Make router keep-alive setting configurable (#1225)

charts/fission-all/templates/deployment.yaml

@@ -197,26 +197,28 @@ spec:
         command: ["/fission-bundle"]
         args: ["--routerPort", "8888", "--executorUrl", "http://executor.{{ .Release.Namespace }}", "--collectorEndpoint", "{{ .Values.traceCollectorEndpoint }}"]
         env:
-        - name: POD_NAMESPACE
-          valueFrom:
-            fieldRef:
-              fieldPath: metadata.namespace
-        - name: ROUTER_ROUND_TRIP_TIMEOUT
-          value: {{ .Values.routerRoundTripTimeout | default "50ms" | quote }}
-        - name: ROUTER_ROUNDTRIP_TIMEOUT_EXPONENT
-          value: {{ .Values.routerRoundTripTimeoutExponent | default 2 | quote }}
-        - name: ROUTER_ROUND_TRIP_KEEP_ALIVE_TIME
-          value: {{ .Values.routerRoundTripKeepAliveTime | default "30s" | quote }}
-        - name: ROUTER_ROUND_TRIP_MAX_RETRIES
-          value: {{ .Values.routerRoundTripMaxRetries | default 10 | quote }}
-        - name: ROUTER_ROUND_TRIP_SVC_ADDRESS_MAX_RETRIES
-          value: {{ .Values.routerRoundTripSvcAddressMaxRetries | default 5 | quote }}
-        - name: ROUTER_ROUND_TRIP_SVC_ADDRESS_UPDATE_TIMEOUT
-          value: {{ .Values.routerRoundTripSvcAddressUpdateTimeout | default 30 | quote }}
-        - name: DEBUG_ENV
-          value: {{ .Values.debugEnv | quote }}
-        - name: TRACING_SAMPLING_RATE
-          value: {{ .Values.traceSamplingRate | default "0.5" | quote }}
+          - name: POD_NAMESPACE
+            valueFrom:
+              fieldRef:
+                fieldPath: metadata.namespace
+          - name: ROUTER_ROUND_TRIP_TIMEOUT
+            value: {{ .Values.router.roundTrip.timeout | default "50ms" | quote }}
+          - name: ROUTER_ROUNDTRIP_TIMEOUT_EXPONENT
+            value: {{ .Values.router.roundTrip.timeoutExponent | default 2 | quote }}
+          - name: ROUTER_ROUND_TRIP_KEEP_ALIVE_TIME
+            value: {{ .Values.router.roundTrip.keepAliveTime | default "30s" | quote }}
+          - name: ROUTER_ROUND_TRIP_DISABLE_KEEP_ALIVE
+            value: {{ .Values.router.roundTrip.disableKeepAlive | default true | quote }}
+          - name: ROUTER_ROUND_TRIP_MAX_RETRIES
+            value: {{ .Values.router.roundTrip.maxRetries | default 10 | quote }}
+          - name: ROUTER_SVC_ADDRESS_MAX_RETRIES
+            value: {{ .Values.router.svcAddressMaxRetries | default 5 | quote }}
+          - name: ROUTER_SVC_ADDRESS_UPDATE_TIMEOUT
+            value: {{ .Values.router.svcAddressUpdateTimeout | default "30s" | quote }}
+          - name: DEBUG_ENV
+            value: {{ .Values.debugEnv | quote }}
+          - name: TRACING_SAMPLING_RATE
+            value: {{ .Values.traceSamplingRate | default "0.5" | quote }}
 {{ if .Values.analytics }}
         - name: ANALYTICS_URL
           value: "https://g.fission.sh/metrics"

charts/fission-all/values.yaml

@@ -57,6 +57,36 @@ logger:
   fluentdImage: fluent/fluent-bit
   fluentdImageTag: 1.0.4
 
+## Router config
+router:
+  svcAddressMaxRetries: 5
+  svcAddressUpdateTimeout: 30s
+  roundTrip:
+    ## If true, router will disable the HTTP keep-alive which result in performance degradation.
+    ## But it ensures that router can redirect new coming requests to new function pods.
+    ##
+    ## If false, router will enable transport keep-alive feature for better performance.
+    ## However, the drawback is it takes longer to switch to newly created function pods
+    ## if using newdeploy as executor type for function. If you want to preserve the
+    ## performance while keeping the short switching time to new function, you can create
+    ## an environment with short grace period by setting flag "--graceperiod" (default 360s),
+    ## so that kubernetes will be able to reap old function pod quickly.
+    ##
+    ## For details, see https://github.com/fission/fission/issues/723
+    disableKeepAlive: false
+
+    ## The keep-alive period for an active network connection to function pod.
+    keepAliveTime: 30s
+
+    ## HTTP transport request timeout
+    timeout: 50ms
+
+    ## The length of request timeout will multiply with timeoutExponent after each retry
+    timeoutExponent: 2
+
+    ## Max retries times of a failed request
+    maxRetries: 10
+
 ## Message queue trigger config
 ### NATS Streaming, enabled by default
 nats:
@@ -146,7 +176,6 @@ preUpgradeChecksImage: fission/pre-upgrade-checks
 ## summary is returned as part of http response
 debugEnv: false
 
-
 ## set this flag to true if prometheus needs to be deployed along with fission
 prometheusDeploy: true
 
@@ -154,7 +183,6 @@ prometheusDeploy: true
 canaryDeployment:
   enabled: true
 
-
 # Use these flags to enable opentracing, the variable is endpoint of Jaeger collector in the format shown below
 #traceCollectorEndpoint: "http://jaeger-collector.jaeger.svc:14268/api/traces?format=jaeger.thrift"
 #traceSamplingRate: 0.75