reworking and adding customizability to probes (valeriano-manassero#206)

cccs-nik · Jun 20, 2024 · 1cfa474 · 1cfa474
1 parent 64984ac
commit 1cfa474
Show file tree

Hide file tree

Showing 6 changed files with 62 additions and 37 deletions.
diff --git a/valeriano-manassero/trino/Chart.yaml b/valeriano-manassero/trino/Chart.yaml
@@ -2,7 +2,7 @@ apiVersion: v1
 appVersion: "433"
 description: High performance, distributed SQL query engine for big data
 name: trino
-version: 10.0.0
+version: 10.1.0
 kubeVersion: ">= 1.24.0-0 < 1.31.0-0"
 home: https://trino.io
 icon: https://trino.io/assets/images/trino-logo/trino-ko_tiny-alt.svg
@@ -27,6 +27,4 @@ keywords:
 annotations:
   artifacthub.io/changes: |
     - kind: changed
-      description: Reduce default jvm configuration to make easier to customize
-    - kind: changed
-      description: Increase kubernetes version support
+      description: Adding customizability to probes and changing what is probed
diff --git a/valeriano-manassero/trino/README.md b/valeriano-manassero/trino/README.md
@@ -1,6 +1,6 @@
 # trino
 
-![Version: 10.0.0](https://img.shields.io/badge/Version-10.0.0-informational?style=flat-square) ![AppVersion: 433](https://img.shields.io/badge/AppVersion-433-informational?style=flat-square)
+![Version: 10.1.0](https://img.shields.io/badge/Version-10.1.0-informational?style=flat-square) ![AppVersion: 433](https://img.shields.io/badge/AppVersion-433-informational?style=flat-square)
 
 High performance, distributed SQL query engine for big data
 
@@ -38,10 +38,12 @@ Kubernetes: `>= 1.24.0-0 < 1.31.0-0`
 | config.coordinator.jvm.gcMethod.type | string | `"UseG1GC"` |  |
 | config.coordinator.jvm.maxRAMPercentage | float | `80` |  |
 | config.coordinator.jvmExtraConfig | string | `"-agentpath:/usr/lib/trino/bin/libjvmkill.so\n-XX:+ExplicitGCInvokesConcurrent\n-XX:+HeapDumpOnOutOfMemoryError\n-XX:+ExitOnOutOfMemoryError\n-XX:-OmitStackTraceInFastThrow\n-XX:ReservedCodeCacheSize=512M\n-XX:PerMethodRecompilationCutoff=10000\n-XX:PerBytecodeRecompilationCutoff=10000\n-Djdk.attach.allowAttachSelf=true\n-Djdk.nio.maxCachedBufferSize=2000000\n-XX:+UnlockDiagnosticVMOptions\n-XX:G1NumCollectionsKeepPinned=10000000"` |  |
+| config.coordinator.livenessProbe | object | `{}` |  |
 | config.coordinator.nodeSelector | object | `{}` |  |
 | config.coordinator.podAnnotations | object | `{}` |  |
 | config.coordinator.podLabels | object | `{}` |  |
 | config.coordinator.query.maxMemoryPerNode | string | `"1GB"` |  |
+| config.coordinator.readinessProbe | object | `{}` |  |
 | config.coordinator.replicas | int | `1` |  |
 | config.coordinator.resources | object | `{}` |  |
 | config.coordinator.tolerations | list | `[]` |  |
@@ -81,10 +83,12 @@ Kubernetes: `>= 1.24.0-0 < 1.31.0-0`
 | config.worker.jvm.maxRAMPercentage | float | `80` |  |
 | config.worker.jvmExtraConfig | string | `"-agentpath:/usr/lib/trino/bin/libjvmkill.so\n-XX:+ExplicitGCInvokesConcurrent\n-XX:+HeapDumpOnOutOfMemoryError\n-XX:+ExitOnOutOfMemoryError\n-XX:-OmitStackTraceInFastThrow\n-XX:ReservedCodeCacheSize=512M\n-XX:PerMethodRecompilationCutoff=10000\n-XX:PerBytecodeRecompilationCutoff=10000\n-Djdk.attach.allowAttachSelf=true\n-Djdk.nio.maxCachedBufferSize=2000000\n-XX:+UnlockDiagnosticVMOptions\n-XX:G1NumCollectionsKeepPinned=10000000"` |  |
 | config.worker.lifecycle | object | `{}` |  |
+| config.worker.livenessProbe | object | `{}` |  |
 | config.worker.nodeSelector | object | `{}` |  |
 | config.worker.podAnnotations | object | `{}` |  |
 | config.worker.podLabels | object | `{}` |  |
 | config.worker.query.maxMemoryPerNode | string | `"1GB"` |  |
+| config.worker.readinessProbe | object | `{}` |  |
 | config.worker.replicas | int | `2` | Replica count when autoscaler is disabled. If autoscaler is enabled, it sets minimum number of replicas. |
 | config.worker.resources | object | `{}` |  |
 | config.worker.tolerations | list | `[]` |  |

diff --git a/valeriano-manassero/trino/templates/configmap-worker.yaml b/valeriano-manassero/trino/templates/configmap-worker.yaml
@@ -58,18 +58,6 @@ data:
   {{- end }}
 {{ end }}
 
----
-apiVersion: v1
-kind: ConfigMap
-metadata:
-  name: {{ template "trino.worker" . }}-health-check
-  labels:
-    {{- include "trino.labels" . | nindent 4 }}
-    app.kubernetes.io/component: worker
-data:
-  health_check.sh: |
-    #!/bin/bash
-    curl -H 'X-trino-User: healthCheck' -m 3 --silent {{ template "trino.fullname" . }}:{{ .Values.config.general.http.port }}/v1/info/state | grep --silent 'ACTIVE'
 ---
 apiVersion: v1
 kind: ConfigMap

diff --git a/valeriano-manassero/trino/templates/deployment-coordinator.yaml b/valeriano-manassero/trino/templates/deployment-coordinator.yaml
@@ -287,12 +287,21 @@ spec:
             {{- end }}
           livenessProbe:
             httpGet:
-              path: /v1/status
+              path: /v1/info
               port: http-coord
+            initialDelaySeconds: {{ .Values.config.coordinator.livenessProbe.initialDelaySeconds | default 30 }}
+            periodSeconds: {{ .Values.config.coordinator.livenessProbe.periodSeconds | default 10 }}
+            timeoutSeconds: {{ .Values.config.coordinator.livenessProbe.timeoutSeconds | default 5 }}
+            failureThreshold: {{ .Values.config.coordinator.livenessProbe.failureThreshold | default 6 }}
+            successThreshold: {{ .Values.config.coordinator.livenessProbe.successThreshold | default 1 }}
           readinessProbe:
-            httpGet:
-              path: /v1/status
-              port: http-coord
+            exec:
+              command: [/usr/lib/trino/bin/health-check]
+            initialDelaySeconds: {{ .Values.config.coordinator.readinessProbe.initialDelaySeconds | default 10 }}
+            periodSeconds: {{ .Values.config.coordinator.readinessProbe.periodSeconds | default 10 }}
+            timeoutSeconds: {{ .Values.config.coordinator.readinessProbe.timeoutSeconds | default 5 }}
+            failureThreshold: {{ .Values.config.coordinator.readinessProbe.failureThreshold | default 6 }}
+            successThreshold: {{ .Values.config.coordinator.readinessProbe.successThreshold | default 1 }}
           resources:
 {{ toYaml .Values.config.coordinator.resources | indent 12 }}
     {{- with .Values.config.coordinator.nodeSelector }}

diff --git a/valeriano-manassero/trino/templates/deployment-worker.yaml b/valeriano-manassero/trino/templates/deployment-worker.yaml
@@ -39,10 +39,6 @@ spec:
             - secret:
                 name: {{ .Values.faultToleranceSecret | default "trino-fault-tolerance" }}
             {{- end }}
-        - name: health-check-volume
-          configMap:
-            name: {{ template "trino.worker" . }}-health-check
-            defaultMode: 0777
         {{- if eq .Values.config.general.catalogsMountType "configmap" }}
         - name: catalog-volume
           configMap:
@@ -130,8 +126,6 @@ spec:
               name: config-volume
             - mountPath: /etc/trino/schemas
               name: schemas-volume
-            - mountPath: /startup/
-              name: health-check-volume
             - mountPath: {{ .Values.config.general.path }}/catalog
               name: catalog-volume
             - mountPath: {{ .Values.config.general.path }}/certs
@@ -160,8 +154,11 @@ spec:
             - name: {{ $mountName }}
             {{- tpl (toYaml $mount) $ | nindent 14 -}}
             {{- end }}
-          {{- if .Values.jmxExporter.worker.enabled }}
           ports:
+            - name: http-worker
+              containerPort: {{ .Values.config.general.http.port }}
+              protocol: TCP
+          {{- if .Values.jmxExporter.worker.enabled }}
             - name: jmx-exporter
               containerPort: {{ .Values.jmxExporter.port | default 9000 }}
               protocol: TCP
@@ -171,17 +168,22 @@ spec:
 {{ toYaml . | nindent 12 }}
           {{- end }}
           livenessProbe:
-            exec:
-              command:
-                - /startup/health_check.sh
-            initialDelaySeconds: 10
-            periodSeconds: 25
+            httpGet:
+              path: /v1/info
+              port: http-worker
+            initialDelaySeconds: {{ .Values.config.worker.livenessProbe.initialDelaySeconds | default 30 }}
+            periodSeconds: {{ .Values.config.worker.livenessProbe.periodSeconds | default 10 }}
+            timeoutSeconds: {{ .Values.config.worker.livenessProbe.timeoutSeconds | default 5 }}
+            failureThreshold: {{ .Values.config.worker.livenessProbe.failureThreshold | default 6 }}
+            successThreshold: {{ .Values.config.worker.livenessProbe.successThreshold | default 1 }}
           readinessProbe:
             exec:
-              command:
-                - /startup/health_check.sh
-            initialDelaySeconds: 5
-            periodSeconds: 10
+              command: [/usr/lib/trino/bin/health-check]
+            initialDelaySeconds: {{ .Values.config.worker.readinessProbe.initialDelaySeconds | default 10 }}
+            periodSeconds: {{ .Values.config.worker.readinessProbe.periodSeconds | default 10 }}
+            timeoutSeconds: {{ .Values.config.worker.readinessProbe.timeoutSeconds | default 5 }}
+            failureThreshold: {{ .Values.config.worker.readinessProbe.failureThreshold | default 6 }}
+            successThreshold: {{ .Values.config.worker.readinessProbe.successThreshold | default 1 }}
           resources:
 {{ toYaml .Values.config.worker.resources | indent 12 }}
     {{- with .Values.config.worker.nodeSelector }}

diff --git a/valeriano-manassero/trino/values.yaml b/valeriano-manassero/trino/values.yaml
@@ -100,6 +100,18 @@ config:
     extraConfig: ""
     query:
       maxMemoryPerNode: "1GB"
+    livenessProbe: {}
+    #  initialDelaySeconds: 20
+    #  periodSeconds: 10
+    #  timeoutSeconds: 5
+    #  failureThreshold: 6
+    #  successThreshold: 1
+    readinessProbe: {}
+    #  initialDelaySeconds: 20
+    #  periodSeconds: 10
+    #  timeoutSeconds: 5
+    #  failureThreshold: 6
+    #  successThreshold: 1
 
   worker:
     # -- Replica count when autoscaler is disabled. If autoscaler is enabled, it sets minimum number of replicas.
@@ -166,6 +178,18 @@ config:
       # localtemp:
       #   hostPath:
       #     path: /mnt
+    livenessProbe: {}
+    #  initialDelaySeconds: 20
+    #  periodSeconds: 10
+    #  timeoutSeconds: 5
+    #  failureThreshold: 6
+    #  successThreshold: 1
+    readinessProbe: {}
+    #  initialDelaySeconds: 20
+    #  periodSeconds: 10
+    #  timeoutSeconds: 5
+    #  failureThreshold: 6
+    #  successThreshold: 1
 
 
 eventListenerProperties: {}