[bitnami/etcd] fix save-snapshot script

bitnami/etcd/Chart.yaml

@@ -1,6 +1,6 @@
 apiVersion: v1
 name: etcd
-version: 4.4.14
+version: 4.5.0
 appVersion: 3.4.3
 description: etcd is a distributed key value store that provides a reliable way to store data across a cluster of machines
 keywords:

bitnami/etcd/README.md

@@ -130,6 +130,7 @@ The following tables lists the configurable parameters of the etcd chart and the
 | `startFromSnapshot.existingClaim`         | PVC containing the existing snapshot                                                                                                                      | `nil`                                                       |
 | `startFromSnapshot.snapshotFilename`      | Snapshot filename                                                                                                                                         | `nil`                                                       |
 | `disasterRecovery.enabled`                | Enable auto disaster recovery by periodically snapshotting the keyspace                                                                                   | `false`                                                     |
+| `disasterRecovery.debug`                  | Enable debug logging for snapshots                                                                                                                        | `false`                                                     |
 | `disasterRecovery.cronjob.schedule`       | Schedule in Cron format to save snapshots                                                                                                                 | `*/30 * * * *`                                              |
 | `disasterRecovery.cronjob.historyLimit`   | Number of successful finished jobs to retain                                                                                                              | `1`                                                         |
 | `disasterRecovery.cronjob.podAnnotations` | Pod annotations for cronjob pods                                                                                                                          | `{}`                                                        |

bitnami/etcd/templates/cronjob.yaml

@@ -31,6 +31,8 @@ spec:
               command:
                 - /scripts/save-snapshot.sh
               env:
+              - name: BITNAMI_SNAPSHOT_DEBUG
+                value: {{ ternary "true" "false" .Values.disasterRecovery.debug | quote }}
               - name: ETCDCTL_API
                 value: "3"
               {{- if .Values.auth.client.secureTransport }}

bitnami/etcd/templates/scripts-configmap.yaml

@@ -245,8 +245,9 @@ data:
     exec 3>&1
     exec 4>&2
 
-    if [ "${BITNAMI_DEBUG}" = true ]; then
+    if [ "${BITNAMI_SNAPSHOT_DEBUG}" = true ]; then
       echo "==> Bash debug is on"
+      set -x
     else
       echo "==> Bash debug is off"
       exec 1>/dev/null
@@ -257,14 +258,24 @@ data:
     AUTH_OPTIONS="{{ $etcdAuthOptions }}"
     ETCDCTL_ENDPOINTS="{{range $i, $e := until $replicaCount }}{{ $etcdClientProtocol }}://{{ $etcdFullname }}-{{ $e }}.{{ $etcdHeadlessServiceName }}.{{ $releaseNamespace }}.svc.{{ $clusterDomain }}:{{ $clientPort }},{{ end }}"
     # Remove the last comma "," introduced in the string
-    export ETCDCTL_ENDPOINTS="$(sed 's/,/ /g' <<< $ETCDCTL_ENDPOINTS | awk '{$1=$1};1' | sed 's/ /,/g')"
+    export ETCDCTL_ENDPOINTS="${ETCDCTL_ENDPOINTS%%,}"
 
     mkdir -p "/snapshots" 1>&3 2>&4
-    if etcdctl $AUTH_OPTIONS endpoint health; then
-        echo "Snapshotting the keyspace..." 1>&3 2>&4
-        etcdctl $AUTH_OPTIONS snapshot save "/snapshots/db" 1>&3 2>&4
-    else
-        echo "Cluster isn't responding!!" 1>&3 2>&4
-        exit 1
-    fi
+
+    # walk over each node in the list, check if node healthy make snapshot and exit, otherwise try another node
+    while read -d, ETCDCTL_ENDPOINTS; do
+        echo "Using endpoint ${ETCDCTL_ENDPOINTS}" 1>&3 2>&4
+        if etcdctl $AUTH_OPTIONS endpoint health; then
+            echo "Snapshotting the keyspace..." 1>&3 2>&4
+            etcdctl $AUTH_OPTIONS snapshot save "/snapshots/db" 1>&3 2>&4
+            exit 0
+        else
+            echo "Warning - etcd endpoint ${ETCDCTL_ENDPOINTS} not healthy" 1>&3 2>&4
+            echo "Trying another endpoint." 1>&3 2>&4
+        fi
+    done <<< "$ETCDCTL_ENDPOINTS"
+
+    # exit with error if all endpoints are bad
+    echo "Error - all etcd endpoints are unhealthy!" 1>&3 2>&4
+    exit 1
 {{- end }}

bitnami/etcd/values-production.yaml

@@ -320,6 +320,8 @@ startFromSnapshot:
 ##
 disasterRecovery:
   enabled: false
+  ## Set to true if you would like to see extra information on logs for snapshotting
+  debug: false
   cronjob:
     ## Schedule in Cron format to save snapshots
     ## See https://en.wikipedia.org/wiki/Cron

bitnami/etcd/values.yaml

@@ -320,6 +320,8 @@ startFromSnapshot:
 ##
 disasterRecovery:
   enabled: false
+  ## Set to true if you would like to see extra information on logs for snapshotting
+  debug: true
   cronjob:
     ## Schedule in Cron format to save snapshots
     ## See https://en.wikipedia.org/wiki/Cron