use both service and pod metrics (#1108)

nuclia · Jul 17, 2023 · c0e1f1f · c0e1f1f · github-actions · Jul 17, 2023
1 parent 7497f57
commit c0e1f1f
Show file tree

Hide file tree

Showing 15 changed files with 71 additions and 7 deletions.
diff --git a/charts/nucliadb_ingest/templates/ingest-headless.svc.yaml b/charts/nucliadb_ingest/templates/ingest-headless.svc.yaml
@@ -23,3 +23,7 @@ spec:
       port: {{ .Values.chitchat.cluster_manager.port }}
       protocol: TCP
       appProtocol: tcp
+    - name: metrics
+      port: {{ .Values.serving.metricsPort}}
+      protocol: TCP
+      appProtocol: http
diff --git a/charts/nucliadb_ingest/templates/ingest-orm-grpc.deploy.yaml b/charts/nucliadb_ingest/templates/ingest-orm-grpc.deploy.yaml
@@ -26,7 +26,6 @@ spec:
         checksum/cm: {{ include (print $.Template.BasePath "/ingest.cm.yaml") . | sha256sum }}
       labels:
         app: ingest-orm-grpc
-        metrics: "enabled"
         version: "{{ .Chart.Version | replace "+" "_" }}"
         chart: "{{ .Chart.Name }}"
         release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_ingest/templates/ingest-orm-grpc.svc.yaml b/charts/nucliadb_ingest/templates/ingest-orm-grpc.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: ingest-orm-grpc
   labels:
     app: ingest-orm-grpc
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"
@@ -21,3 +22,7 @@ spec:
       port: {{ .Values.chitchat.cluster_manager.port }}
       protocol: TCP
       appProtocol: tcp
+    - name: metrics
+      port: {{ .Values.serving.metricsPort}}
+      protocol: TCP
+      appProtocol: http
diff --git a/charts/nucliadb_ingest/templates/ingest.svc.yaml b/charts/nucliadb_ingest/templates/ingest.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: ingest
   labels:
     app: ingest
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"
@@ -21,3 +22,7 @@ spec:
       port: {{ .Values.chitchat.cluster_manager.port }}
       protocol: TCP
       appProtocol: tcp
+    - name: metrics
+      port: {{ .Values.serving.metricsPort}}
+      protocol: TCP
+      appProtocol: http
diff --git a/charts/nucliadb_node/templates/node.sm.yaml b/charts/nucliadb_node/templates/node.sm.yaml
@@ -0,0 +1,32 @@
+{{- if .Values.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: node
+  labels:
+    app: node
+    version: "{{ .Chart.Version | replace "+" "_" }}"
+    chart: "{{ .Chart.Name }}"
+    release: "{{ .Release.Name }}"
+    heritage: "{{ .Release.Service }}"
+spec:
+  namespaceSelector:
+    matchNames:
+    - "{{ .Release.Namespace }}"
+  selector:
+    matchLabels:
+      app: node
+  endpoints:
+  # node sidecar metrics
+  - targetPort: {{ .Values.serving.metricsPort }}
+    interval: 10s
+    path: /metrics
+  # reader
+  - targetPort: 3031
+    interval: 10s
+    path: /metrics
+  # writer
+  - targetPort: 3032
+    interval: 10s
+    path: /metrics
+{{- end }}
diff --git a/charts/nucliadb_node/templates/node.sts.yaml b/charts/nucliadb_node/templates/node.sts.yaml
@@ -26,13 +26,12 @@ spec:
     metadata:
       annotations:
         traffic.sidecar.istio.io/excludeOutboundPorts: "{{.Values.nats.port }}"
-        traffic.sidecar.istio.io/excludeInboundPorts: "{{.Values.serving.metricsPort }}"
+        traffic.sidecar.istio.io/excludeInboundPorts: "{{.Values.serving.metricsPort }},3031,3032"
         # do not have access to dependency chart cm this component references
         checksum/cm: {{ include (print $.Template.BasePath "/node.cm.yaml") . | sha256sum }}
       name: node
       labels:
         app: node
-        metrics: "enabled"
         app.kubernetes.io/name: node
         app.kubernetes.io/instance: {{ .Release.Name }}
         app.kubernetes.io/managed-by: {{ .Release.Service }}

diff --git a/charts/nucliadb_reader/templates/reader.deploy.yaml b/charts/nucliadb_reader/templates/reader.deploy.yaml
@@ -26,7 +26,6 @@ spec:
         checksum/cm: {{ include (print $.Template.BasePath "/reader.cm.yaml") . | sha256sum }}
       labels:
         app: reader
-        metrics: "enabled"
         version: "{{ .Chart.Version | replace "+" "_" }}"
         chart: "{{ .Chart.Name }}"
         release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_reader/templates/reader.svc.yaml b/charts/nucliadb_reader/templates/reader.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: reader
   labels:
     app: reader
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_search/templates/search.deploy.yaml b/charts/nucliadb_search/templates/search.deploy.yaml
@@ -26,7 +26,6 @@ spec:
         checksum/cm: {{ include (print $.Template.BasePath "/search.cm.yaml") . | sha256sum }}
       labels:
         app: search
-        metrics: "enabled"
         version: "{{ .Chart.Version | replace "+" "_" }}"
         chart: "{{ .Chart.Name }}"
         release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_search/templates/search.svc.yaml b/charts/nucliadb_search/templates/search.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: search
   labels:
     app: search
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_shared/templates/nucliadb.sm.yaml b/charts/nucliadb_shared/templates/nucliadb.sm.yaml
@@ -0,0 +1,20 @@
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: ndb-sm
+  labels:
+    version: "{{ .Chart.Version | replace "+" "_" }}"
+    chart: "{{ .Chart.Name }}"
+    release: "{{ .Release.Name }}"
+    heritage: "{{ .Release.Service }}"
+spec:
+  namespaceSelector:
+    matchNames:
+    - "{{ .Release.Namespace }}"
+  selector:
+    matchLabels:
+      metrics: "enabled"
+  endpoints:
+  - port: metrics
+    interval: 10s
+    path: /metrics
diff --git a/charts/nucliadb_train/templates/train.deploy.yaml b/charts/nucliadb_train/templates/train.deploy.yaml
@@ -26,7 +26,6 @@ spec:
         checksum/cm: {{ include (print $.Template.BasePath "/train.cm.yaml") . | sha256sum }}
       labels:
         app: train
-        metrics: "enabled"
         version: "{{ .Chart.Version | replace "+" "_" }}"
         chart: "{{ .Chart.Name }}"
         release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_train/templates/train.svc.yaml b/charts/nucliadb_train/templates/train.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: train
   labels:
     app: train
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_writer/templates/writer.deploy.yaml b/charts/nucliadb_writer/templates/writer.deploy.yaml
@@ -21,7 +21,6 @@ spec:
       name: writer
       labels:
         app: writer
-        metrics: "enabled"
         version: "{{ .Chart.Version | replace "+" "_" }}"
         chart: "{{ .Chart.Name }}"
         release: "{{ .Release.Name }}"

diff --git a/charts/nucliadb_writer/templates/writer.svc.yaml b/charts/nucliadb_writer/templates/writer.svc.yaml
@@ -4,6 +4,7 @@ metadata:
   name: writer
   labels:
     app: writer
+    metrics: "enabled"
     version: "{{ .Chart.Version | replace "+" "_" }}"
     chart: "{{ .Chart.Name }}"
     release: "{{ .Release.Name }}"
Benchmark suite	Current: `c0e1f1f`	Previous: `a40981d`	Ratio
`nucliadb/tests/benchmarks/test_search.py::test_search_returns_labels`	`51.29825067787303` iter/sec (`stddev: 0.0008826635384177155`)	`60.779932309336715` iter/sec (`stddev: 0.0019119907918232523`)	`1.18`
`nucliadb/tests/benchmarks/test_search.py::test_search_relations`	`136.80337750037836` iter/sec (`stddev: 0.0004377745409201615`)	`182.57436721258293` iter/sec (`stddev: 0.0002220745559283828`)	`1.33`