Merge pull request #110 from sustainable-computing-io/issue-109

Improve configuration and debugging
sustainable-computing-io · Jun 21, 2024 · 0ba84eb · 0ba84eb
2 parents e0ea62c + f38ebee
commit 0ba84eb
Show file tree

Hide file tree

Showing 7 changed files with 57 additions and 14 deletions.
diff --git a/Dockerfile b/Dockerfile
@@ -26,8 +26,11 @@ RUN CGO_ENABLED=0 GOOS=${TARGETOS:-linux} GOARCH=${TARGETARCH} go build -a -o ma
 # Use distroless as minimal base image to package the manager binary
 # Refer to https://github.com/GoogleContainerTools/distroless for more details
 FROM gcr.io/distroless/static:nonroot
+# FROM gcr.io/distroless/static:debug
 WORKDIR /
 COPY --from=builder /workspace/manager .
+COPY cmd/debug-entrypoint.sh .
 USER 65532:65532
 
 ENTRYPOINT ["/manager"]
+# ENTRYPOINT ["/debug-entrypoint.sh"]
diff --git a/bundle/manifests/susql-operator.clusterserviceversion.yaml b/bundle/manifests/susql-operator.clusterserviceversion.yaml
@@ -23,7 +23,7 @@ metadata:
     capabilities: Basic Install
     categories: Monitoring
     containerImage: quay.io/sustainable_computing_io/susql_operator:0.0.19
-    createdAt: "2024-06-14T17:42:05Z"
+    createdAt: "2024-06-21T03:56:29Z"
     description: 'Aggregates energy data from pods tagged with SusQL labels '
     operators.operatorframework.io/builder: operator-sdk-v1.34.1
     operators.operatorframework.io/project_layout: go.kubebuilder.io/v4
@@ -197,14 +197,14 @@ spec:
                     drop:
                     - ALL
               - args:
-                - --leader-elect
+                - --leader-elect=$(LEADER-ELECT)
                 - --kepler-prometheus-url=$(KEPLER-PROMETHEUS-URL)
                 - --kepler-metric-name=$(KEPLER-METRIC-NAME)
                 - --susql-prometheus-database-url=$(SUSQL-PROMETHEUS-DATABASE-URL)
                 - --susql-prometheus-metrics-url=$(SUSQL-PROMETHEUS-METRICS-URL)
                 - --sampling-rate=$(SAMPLING-RATE)
-                - --health-probe-bind-address=:8081
-                - --metrics-bind-address=127.0.0.1:9999
+                - --health-probe-bind-address=$(HEALTH-PROBE-BIND-ADDRESS)
+                - --metrics-bind-address=$(METRICS-BIND-ADDRESS)
                 command:
                 - /manager
                 env:
@@ -218,6 +218,12 @@ spec:
                   value: http://0.0.0.0:8082
                 - name: SAMPLING-RATE
                   value: "2"
+                - name: LEADER-ELECT
+                  value: "true"
+                - name: HEALTH-PROBE-BIND-ADDRESS
+                  value: :8081
+                - name: METRICS-BIND-ADDRESS
+                  value: 127.0.0.1:9999
                 image: quay.io/sustainable_computing_io/susql_operator:0.0.19
                 imagePullPolicy: IfNotPresent
                 livenessProbe:

diff --git a/cmd/debug-entrypoint.sh b/cmd/debug-entrypoint.sh
@@ -0,0 +1,16 @@
+#!/busybox/sh
+
+echo
+date
+echo entered entrypoint.sh
+
+/manager
+ec=$?
+
+echo
+echo MANAGER TERMINATED!  ERROR=${ec} AT $(date)
+echo
+
+echo sleeping
+sleep 9999
+echo terminating 
diff --git a/cmd/main.go b/cmd/main.go
@@ -61,24 +61,29 @@ func main() {
 	var susqlPrometheusDatabaseUrl string
 	var samplingRate string
 
-	flag.StringVar(&metricsAddr, "metrics-bind-address", ":8082", "The address the metric endpoint binds to.")
-	flag.StringVar(&probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.")
-	flag.BoolVar(&enableLeaderElection, "leader-elect", false,
-		"Enable leader election for controller manager. "+
-			"Enabling this will ensure there is only one active controller manager.")
-
 	// NOTE: these can be set as env or flag, flag takes precedence over env
 	keplerPrometheusUrlEnv := os.Getenv("KEPLER-PROMETHEUS-URL")
 	keplerMetricNameEnv := os.Getenv("KEPLER-METRIC-NAME")
 	susqlPrometheusDatabaseUrlEnv := os.Getenv("SUSQL-PROMETHEUS-DATABASE-URL")
 	susqlPrometheusMetricsUrlEnv := os.Getenv("SUSQL-PROMETHEUS-METRICS-URL")
 	samplingRateEnv := os.Getenv("SAMPLING-RATE")
+	metricsAddrEnv := os.Getenv("METRICS-BIND-ADDRESS")
+	probeAddrEnv := os.Getenv("HEALTH-PROBE-BIND-ADDRESS")
+	enableLeaderElectionEnv, err := strconv.ParseBool(os.Getenv("LEADER-ELECT"))
+	if err != nil {
+		enableLeaderElectionEnv = false
+	}
 
 	flag.StringVar(&keplerPrometheusUrl, "kepler-prometheus-url", keplerPrometheusUrlEnv, "The URL for the Prometheus server where Kepler stores the energy data")
 	flag.StringVar(&keplerMetricName, "kepler-metric-name", keplerMetricNameEnv, "The metric name to be queried in the kepler Prometheus server")
 	flag.StringVar(&susqlPrometheusDatabaseUrl, "susql-prometheus-database-url", susqlPrometheusDatabaseUrlEnv, "The URL for the Prometheus database where SusQL stores the energy data")
 	flag.StringVar(&susqlPrometheusMetricsUrl, "susql-prometheus-metrics-url", susqlPrometheusMetricsUrlEnv, "The URL for the Prometheus metrics where SusQL exposes the energy data")
 	flag.StringVar(&samplingRate, "sampling-rate", samplingRateEnv, "Sampling rate in seconds")
+	flag.StringVar(&metricsAddr, "metrics-bind-address", metricsAddrEnv, "The address the metric endpoint binds to.")
+	flag.StringVar(&probeAddr, "health-probe-bind-address", probeAddrEnv, "The address the probe endpoint binds to.")
+	flag.BoolVar(&enableLeaderElection, "leader-elect", enableLeaderElectionEnv,
+		"Enable leader election for controller manager. "+
+			"Enabling this will ensure there is only one active controller manager.")
 
 	opts := zap.Options{
 		Development: true,

diff --git a/config/manager/manager.yaml b/config/manager/manager.yaml
@@ -67,6 +67,7 @@ spec:
       containers:
       - command:
         - /manager
+          #        - /debug-entrypoint.sh
         env:
           - name: KEPLER-PROMETHEUS-URL
             value: "https://thanos-querier.openshift-monitoring.svc.cluster.local:9091"
@@ -79,16 +80,22 @@ spec:
             value: "http://0.0.0.0:8082"
           - name: SAMPLING-RATE
             value: "2"
+          - name: LEADER-ELECT
+            value: "true"
+          - name: HEALTH-PROBE-BIND-ADDRESS
+            value: ":8081"
+          - name: METRICS-BIND-ADDRESS
+            value: "127.0.0.1:9999"
         args:
-        - --leader-elect
+        - --leader-elect=$(LEADER-ELECT)
         # - --deployment-namespace=susql-operator
         - --kepler-prometheus-url=$(KEPLER-PROMETHEUS-URL)
         - --kepler-metric-name=$(KEPLER-METRIC-NAME)
         - --susql-prometheus-database-url=$(SUSQL-PROMETHEUS-DATABASE-URL)
         - --susql-prometheus-metrics-url=$(SUSQL-PROMETHEUS-METRICS-URL)
         - --sampling-rate=$(SAMPLING-RATE)
-        - --health-probe-bind-address=:8081
-        - --metrics-bind-address=127.0.0.1:9999
+        - --health-probe-bind-address=$(HEALTH-PROBE-BIND-ADDRESS)
+        - --metrics-bind-address=$(METRICS-BIND-ADDRESS)
         image: '<OPERATOR_IMG>'
         imagePullPolicy: IfNotPresent
         name: manager

diff --git a/deployment/susql-controller/templates/deployment.yaml b/deployment/susql-controller/templates/deployment.yaml
@@ -25,10 +25,13 @@ spec:
                   imagePullPolicy: {{ .Values.imagePullPolicy | default "IfNotPresent" }}
                   args:
                       - "--kepler-prometheus-url={{ .Values.keplerPrometheusUrl }}"
-                      - "--kepler-metric-name={{ .Values.KeplerMetricName }}"
+                      - "--kepler-metric-name={{ .Values.keplerMetricName }}"
                       - "--susql-prometheus-database-url={{ .Values.susqlPrometheusDatabaseUrl }}"
                       - "--susql-prometheus-metrics-url={{ .Values.susqlPrometheusMetricsUrl }}"
                       - "--sampling-rate={{ .Values.samplingRate }}"
+                      - "--metrics-bind-address={{ .Values.metricsAddr }}"
+                      - "--health-prove-bind-address={{ .Values.healthProbeAddr }}"
+                      - "--leader-elect={{ .Values.leaderElect }}"
                   ports:
                       - name: metrics
                         containerPort: 8082

diff --git a/deployment/susql-controller/values.yaml b/deployment/susql-controller/values.yaml
@@ -24,3 +24,6 @@ keplerMetricName: "kepler_container_joules_total"
 susqlPrometheusDatabaseUrl: "http://prometheus-susql.openshift-kepler-operator.svc.cluster.local:9090"
 susqlPrometheusMetricsUrl: "http://0.0.0.0:8082"
 samplingRate: "2"
+metricsAddr: "127.0.0.1:9999"
+healthProbeAddr: ":8081"
+leaderElect: "true"