Increase timing between polling (#102)

* increasing time between probe checks * adjusting startup delay to 60 seconds
microsoft · Aug 1, 2024 · b41347e · b41347e
1 parent 86ae868
commit b41347e
Show file tree

Hide file tree

Showing 3 changed files with 30 additions and 6 deletions.
diff --git a/chart/templates/_secrets.tpl b/chart/templates/_secrets.tpl
@@ -22,6 +22,7 @@ data:
   spacefx_dir_xfer: {{ printf "%s/%s/%s" $globalValues.spacefxDirectories.base $globalValues.spacefxDirectories.xfer $serviceValues.appName | b64enc }}
   heartbeatpulsetimingms: {{ printf "%.0f" $globalValues.appConfig.heartBeatPulseTimingMS | b64enc }}
   heartbeatreceivedtolerancems: {{ printf "%.0f" $globalValues.appConfig.heartBeatReceivedToleranceMS | b64enc }}
+  heartbeatreceivedcriticaltolerancems: {{ printf "%.0f" $globalValues.appConfig.heartBeatReceivedCriticalToleranceMS | b64enc }}
   resourcemonitorenabled: {{ $globalValues.appConfig.resourceMonitorEnabled | ternary "true" "false" | b64enc }}
   resourcemonitortimingms: {{ printf "%.0f" $globalValues.appConfig.resourceMonitorTimingMS | b64enc }}
   resourcescavengerenabled: {{ $globalValues.appConfig.resourceScavengerEnabled | ternary "true" "false" | b64enc }}

diff --git a/chart/templates/_service.tpl b/chart/templates/_service.tpl
@@ -39,26 +39,35 @@ spec:
 {{- include "spacefx.daprannotations" (dict "globalValues" $globalValues "serviceValues" $serviceValues) | indent 8 }}
     spec:
       serviceAccountName: {{  $serviceValues.appName | quote }}
+      terminationGracePeriodSeconds: 1
       {{- if eq $serviceValues.appName "platform-mts" }}
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
       {{- end }}
       containers:
         - name: {{ $serviceValues.appName | quote }}
-        {{- if or $serviceValues.appHealthChecks $serviceValues.debugShim }}
           ports:
-          - name: liveness-port
+          - name: app-port
             containerPort: 50051
+        {{- if or $serviceValues.appHealthChecks $serviceValues.debugShim }}
+          {{- if $globalValues.probes.liveness.enabled }}
           livenessProbe:
             grpc:
               port: 50051
-            failureThreshold: 2
-            periodSeconds: 5
+            failureThreshold: {{ $globalValues.probes.liveness.failureThreshold }}
+            periodSeconds: {{ $globalValues.probes.liveness.periodSeconds }}
+            initialDelaySeconds: {{ $globalValues.probes.liveness.initialDelaySeconds }}
+            timeoutSeconds: {{ $globalValues.probes.liveness.timeoutSeconds }}
+          {{- end }}
+          {{- if $globalValues.probes.startup.enabled }}
           startupProbe:
             grpc:
               port: 50051
-            failureThreshold: 100
-            periodSeconds: 5
+            failureThreshold: {{ $globalValues.probes.startup.failureThreshold }}
+            periodSeconds: {{ $globalValues.probes.startup.periodSeconds }}
+            initialDelaySeconds: {{ $globalValues.probes.startup.initialDelaySeconds }}
+            timeoutSeconds: {{ $globalValues.probes.startup.timeoutSeconds }}
+          {{- end }}
         {{- end }}
         {{- if $serviceValues.securityContext }}
           securityContext:

diff --git a/chart/values.yaml b/chart/values.yaml
@@ -3,10 +3,24 @@ global:
   appConfig:
     heartBeatPulseTimingMS: 2000
     heartBeatReceivedToleranceMS: 10000
+    heartBeatReceivedCriticalToleranceMS: 60000
     resourceMonitorEnabled: false
     resourceMonitorTimingMS: 5000
     resourceScavengerEnabled: true
     resourceScavengerTimingMS: 30000
+  probes:
+    liveness:
+      enabled: true
+      initialDelaySeconds: 30
+      periodSeconds: 30
+      timeoutSeconds: 5
+      failureThreshold: 3
+    startup:
+      enabled: true
+      initialDelaySeconds: 60
+      periodSeconds: 10
+      timeoutSeconds: 5
+      failureThreshold: 100
   containerCommand: dotnet
   containerRegistry: registry.spacefx.local:5000
   containerRegistryInternal: coresvc-registry.coresvc.svc.cluster.local:5000