Use a better readiness probe

Use the health check script as the readiness probe, because it also
checks the response of the `/v1/info` endpoint to make sure the server
actually finished starting up. This only makes sense for the readiness
probe, so keep the liveness probe as-is.

Author: nineinchnick

charts/trino/templates/deployment-coordinator.yaml

@@ -172,16 +172,15 @@ spec:
             httpGet:
               path: /v1/info
               port: http
-            initialDelaySeconds: {{ .Values.coordinator.livenessProbe.initialDelaySeconds | default 15 }}
+            initialDelaySeconds: {{ .Values.coordinator.livenessProbe.initialDelaySeconds | default 30 }}
             periodSeconds: {{ .Values.coordinator.livenessProbe.periodSeconds | default 10 }}
             timeoutSeconds: {{ .Values.coordinator.livenessProbe.timeoutSeconds | default 5 }}
             failureThreshold: {{ .Values.coordinator.livenessProbe.failureThreshold | default 6 }}
             successThreshold: {{ .Values.coordinator.livenessProbe.successThreshold | default 1 }}
           readinessProbe:
-            httpGet:
-              path: /v1/info
-              port: http
-            initialDelaySeconds: {{ .Values.coordinator.readinessProbe.initialDelaySeconds | default 15 }}
+            exec:
+              command: [/usr/lib/trino/bin/health-check]
+            initialDelaySeconds: {{ .Values.coordinator.readinessProbe.initialDelaySeconds | default 10 }}
             periodSeconds: {{ .Values.coordinator.readinessProbe.periodSeconds | default 10 }}
             timeoutSeconds: {{ .Values.coordinator.readinessProbe.timeoutSeconds | default 5 }}
             failureThreshold: {{ .Values.coordinator.readinessProbe.failureThreshold | default 6 }}

charts/trino/templates/deployment-worker.yaml

@@ -138,16 +138,15 @@ spec:
             httpGet:
               path: /v1/info
               port: http
-            initialDelaySeconds: {{ .Values.worker.livenessProbe.initialDelaySeconds | default 15 }}
+            initialDelaySeconds: {{ .Values.worker.livenessProbe.initialDelaySeconds | default 30 }}
             periodSeconds: {{ .Values.worker.livenessProbe.periodSeconds | default 10 }}
             timeoutSeconds: {{ .Values.worker.livenessProbe.timeoutSeconds | default 5 }}
             failureThreshold: {{ .Values.worker.livenessProbe.failureThreshold | default 6 }}
             successThreshold: {{ .Values.worker.livenessProbe.successThreshold | default 1 }}
           readinessProbe:
-            httpGet:
-              path: /v1/info
-              port: http
-            initialDelaySeconds: {{ .Values.worker.readinessProbe.initialDelaySeconds | default 15 }}
+            exec:
+              command: [/usr/lib/trino/bin/health-check]
+            initialDelaySeconds: {{ .Values.worker.readinessProbe.initialDelaySeconds | default 10 }}
             periodSeconds: {{ .Values.worker.readinessProbe.periodSeconds | default 10 }}
             timeoutSeconds: {{ .Values.worker.readinessProbe.timeoutSeconds | default 5 }}
             failureThreshold: {{ .Values.worker.readinessProbe.failureThreshold | default 6 }}