release: sourcegraph@4.5.0

base/monitoring/cadvisor/cadvisor.DaemonSet.yaml

@@ -26,7 +26,7 @@ spec:
       serviceAccountName: cadvisor
       containers:
         - name: cadvisor
-          image: index.docker.io/sourcegraph/cadvisor:4.4.2@sha256:4c3af0c4fd9ea4425d38f7d1a784833c5fd542542cdbb81292044773e686fa60
+          image: index.docker.io/sourcegraph/cadvisor:4.5.0@sha256:5117f2bc817c16fb129acb6f9b070af8f1be09d3d9a8f88e3297f7adfff9af0d
           args:
             # Kubernetes-specific flags below (other flags are baked into the Docker image)
             #

base/monitoring/grafana/grafana.StatefulSet.yaml

@@ -26,7 +26,7 @@ spec:
     spec:
       containers:
         - name: grafana
-          image: index.docker.io/sourcegraph/grafana:4.4.2@sha256:69777c3a895a03eee035c173c91c0f25893285118c06e51a67728ec4259e2296
+          image: index.docker.io/sourcegraph/grafana:4.5.0@sha256:f70a7f79c5c90cab0d5cfb8f3dbca4dc60ed390b045aff1a86079c87bfe9a8af
           terminationMessagePolicy: FallbackToLogsOnError
           ports:
             - containerPort: 3370

base/monitoring/grafana/rbac/kustomization.yaml

@@ -0,0 +1,4 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+resources:
+  - grafana.ServiceAccount.yaml

base/monitoring/jaeger/jaeger.Deployment.yaml

@@ -30,7 +30,7 @@ spec:
     spec:
       containers:
         - name: jaeger
-          image: index.docker.io/sourcegraph/jaeger-all-in-one:insiders@sha256:462ef3b4a5fa9227f04c2f4bc2968970fad0fcc9efbaf89adaad0ef98a24b53f
+          image: index.docker.io/sourcegraph/jaeger-all-in-one:4.5.0@sha256:461476b01968324a0d8cb43a0176713e006f99cdb1f2efc3ab2210fd0bb812c2
           args: ["--memory.max-traces=20000"]
           ports:
             - containerPort: 5775

base/monitoring/node-exporter/node-exporter.DaemonSet.yaml

@@ -24,7 +24,7 @@ spec:
     spec:
       containers:
         - name: node-exporter
-          image: index.docker.io/sourcegraph/node-exporter:4.4.2@sha256:fa8e5700b7762fffe0674e944762f44bb787a7e44d97569fe55348260453bf80
+          image: index.docker.io/sourcegraph/node-exporter:4.5.0@sha256:fa8e5700b7762fffe0674e944762f44bb787a7e44d97569fe55348260453bf80
           imagePullPolicy: IfNotPresent
           resources:
             limits:

base/monitoring/otel-collector/otel-agent.DaemonSet.yaml

@@ -26,7 +26,7 @@ spec:
     spec:
       containers:
         - name: otel-agent
-          image: index.docker.io/sourcegraph/opentelemetry-collector:4.4.2@sha256:f0723c96c973258ad3123ddc479261bb8f5827bbac1d091b6a683fde55334413
+          image: index.docker.io/sourcegraph/opentelemetry-collector:4.5.0@sha256:12f3fc137edea8319ebf574e15e6c27c19fb0b7ca17165973f98c8d8c342ca1d
           command:
             - "/bin/otelcol-sourcegraph"
             - "--config=/etc/otel-agent/config.yaml"

base/monitoring/otel-collector/otel-collector.Deployment.yaml

@@ -26,7 +26,7 @@ spec:
     spec:
       containers:
         - name: otel-collector
-          image: index.docker.io/sourcegraph/opentelemetry-collector:4.4.2@sha256:f0723c96c973258ad3123ddc479261bb8f5827bbac1d091b6a683fde55334413
+          image: index.docker.io/sourcegraph/opentelemetry-collector:4.5.0@sha256:12f3fc137edea8319ebf574e15e6c27c19fb0b7ca17165973f98c8d8c342ca1d
           command:
             - "/bin/otelcol-sourcegraph"
             # To use a custom configuration, edit otel-collector.ConfigMap.yaml

base/monitoring/prometheus/prometheus.ConfigMap.yaml

@@ -10,7 +10,7 @@ data:
   prometheus.yml: |
     # Prometheus global config
     global:
-      scrape_interval:     30s
+      scrape_interval: 30s
       evaluation_interval: 30s
       # scrape_timeout is set to the global default (10s).
 
@@ -19,47 +19,36 @@ data:
       alertmanagers:
         # bundled alertmanager, started by prom-wrapper
         - static_configs:
-            - targets: ['127.0.0.1:9093']
+            - targets: ["127.0.0.1:9093"]
           path_prefix: /alertmanager
         # add more alertmanagers here
 
     # Load rules once and periodically evaluate them according to the global 'evaluation_interval'.
     rule_files:
-      - '/sg_config_prometheus/*_rules.yml'
-      - '/sg_prometheus_add_ons/*_rules.yml'
+      - "/sg_config_prometheus/*_rules.yml"
+      - "/sg_prometheus_add_ons/*_rules.yml"
 
     # Configure targets to scrape
     scrape_configs:
-
       # Scrape prometheus itself for metrics.
-      - job_name: 'builtin-prometheus'
+      - job_name: "builtin-prometheus"
         static_configs:
-          - targets: ['127.0.0.1:9092']
+          - targets: ["127.0.0.1:9092"]
 
-      - job_name: 'builtin-alertmanager'
+      - job_name: "builtin-alertmanager"
         metrics_path: /alertmanager/metrics
         static_configs:
-          - targets: ['127.0.0.1:9093']
+          - targets: ["127.0.0.1:9093"]
 
-      - job_name: 'sourcegraph-services'
-        relabel_configs:
-          - source_labels: [__address__]
-            target_label: instance
-            regex: (.*)\.(.*)
-            replacement: ${1}_${2}
-        metric_relabel_configs:
-          - source_labels: [container_label_io_kubernetes_pod_namespace]
-            regex: kube-system
-            action: drop
-        file_sd_configs:
-          - files:
-              - '/sg_prometheus_add_ons/*_targets.yml'
-
-      - job_name: 'cadvisor'
+      ##########################################################################################
+      # cadvisor
+      ##########################################################################################
+
+      - job_name: "kubernetes-pods"
         dns_sd_configs:
           - names:
-            - 'cadvisor.default.svc.cluster.local'
-            - 'cadvisor.ns-sourcegraph.svc.cluster.local'
+              - "cadvisor.default.svc.cluster.local"
+              - "cadvisor.ns-sourcegraph.svc.cluster.local"
             type: A
             port: 48080
         relabel_configs:
@@ -73,59 +62,97 @@ data:
           - source_labels: [container_label_io_kubernetes_pod_namespace]
             regex: kube-system
             action: drop
-          - source_labels: [container_label_io_kubernetes_container_name, container_label_io_kubernetes_pod_name]
+          - source_labels:
+              [
+                container_label_io_kubernetes_container_name,
+                container_label_io_kubernetes_pod_name,
+              ]
             regex: (.+)
             action: replace
             target_label: name
-            separator: '-'
-          # - source_labels: [container_label_io_kubernetes_pod_namespace]
-          #   regex: ^$|ns-sourcegraph # ACTION: replace ns-sourcegraph with your namespace
-          #   action: keep
-
-      - job_name: 'sourcegraph-statefulsets'
+            separator: "-"
+
+      ##########################################################################################
+      # sourcegraph-services
+      ##########################################################################################
+
+      - job_name: "sourcegraph-services"
+        relabel_configs:
+          - source_labels: [__address__]
+            target_label: instance
+            regex: (.*)\.(.*)
+            replacement: ${1}_${2}
+        metric_relabel_configs:
+          - source_labels: [container_label_io_kubernetes_pod_namespace]
+            regex: kube-system
+            action: drop
+        file_sd_configs:
+          - files:
+              - "/sg_prometheus_add_ons/*_targets.yml"
+
+      - job_name: "sourcegraph-statefulsets"
         dns_sd_configs:
           - names:
-            - 'symbols.default.svc.cluster.local'
-            - 'symbols.ns-sourcegraph.svc.cluster.local'
-            - 'searcher.default.svc.cluster.local'
-            - 'searcher.ns-sourcegraph.svc.cluster.local'
-            - 'gitserver.default.svc.cluster.local'
-            - 'gitserver.ns-sourcegraph.svc.cluster.local'
-            - 'sourcegraph-frontend.default.svc.cluster.local'
-            - 'sourcegraph-frontend.ns-sourcegraph.svc.cluster.local'
-            type: A
-            port: 6060
-          - names:
-            - 'indexed-search.default.svc.cluster.local'
-            - 'indexed-search.ns-sourcegraph.svc.cluster.local'
-            type: A
-            port: 6070
-          - names:
-            - 'indexed-search-indexer.default.svc.cluster.local'
-            - 'indexed-search-indexer.ns-sourcegraph.svc.cluster.local'
-            type: A
-            port: 6072
+              - "symbols.default.svc.cluster.local"
+              - "symbols.ns-sourcegraph.svc.cluster.local"
+              - "symbols.$SG_NAMESPACE.svc.cluster.local"
+              - "searcher.default.svc.cluster.local"
+              - "searcher.ns-sourcegraph.svc.cluster.local"
+              - "searcher.$SG_NAMESPACE.svc.cluster.local"
+              - "gitserver.default.svc.cluster.local"
+              - "gitserver.ns-sourcegraph.svc.cluster.local"
+              - "gitserver.$SG_NAMESPACE.svc.cluster.local"
+              - "sourcegraph-frontend.default.svc.cluster.local"
+              - "sourcegraph-frontend.ns-sourcegraph.svc.cluster.local"
+              - "sourcegraph-frontend.$SG_NAMESPACE.svc.cluster.local"
+              - "indexed-search.default.svc.cluster.local"
+              - "indexed-search.ns-sourcegraph.svc.cluster.local"
+              - "indexed-search.$SG_NAMESPACE.svc.cluster.local"
+              - "indexed-search-indexer.default.svc.cluster.local"
+              - "indexed-search-indexer.ns-sourcegraph.svc.cluster.local"
+              - "indexed-search-indexer.$SG_NAMESPACE.svc.cluster.local"
+            type: SRV
         relabel_configs:
+          - source_labels: [__meta_dns_srv_record_target]
+            target_label: __address__
+            regex: (.*)\.
+            replacement: ${1}:6060
+          - source_labels: [__meta_dns_srv_record_target]
+            target_label: __address__
+            regex: ^(indexed-search.*)\.
+            replacement: ${1}:6070
+          - source_labels: [__meta_dns_srv_record_target]
+            target_label: __address__
+            regex: (.*)\.(indexed-search-indexer.*)\.
+            replacement: ${1}.${2}:6072
+          - source_labels: [__meta_dns_srv_record_port]
+            target_label: __meta_dns_srv_record_port
+            replacement: 6060
+          - source_labels: [__address__]
+            regex: ^(indexed-search).*$
+            target_label: __meta_dns_srv_record_port
+            replacement: 6070
           - source_labels: [__meta_dns_name]
-            target_label: service_name
+            target_label: job
             regex: (.*)\..*\..*\..*\..*
             replacement: ${1}
+          - source_labels: [__meta_dns_srv_record_target]
+            regex: (.*)\.(.*)\..*\..*\..*\..*\..*
+            target_label: instance
+            replacement: ${2}_${1}
+        metric_relabel_configs:
+          - source_labels: [container_label_io_kubernetes_pod_namespace]
+            regex: kube-system
+            action: drop
+          - source_labels: [__address__]
+            target_label: instance
+            regex: (.*)\:.*
+            replacement: $1:6060
+          - source_labels: [__address__]
+            target_label: instance
+            regex: (.*)\.(.*)\..*\..*\..*\..*\..*
+            replacement: ${2}_${1}
 
-  # Extra rules
-  extra_rules.yml: |
-    groups:
-    - name: container.rules
-      rules:
-        - record: container:process_cpu_seconds_total:ratio_rate5m
-          expr: sum by (instance) (rate(process_cpu_seconds_total[5m])) / engine_daemon_engine_cpus_cpus
-        - record: container:process_cpu_seconds_total:sum
-          expr: sum by (instance) (irate(process_cpu_seconds_total[1m]))
-        - record: container:process_resident_memory_bytes:max
-          expr: max by (instance) (process_resident_memory_bytes)
-        - record: container:process_virtual_memory_bytes:max
-          expr: max by (instance) (process_virtual_memory_bytes)
-
-  # List of static targets
   prometheus_targets.yml: |
     - labels:
         nodename: "sourcegraph-services"
@@ -202,40 +229,15 @@ data:
         job: otel-collector
       targets:
         - otel-collector:8888
-
-  # Add new targets based on replica count of symbols
-  symbols_targets.yml: |
-    - labels:
-        nodename: "sourcegraph-services"
-        job: symbols
-      targets:
-        - symbols-0.symbols:6060
-
-  # Add new targets based on replica count of searcher
-  searcher_targets.yml: |
-    - labels:
-        nodename: "sourcegraph-services"
-        job: searcher
-      targets:
-        - searcher-0.searcher:6060
-
-  # Add new targets based on replica count of gitserver
-  gitserver_targets.yml: |
-    - labels:
-        nodename: "sourcegraph-services"
-        job: gitserver
-      targets:
-        - gitserver-0.gitserver:6060
-
-  # Add new targets based on replica count of indexed-search
-  indexed-search_targets.yml: |
-    - labels:
-        nodename: "sourcegraph-services"
-        job: zoekt-indexserver
-      targets:
-        - indexed-search-0.indexed-search:6072
-    - labels:
-        nodename: "sourcegraph-services"
-        job: zoekt-webserver
-      targets:
-        - indexed-search-0.indexed-search:6070
+  extra_rules.yml: |
+    groups:
+    - name: container.rules
+      rules:
+        - record: container:process_cpu_seconds_total:ratio_rate5m
+          expr: sum by (instance) (rate(process_cpu_seconds_total[5m])) / engine_daemon_engine_cpus_cpus
+        - record: container:process_cpu_seconds_total:sum
+          expr: sum by (instance) (irate(process_cpu_seconds_total[1m]))
+        - record: container:process_resident_memory_bytes:max
+          expr: max by (instance) (process_resident_memory_bytes)
+        - record: container:process_virtual_memory_bytes:max
+          expr: max by (instance) (process_virtual_memory_bytes)

base/monitoring/prometheus/prometheus.Deployment.yaml

@@ -25,10 +25,10 @@ spec:
     spec:
       containers:
         - name: prometheus
-          image: index.docker.io/sourcegraph/prometheus:4.4.2@sha256:d833d00a39937cf700f276f816dc789615d6396979418a7d9362386513b1fc9d
+          image: index.docker.io/sourcegraph/prometheus:4.5.0@sha256:4fe9a5fdee206b1aac9d32afb31ad57e1882394aad9e7e9f719a1b2741afcae5
           terminationMessagePolicy: FallbackToLogsOnError
           env:
-            - name: MY_POD_NAMESPACE
+            - name: SG_NAMESPACE
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
@@ -70,7 +70,6 @@ spec:
         runAsUser: 100
         fsGroup: 100
         fsGroupChangePolicy: "OnRootMismatch"
-      # serviceAccountName: prometheus
       volumes:
         - name: data
           persistentVolumeClaim:

base/monitoring/prometheus/prometheus.PersistentVolumeClaim.yaml

@@ -11,4 +11,4 @@ spec:
     - ReadWriteOnce
   resources:
     requests:
-      storage: 50Gi
+      storage: 200Gi

base/monitoring/prometheus/rbac/kustomization.yaml

@@ -0,0 +1,7 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+resources:
+  - prometheus.ConfigMap.yaml
+  - prometheus.ClusterRole.yaml
+  - prometheus.ClusterRoleBinding.yaml
+  - prometheus.ServiceAccount.yaml