helm: add support for PMM

docker/k8s/pmm-client/Dockerfile

@@ -0,0 +1,17 @@
+FROM vitess/base AS base
+
+FROM debian:stretch-slim
+
+# Copy CA certs for https calls
+COPY --from=base /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
+
+RUN apt-get update && \
+    apt-get upgrade -qq && \
+    apt-get install wget -qq --no-install-recommends && \
+    wget https://www.percona.com/redir/downloads/pmm-client/1.6.1/binary/debian/stretch/x86_64/pmm-client_1.6.1-1.stretch_amd64.deb && \
+    dpkg -i pmm-client_1.6.1-1.stretch_amd64.deb && \
+    rm pmm-client_1.6.1-1.stretch_amd64.deb && \
+    apt-get purge wget -qq && \
+    apt-get autoremove -qq && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*

helm/vitess/README.md

@@ -217,3 +217,36 @@ vttablet:
       requests:
         storage: "100Gi"
 ```
+
+### Enable PMM (Percona Monitoring and Management)
+
+```
+topology:
+  cells:
+    ...
+
+pmm:
+  enabled: true
+  pmmTag: "1.6.1"
+  client:
+    resources:
+      requests:
+        cpu: 50m
+        memory: 128Mi
+      limits:
+        cpu: 200m
+        memory: 256Mi
+  server:
+    resources:
+      limits:
+        cpu: 2
+        memory: 4Gi
+    dataVolumeClaimSpec:
+      storageClassName: "default"
+      accessModes: ["ReadWriteOnce"]
+      resources:
+        requests:
+          storage: "150Gi"
+    env:
+      metricsMemory: "3000000"
+```

helm/vitess/templates/NOTES.txt

@@ -10,4 +10,4 @@ Then use the following URLs:
 
 vtctld: {{$proxyURL}}/services/vtctld:web/app/
 vtgate: {{$proxyURL}}/services/vtgate-{{$cell}}:web/
-
+{{ if $.Values.pmm.enabled }}   pmm: {{$proxyURL}}/services/pmm:web/{{ end }}

helm/vitess/templates/_helpers.tpl

@@ -57,7 +57,6 @@ x{{$replaced_label}}
 securityContext:
   runAsUser: 1000
   fsGroup: 2000
-  runAsNonRoot: true
 {{- end -}}
 
 #############################

helm/vitess/templates/_pmm.tpl

@@ -0,0 +1,186 @@
+###################################
+# pmm Service + Deployment
+###################################
+{{- define "pmm" -}}
+# set tuple values to more recognizable variables
+{{- $pmm := index . 0 -}}
+{{- $namespace := index . 1 -}}
+
+###################################
+# pmm Service
+###################################
+kind: Service
+apiVersion: v1
+metadata:
+  name: pmm
+  labels:
+    component: pmm
+    app: vitess
+spec:
+  ports:
+    - name: web
+      port: 80
+
+  selector:
+    component: pmm
+    app: vitess
+  type: ClusterIP
+---
+###################################
+# pmm StatefulSet
+###################################
+apiVersion: apps/v1beta1
+kind: StatefulSet
+metadata:
+  name: pmm
+spec:
+  serviceName: pmm
+  replicas: 1
+  updateStrategy: 
+    type: RollingUpdate
+  selector:
+    matchLabels:
+      app: vitess
+      component: pmm
+  template:
+    metadata:
+      labels:
+        app: vitess
+        component: pmm
+    spec:
+      containers:
+        - name: pmm
+          image: "percona/pmm-server:{{ $pmm.pmmTag }}"
+
+          ports:
+            - name: web
+              containerPort: 80
+
+          volumeMounts:
+            - name: pmmdata
+              mountPath: /pmmdata
+
+          resources:
+{{ toYaml $pmm.server.resources | indent 12 }}
+
+          env:
+            - name: DISABLE_UPDATES
+              value: "true"
+
+            - name: DISABLE_TELEMETRY
+              value: {{ $pmm.server.env.disableTelemetry | quote }}
+
+            - name: METRICS_RESOLUTION
+              value: {{ $pmm.server.env.metricsResolution | quote }}
+
+            - name: METRICS_RETENTION
+              value: {{ $pmm.server.env.metricsRetention | quote }}
+
+            - name: QUERIES_RETENTION
+              value: {{ $pmm.server.env.queriesRetention | quote }}
+
+            - name: METRICS_MEMORY
+              value: {{ $pmm.server.env.metricsMemory | quote }}
+
+          command: ["bash"]
+          args:
+            - "-c"
+            - |
+              set -ex
+
+
+              if [ ! -f /pmmdata/vitess-init ]; then
+                  # the PV hasn't been initialized, so copy over default
+                  # pmm-server directories before symlinking
+                  mkdir -p /pmmdata
+
+                  mv /opt/prometheus/data /pmmdata/data
+                  mv /opt/consul-data /pmmdata
+                  mv /var/lib/mysql /pmmdata
+                  mv /var/lib/grafana /pmmdata
+
+                  # initialize the PV and then mark it complete
+                  touch /pmmdata/vitess-init
+              else
+                  # remove the default directories so we can symlink the
+                  # existing PV directories
+                  rm -Rf /opt/prometheus/data
+                  rm -Rf /opt/consul-data
+                  rm -Rf /var/lib/mysql
+                  rm -Rf /var/lib/grafana
+              fi
+
+              # symlink pmm-server paths to point to our PV
+              ln -s /pmmdata/data /opt/prometheus/
+              ln -s /pmmdata/consul-data /opt/
+              ln -s /pmmdata/mysql /var/lib/
+              ln -s /pmmdata/grafana /var/lib/
+
+              /opt/entrypoint.sh
+
+  volumeClaimTemplates:
+    - metadata:
+        name: pmmdata
+        annotations:
+{{ toYaml $pmm.server.dataVolumeClaimAnnotations | indent 10 }}
+      spec:
+{{ toYaml $pmm.server.dataVolumeClaimSpec | indent 8 }}
+
+{{- end -}}
+
+###################################
+# sidecar container running pmm-client
+###################################
+{{- define "cont-pmm-client" -}}
+{{- $pmm := index . 0 -}}
+{{- $namespace := index . 1 -}}
+
+- name: "pmm-client"
+  image: "vitess/pmm-client:{{ $pmm.pmmTag }}"
+  imagePullPolicy: IfNotPresent
+  volumeMounts:
+    - name: vtdataroot
+      mountPath: "/vtdataroot"
+  ports:
+    - containerPort: 42001
+      name: query-data
+    - containerPort: 42002
+      name: mysql-metrics
+
+  securityContext:
+    # PMM requires root privileges
+    runAsUser: 0
+
+  resources:
+{{ toYaml $pmm.client.resources | indent 4 }}
+
+  command: ["bash"]
+  args:
+    - "-c"
+    - |
+      set -ex
+
+      mkdir -p /vtdataroot/pmm
+
+      # redirect logs to PV
+      ln -s /vtdataroot/pmm/pmm-mysql-metrics-42002.log /var/log/pmm-mysql-metrics-42002.log
+
+      # --force is used because the pod ip address may have changed
+      pmm-admin config --server pmm.{{ $namespace }} --force
+
+      # creates a systemd service
+      # TODO: remove "|| true" after https://jira.percona.com/projects/PMM/issues/PMM-1985 is resolved
+      pmm-admin add mysql:metrics --user root --socket /vtdataroot/tabletdata/mysql.sock --force || true
+
+      # keep the container alive but still responsive to stop requests
+      trap : TERM INT; sleep infinity & wait
+
+- name: pmm-client-metrics-log
+  image: busybox
+  command: ["/bin/sh"]
+  args: ["-c", "tail -n+1 -F /vtdataroot/pmm/pmm-mysql-metrics-42002.log"]
+  volumeMounts:
+    - name: vtdataroot
+      mountPath: /vtdataroot
+
+{{- end -}}

helm/vitess/templates/_vttablet.tpl

@@ -2,6 +2,8 @@
 # vttablet Service
 ###################################
 {{- define "vttablet-service" -}}
+# set tuple values to more recognizable variables
+{{- $pmm := index . 0 -}}
 apiVersion: v1
 kind: Service
 metadata:
@@ -17,6 +19,12 @@ spec:
       name: web
     - port: 16002
       name: grpc
+{{ if $pmm.enabled }}
+    - port: 42001
+      name: query-data
+    - port: 42002
+      name: mysql-metrics
+{{ end }}
   clusterIP: None
   selector:
     app: vitess
@@ -37,6 +45,7 @@ spec:
 {{- $defaultVttablet := index . 5 -}}
 {{- $namespace := index . 6 -}}
 {{- $config := index . 7 -}}
+{{- $pmm := index . 8 -}}
 
 # sanitize inputs to create tablet name
 {{- $cellClean := include "clean-label" $cell.name -}}
@@ -96,6 +105,7 @@ spec:
 {{ include "cont-vttablet" (tuple $topology $cell $keyspace $shard $tablet $defaultVttablet $vitessTag $uid $namespace $config) | indent 8 }}
 {{ include "cont-mysql-errorlog" . | indent 8 }}
 {{ include "cont-mysql-slowlog" . | indent 8 }}
+{{ if $pmm.enabled }}{{ include "cont-pmm-client" (tuple $pmm $namespace) | indent 8 }}{{ end }}
 
       volumes:
         - name: vt

helm/vitess/templates/vitess.yaml

@@ -1,8 +1,14 @@
 # Create global resources.
 ---
 # create a single vttablet service
-{{ include "vttablet-service" $ }}
+{{ include "vttablet-service" (tuple $.Values.pmm) }}
 ---
+{{ if $.Values.pmm.enabled }}
+# create the pmm service and stateful set
+{{ include "pmm" (tuple $.Values.pmm $.Release.Namespace) }}
+---
+{{ end }}
+
 # create an etcd cluster for the global topology
 {{- $replicas := $.Values.topology.globalCell.replicas | default $.Values.etcd.replicas -}}
 {{- $version := $.Values.topology.globalCell.version | default $.Values.etcd.version -}}
@@ -33,7 +39,7 @@
     {{ range $shard := $keyspace.shards }}
       {{ range $tablet := $shard.tablets }}
 ---
-{{ include "vttablet" (tuple $.Values.topology $cell $keyspace $shard $tablet $.Values.vttablet $.Release.Namespace $.Values.config) }}
+{{ include "vttablet" (tuple $.Values.topology $cell $keyspace $shard $tablet $.Values.vttablet $.Release.Namespace $.Values.config $.Values.pmm) }}
       {{ end }} # range $tablet
     {{ end }} # range $shard
   {{ end }} # range $keyspace

helm/vitess/values.yaml

@@ -152,3 +152,58 @@ vttablet:
     resources:
       requests:
         storage: "10Gi"
+
+# Default values for pmm
+pmm:
+  enabled: false
+  pmmTag: "1.6.1"
+  client:
+    resources:
+      requests:
+        cpu: 50m
+        memory: 128Mi
+      limits:
+        cpu: 200m
+        memory: 256Mi
+  server:
+    resources:
+      limits:
+        cpu: 500m
+        memory: 1Gi
+    # PVC for pmm
+    dataVolumeClaimAnnotations:
+    dataVolumeClaimSpec:
+      # storageClassName: "pd-ssd"
+      accessModes: ["ReadWriteOnce"]
+      resources:
+        requests:
+          storage: "10Gi"
+    env:
+      # DISABLE_TELEMETRY
+      # With telemetry enabled, your PMM Server sends some statistics to v.percona.com every 24 hours
+      disableTelemetry: true
+
+      # METRICS_RESOLUTION (Option)
+      # This option sets the minimum resolution for checking metrics. You should set it if the latency is higher than 1 second
+      metricsResolution: 1s
+
+      # METRICS_RETENTION (Option)
+      # This option determines how long metrics are stored at PMM Server. 
+      # The value is passed as a combination of hours, minutes, and seconds, such as 720h0m0s. 
+      # The minutes (a number followed by m) and seconds (a number followed by s) are optional.
+      metricsRetention: 720h
+
+      # QUERIES_RETENTION
+      # This option determines how many days queries are stored at PMM Server
+      queriesRetention: 8
+
+      # METRICS_MEMORY (Option) -- TODO: automatically calculate based on resource limits
+      # NOTE: The value must be passed in kilobytes
+      # NOTE: Make sure to quote this value so it isn't converted into scientific notation
+
+      # By default, Prometheus in PMM Server uses up to 768 MB of memory for storing the most recently used data chunks. 
+      # Depending on the amount of data coming into Prometheus, you may require a higher limit to avoid throttling data ingestion,
+      # or allow less memory consumption if it is needed for other processes.
+      # The limit affects only memory reserved for data chunks. Actual RAM usage by Prometheus is higher. 
+      # It is recommended to set this limit to roughly 2/3 of the total memory that you are planning to allow for Prometheus.
+      metricsMemory: "600000"