 kubectl -n "${TARGET_NAMESPACE}" delete secret "restore-credentials-${BACKUP_NAME}"
-kubectl delete secret "statefulset-replicas-${CLAIM_NAME}-${BACKUP_NAME}"
+kubectl delete secret "statefulset-replicas-${SOURCE_CLAIM_NAME}-${BACKUP_NAME}"
diff --git a/pkg/comp-functions/functions/vshnredis/script/prepRestore.sh b/pkg/comp-functions/functions/vshnredis/script/prepRestore.sh
@@ -2,7 +2,7 @@
 
 set -euo pipefail
 
-source_namespace=$(kubectl -n "${CLAIM_NAMESPACE}" get vshnredis "${CLAIM_NAME}" -ojson | jq -r '.status.instanceNamespace')
+source_namespace=$(kubectl -n "${CLAIM_NAMESPACE}" get vshnredis "${SOURCE_CLAIM_NAME}" -ojson | jq -r '.status.instanceNamespace')
 
 echo "copy secret"
 
@@ -14,7 +14,7 @@ backup_path=$(kubectl -n "${source_namespace}" get snapshots.k8up.io "${BACKUP_N
 backup_name=$(kubectl -n "${source_namespace}" get snapshots.k8up.io "${BACKUP_NAME}" -o jsonpath='{.spec.id}')
 num_replicas=$(kubectl -n "${TARGET_NAMESPACE}" get statefulset redis-master -o jsonpath='{.spec.replicas}')
 kubectl -n "${TARGET_NAMESPACE}" create secret generic "restore-credentials-${BACKUP_NAME}" --from-literal AWS_ACCESS_KEY_ID="${access_key}" --from-literal AWS_SECRET_ACCESS_KEY="${secret_key}" --from-literal RESTIC_PASSWORD="${restic_password}" --from-literal RESTIC_REPOSITORY="${restic_repository}" --from-literal BACKUP_PATH="${backup_path}" --from-literal BACKUP_NAME="${backup_name}"
-kubectl create secret generic "statefulset-replicas-${CLAIM_NAME}-${BACKUP_NAME}" --from-literal NUM_REPLICAS="${num_replicas}"
+kubectl create secret generic "statefulset-replicas-${SOURCE_CLAIM_NAME}-${BACKUP_NAME}" --from-literal NUM_REPLICAS="${num_replicas}"
 echo "scaling down redis"
 
 until kubectl -n "${TARGET_NAMESPACE}" get statefulset redis-master > /dev/null 2>&1