Remove healthcheck sidecar from etcd

Also make a switch from auxiliary to permanent cluster.

Signed-off-by: Ruslan Aliev <raliev@mirantis.com>
Change-Id: I7918072a6ba5a6b22b359d1616def8c31425462d

charts/etcd/templates/bin/_etcdctl_anchor.tpl

@@ -61,6 +61,7 @@ cleanup_host () {
 firstrun=true
 saddness_duration=0
 while true; do
+    date
     # TODO(mark-burnett) Need to monitor a file(s) when shutting down/starting
     # up so I don't try to take two actions on the node at once.
     {{- if .Values.bootstrapping.enabled  }}
@@ -117,7 +118,10 @@ while true; do
         fi
         echo Successfully added $HOSTNAME to cluster members.
         # Refresh member list so we start with the right configuration.
-        etcdctl member list > /tmp/members
+        if ! etcdctl member list > /tmp/members; then
+          echo Could not get a member list, trying again.
+          continue
+        fi
     elif grep $PEER_ENDPOINT /tmp/members | grep '\bunstarted\b'; then
         # This member is in the cluster but not started
         if [ $saddness_duration -ge {{ .Values.anchor.saddness_threshold }} ]

charts/etcd/templates/etc/_kubernetes-etcd.yaml.tpl

@@ -16,22 +16,12 @@
 
 {{- $envAll := . }}
 {{- define "etcdreadinessProbeTemplate" }}
-exec:
-  command:
-    - /bin/sh
-    - -c
-    - |-
-      etcdctl endpoint health
-      exit $?
+grpc:
+  port: {{ .Values.network.service_client.port }}
 {{- end }}
 {{- define "etcdlivenessProbeTemplate" }}
-exec:
-  command:
-    - /bin/sh
-    - -c
-    - |-
-      etcdctl endpoint status
-      exit $?
+grpc:
+  port: {{ .Values.network.service_client.port }}
 {{- end }}
 # Strip off "etcd" from service name to get the application name
 # Note that application can either be kubernetes or calico for now
@@ -123,34 +113,11 @@ spec:
         - name: MANIFEST_PATH
           value: /manifests/{{ .Values.service.name }}.yaml
 {{ include "helm-toolkit.utils.to_k8s_env_vars" .Values.pod.env.etcd | indent 8 }}
-      volumeMounts:
-        - name: data
-          mountPath: /var/lib/etcd
-        - name: etc
-          mountPath: /etc/etcd
-    - name: etcd-health-check
-      image: {{ .Values.images.tags.etcdctl }}
-      imagePullPolicy: {{ .Values.images.pull_policy }}
-{{ tuple $envAll $envAll.Values.pod.resources.etcd_pod_sidecar | include "helm-toolkit.snippets.kubernetes_resources" | indent 6 }}
-{{ dict "envAll" $envAll "application" "etcd" "container" "etcd" | include "helm-toolkit.snippets.kubernetes_container_security_context" | indent 6 }}
-      env:
-        - name: ETCDCTL_API
-          value: "{{ .Values.etcd.etcdctl_api }}"
-        - name: ETCDCTL_DIAL_TIMEOUT
-          value: "3s"
-        - name: ETCDCTL_ENDPOINTS
-          value: "https://127.0.0.1:{{ .Values.network.service_client.target_port }}"
-        - name: ETCDCTL_CACERT
-          value: "/etc/etcd/tls/client-ca.pem"
-        - name: ETCDCTL_CERT
-          value: "/etc/etcd/tls/etcd-client.pem"
-        - name: ETCDCTL_KEY
-          value: "/etc/etcd/tls/etcd-client-key.pem"
-      command: ["/bin/sh", "-c", "--"]
-      args: ["while true; do sleep 30; done;"]
 {{ dict "envAll" $envAll "component" "etcd" "container" "etcd" "type" "readiness" "probeTemplate" (include "etcdreadinessProbeTemplate" $envAll | fromYaml) | include "helm-toolkit.snippets.kubernetes_probe" | indent 6 }}
 {{ dict "envAll" $envAll "component" "etcd" "container" "etcd" "type" "liveness" "probeTemplate" (include "etcdlivenessProbeTemplate" $envAll | fromYaml) | include "helm-toolkit.snippets.kubernetes_probe" | indent 6 }}
       volumeMounts:
+        - name: data
+          mountPath: /var/lib/etcd
         - name: etc
           mountPath: /etc/etcd
   volumes:

charts/etcd/values.yaml

@@ -261,13 +261,6 @@ pod:
       limits:
         memory: "1024Mi"
         cpu: "2000m"
-    etcd_pod_sidecar:
-      requests:
-        memory: "128Mi"
-        cpu: "100m"
-      limits:
-        memory: "1024Mi"
-        cpu: "2000m"
     test:
       limits:
         memory: "128Mi"

promenade/templates/include/genesis-etcd/server-container.yaml

@@ -55,29 +55,45 @@
           value: $(ETCD_KEY_FILE)
 {%- if config['Genesis:etcd.heartbeat_interval'] is defined %}
         - name: ETCD_HEARTBEAT_INTERVAL
-          value: {{ config['Genesis:etcd.heartbeat_interval'] }}
+          value: "{{ config['Genesis:etcd.heartbeat_interval'] }}"
 {%- endif %}
 {%- if config['Genesis:etcd.election_timeout'] is defined %}
         - name: ETCD_ELECTION_TIMEOUT
-          value: {{ config['Genesis:etcd.election_timeout'] }}
+          value: "{{ config['Genesis:etcd.election_timeout'] }}"
 {%- endif %}
 {%- if config['Genesis:etcd.snapshot_count'] is defined %}
         - name: ETCD_SNAPSHOT_COUNT
-          value: {{ config['Genesis:etcd.snapshot_count'] }}
+          value: "{{ config['Genesis:etcd.snapshot_count'] }}"
 {%- endif %}
 {%- if config['Genesis:etcd.cipher_suites'] is defined %}
         - name: ETCD_CIPHER_SUITES
-          value: {{ config['Genesis:etcd.cipher_suites'] }}
+          value: "{{ config['Genesis:etcd.cipher_suites'] }}"
 {%- endif %}
 {%- if config['Genesis:etcd.gomaxprocs'] is defined %}
         - name: GOMAXPROCS
-          value: {{ config['Genesis:etcd.gomaxprocs'] }}
+          value: "{{ config['Genesis:etcd.gomaxprocs'] }}"
 {%- endif %}
       ports:
         - name: client
           containerPort: {{ client_port }}
         - name: peer
           containerPort: {{ peer_port }}
+      livenessProbe:
+        grpc:
+          port: {{ client_port }}
+        initialDelaySeconds: 15
+        timeoutSeconds: 10
+        periodSeconds: 30
+        successThreshold: 1
+        failureThreshold: 10
+      readinessProbe:
+        grpc:
+          port: {{ client_port }}
+        initialDelaySeconds: 15
+        timeoutSeconds: 10
+        periodSeconds: 30
+        successThreshold: 1
+        failureThreshold: 10
       volumeMounts:
         - name: data-{{ etcd_name }}
           mountPath: /var/lib/etcd

promenade/templates/roles/genesis/etc/kubernetes/manifests/auxiliary-kubernetes-etcd.yaml

@@ -38,14 +38,56 @@ spec:
               MEMBER_NAME=$1
               MEMBER_ID=$(etcdctl member list | grep "${MEMBER_NAME}" | awk -F ', ' '{ print $1 }')
               if [ -n "${MEMBER_ID}" ]; then
+                  if is_leader "$MEMBER_ID"; then
+                    abdicate
+                  fi
                   etcdctl member remove $MEMBER_ID
+                  sleep 10
               fi
           }
 
+          abdicate () {
+            NEW_LEADER=$(etcdctl member list | grep '\bstarted\b' | grep -Ev '\\b(auxiliary-0|auxiliary-1)\\b' | head -1 | awk -F ', ' '{print $1}')
+            if [ -n "$NEW_LEADER" ]; then
+              if ! etcdctl move-leader "$NEW_LEADER"; then
+                echo "Attempted abdication, but failed."
+                return
+              fi
+              sleep 5
+              return
+            fi
+          }
+
+          abdicate_if_needed () {
+            AUX_MEMBERS=$(etcdctl member list | grep '\bstarted\b' | grep -E "\\b(auxiliary-0|auxiliary-1)\\b" | awk -F ', ' '{print $1}')
+            for m in $AUX_MEMBERS; do
+              if is_leader "$m"; then
+                abdicate
+              fi
+            done
+          }
+
+          is_leader () {
+            MEMBER_ID="$1"
+            MEMBER_EP=$(etcdctl member list | grep "$MEMBER_ID" | awk -F ', ' '{print $5}')
+            IS_LEADER=$(ETCDCTL_ENDPOINTS="$MEMBER_EP" etcdctl endpoint status | awk -F ', ' '{ print $5 }')
+            if [ "$IS_LEADER" = "true" ]; then
+              return 0
+            else
+              return 1
+            fi
+          }
+
+          aux_endpoint_present () {
+            awk '/- name: ETCD_ENDPOINTS/{getline; sub(/.*: "/, ""); sub(/".*/, ""); print}' /manifests/kubernetes-apiserver.yaml \
+                          | grep -E "\\b(12379|22379)\\b"
+          }
+
           auxiliary_threshold="{{ config.get_first('Genesis:etcd.auxiliary_threshold', default=3) }}"
           # NOTE(sh8121att): If there are enough (a fully resilient contigent) non-auxiliary members,
           # then we are ready to remove the auxiliary members.  Otherwise, wait.
-          while [ ! "$(external_member_count)" -ge "$auxiliary_threshold" ]; do
+          while [ ! "$(external_member_count)" -ge "$auxiliary_threshold" ] && [ "$(aux_endpoint_present)" ]; do
+              abdicate_if_needed
               sleep 30
           done