commit 8ec04145d3fbd2581314d0b831b2e737a03440a9
parent a455042fc7d554ba98b20beb800f584fdb7ac4a7
Author: MTRNord <MTRNord@users.noreply.github.com>
Date: Sat, 28 Jun 2025 20:41:12 +0200
Adjust the canary deployment a bit
Diffstat:
5 files changed, 114 insertions(+), 95 deletions(-)
diff --git a/apps/base/matrix/connectivity-tester/deployment.yaml b/apps/base/matrix/connectivity-tester/deployment.yaml
@@ -1,14 +1,13 @@
----
apiVersion: apps/v1
kind: Deployment
metadata:
name: connectivity-tester
namespace: matrix
spec:
+ progressDeadlineSeconds: 120
strategy:
rollingUpdate:
- maxSurge: 25%
- maxUnavailable: 25%
+ maxUnavailable: 0
type: RollingUpdate
selector:
matchLabels:
@@ -21,11 +20,11 @@ spec:
podAntiAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
- matchExpressions:
- - key: app
- operator: In
- values:
- - connectivity-tester
+ matchExpressions:
+ - key: app
+ operator: In
+ values:
+ - connectivity-tester
topologyKey: "kubernetes.io/hostname"
imagePullSecrets:
- name: ghcr-pull
@@ -35,62 +34,78 @@ spec:
image: ghcr.io/mtrnord/rust-federation-tester:main
imagePullPolicy: Always
resources:
- limits: {}
- requests:
- memory: "344Mi"
- cpu: "252m"
+ limits: {}
+ requests:
+ memory: "344Mi"
+ cpu: "252m"
securityContext:
- runAsUser: 1000
- runAsGroup: 1000
- readOnlyRootFilesystem: true
+ runAsUser: 1000
+ runAsGroup: 1000
+ readOnlyRootFilesystem: true
ports:
- - containerPort: 8080
- name: api
- protocol: TCP
+ - containerPort: 8080
+ name: api
+ protocol: TCP
+ readinessProbe:
+ httpGet:
+ path: /healthz
+ port: api
+ scheme: HTTP
livenessProbe:
- httpGet:
- path: /healthz
- port: api
- scheme: HTTP
+ httpGet:
+ path: /healthz
+ port: api
+ scheme: HTTP
startupProbe:
- httpGet:
- path: /healthz
- port: api
+ httpGet:
+ path: /healthz
+ port: api
- name: connection-checker-ui
image: ghcr.io/mtrnord/matrix-connection-tester-ui:main
imagePullPolicy: Always
+ lifecycle:
+ preStop:
+ exec:
+ command:
+ - sleep
+ - "10"
resources:
- limits: {}
- requests:
- memory: "344Mi"
- cpu: "252m"
+ limits: {}
+ requests:
+ memory: "344Mi"
+ cpu: "252m"
ports:
- - containerPort: 3000
- name: web
- protocol: TCP
+ - containerPort: 3000
+ name: web
+ protocol: TCP
volumeMounts:
- - name: configs
- mountPath: "/usr/share/nginx/html/config.json"
- subPath: config.json
- readOnly: true
- - mountPath: /tmp
- name: tmp
+ - name: configs
+ mountPath: "/usr/share/nginx/html/config.json"
+ subPath: config.json
+ readOnly: true
+ - mountPath: /tmp
+ name: tmp
livenessProbe:
- httpGet:
- path: /
- port: web
- scheme: HTTP
+ httpGet:
+ path: /
+ port: web
+ scheme: HTTP
+ readinessProbe:
+ httpGet:
+ path: /
+ port: web
+ scheme: HTTP
startupProbe:
- httpGet:
- path: /
- port: web
+ httpGet:
+ path: /
+ port: web
volumes:
- name: configs
configMap:
- name: connectivity-tester-config
+ name: connectivity-tester-config
- name: tmp
emptyDir:
- sizeLimit: 2048Mi
+ sizeLimit: 2048Mi
---
apiVersion: flagger.app/v1beta1
kind: MetricTemplate
@@ -172,6 +187,10 @@ spec:
gatewayRefs:
- name: envoy-gateway
namespace: envoy-gateway
+ retries:
+ attempts: 10
+ perTryTimeout: 5s
+ retryOn: "gateway-error,connect-failure,refused-stream"
analysis:
# schedule interval (default 60s)
interval: 1m
@@ -188,18 +207,18 @@ spec:
# max error rate (5xx responses)
# percentage (0-100)
templateRef:
- name: connectivity-tester-error-rate
- namespace: matrix
+ name: connectivity-tester-error-rate
+ namespace: matrix
thresholdRange:
- max: 1
+ max: 1
interval: 1m
- name: latency
templateRef:
- name: connectivity-tester-latency
- namespace: matrix
+ name: connectivity-tester-latency
+ namespace: matrix
# seconds
thresholdRange:
- max: 0.5
+ max: 0.5
interval: 30s
# testing (optional)
webhooks:
@@ -208,10 +227,11 @@ spec:
url: http://flagger-loadtester.flagger-system/
timeout: 15s
metadata:
- type: bash
- cmd: "curl -s http://connectivity-tester.matrix:3000 && curl -s http://connectivity-tester.matrix:8080/healthz | grep 'OK'"
+ type: bash
+ cmd: "curl -s http://connectivity-tester.matrix:3000 && curl -s http://connectivity-tester.matrix:8080/healthz
+ | grep 'OK'"
- name: load-test
url: http://flagger-loadtester.flagger-system/
timeout: 5s
metadata:
- cmd: "hey -z 2m -q 10 -c 2 -host connectivity-tester.mtrnord.blog http://connectivity-tester.matrix:3000/"
+ cmd: "hey -z 2m -q 10 -c 2 -host connectivity-tester.mtrnord.blog http://connectivity-tester.matrix:3000/"
diff --git a/apps/base/matrix/connectivity-tester/kustomization.yaml b/apps/base/matrix/connectivity-tester/kustomization.yaml
@@ -8,4 +8,4 @@ configMapGenerator:
- name: connectivity-tester-config
namespace: matrix
files:
- - config.json=config.json
+ - config.json=config.json
diff --git a/apps/base/matrix/connectivity-tester/service.yaml b/apps/base/matrix/connectivity-tester/service.yaml
@@ -28,12 +28,12 @@ spec:
- federationtester.mtrnord.blog
rules:
- matches:
- - path:
- value: /api
- type: PathPrefix
+ - path:
+ value: /api
+ type: PathPrefix
backendRefs:
- - name: connectivity-tester
- port: 8080
+ - name: connectivity-tester
+ port: 8080
timeouts:
- request: 240s
- backendRequest: 0s
+ request: 240s
+ backendRequest: 0s
diff --git a/apps/base/victoria-metrics-k8s-stack/dashboards/kustomization.yaml b/apps/base/victoria-metrics-k8s-stack/dashboards/kustomization.yaml
@@ -5,45 +5,45 @@ configMapGenerator:
- name: grafana-dashboards-matrix
namespace: monitoring
files:
- - draupnir4all_status.json=draupnir4all_status.json
- - synapse.json=synapse.json
+ - draupnir4all_status.json=draupnir4all_status.json
+ - synapse.json=synapse.json
options:
- labels:
- grafana_dashboard: "1"
- annotations:
- grafana_folder: Matrix
+ labels:
+ grafana_dashboard: "1"
+ annotations:
+ grafana_folder: Matrix
- name: grafana-dashboards-flux
namespace: monitoring
files:
- - flux-control-plane-dashboard.json=flux-control-plane-dashboard.json
- - flux-cluster-dashboard.json=flux-cluster-dashboard.json
- - flux-logs-dashboard.json=flux-logs-dashboard.json
- - flagger.json=flagger.json
+ - flux-control-plane-dashboard.json=flux-control-plane-dashboard.json
+ - flux-cluster-dashboard.json=flux-cluster-dashboard.json
+ - flux-logs-dashboard.json=flux-logs-dashboard.json
+ - flagger.json=flagger.json
options:
- labels:
- grafana_dashboard: "1"
- annotations:
- grafana_folder: Flux
+ labels:
+ grafana_dashboard: "1"
+ annotations:
+ grafana_folder: Flux
- name: grafana-dashboards-ceph
namespace: monitoring
files:
- - ceph-cluster-dashboard.json=ceph-cluster-dashboard.json
- - ceph-osd-single-dashboard.json=ceph-osd-single-dashboard.json
- - ceph-pools-dashboard.json=ceph-pools-dashboard.json
+ - ceph-cluster-dashboard.json=ceph-cluster-dashboard.json
+ - ceph-osd-single-dashboard.json=ceph-osd-single-dashboard.json
+ - ceph-pools-dashboard.json=ceph-pools-dashboard.json
options:
- labels:
- grafana_dashboard: "1"
- annotations:
- grafana_folder: Ceph
+ labels:
+ grafana_dashboard: "1"
+ annotations:
+ grafana_folder: Ceph
- name: grafana-dashboards-envoy
namespace: monitoring
files:
- - envoy-clusters.json=envoy-clusters.json
- - envoy-gateway-global.json=envoy-gateway-global.json
- - envoy-proxy-global.json=envoy-proxy-global.json
- - global-ratelimit.json=global-ratelimit.json
+ - envoy-clusters.json=envoy-clusters.json
+ - envoy-gateway-global.json=envoy-gateway-global.json
+ - envoy-proxy-global.json=envoy-proxy-global.json
+ - global-ratelimit.json=global-ratelimit.json
options:
- labels:
- grafana_dashboard: "1"
- annotations:
- grafana_folder: Envoy
+ labels:
+ grafana_dashboard: "1"
+ annotations:
+ grafana_folder: Envoy
diff --git a/apps/base/victoria-metrics-k8s-stack/pg_dashboard_configmap.yaml b/apps/base/victoria-metrics-k8s-stack/pg_dashboard_configmap.yaml
@@ -7,8 +7,7 @@ metadata:
annotations:
grafana_folder: Postgres
data:
- postgres.json:
- "{\n\"annotations\": {\n \"list\": [\n {\n \"builtIn\": 1,\n \"datasource\": \"-- Grafana
+ postgres.json: "{\n\"annotations\": {\n \"list\": [\n {\n \"builtIn\": 1,\n \"datasource\": \"-- Grafana
--\",\n \"enable\": true,\n \"hide\": true,\n \"iconColor\": \"rgba(0, 211, 255, 1)\",\n \"name\":
\"Annotations & Alerts\",\n \"type\": \"dashboard\"\n }\n ]\n},\n\"description\": \"Performance metrics
for Postgres\",\n\"editable\": true,\n\"gnetId\": 455,\n\"graphTooltip\": 0,\n\"id\": 1,\n\"iteration\": 1603191461722,\n\"links\":
