OPSEXP-2253 Add healthchecks and conditional service startup based on…

… service health in docker compose (#1232) Co-authored-by: Giovanni Toraldo <[email protected]>
Alfresco · Oct 25, 2024 · 78132d9 · 78132d9
1 parent ea127d9
commit 78132d9
Show file tree

Hide file tree

Showing 9 changed files with 638 additions and 57 deletions.
diff --git a/.github/workflows/docker-compose-community.yml b/.github/workflows/docker-compose-community.yml
@@ -21,7 +21,10 @@ jobs:
     name: Docker Compose community
     runs-on: ubuntu-latest
     steps:
+      - uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938  # v4.2.0
       - uses: >-
-          Alfresco/alfresco-build-tools/.github/actions/dbp-charts/verify-compose@v7.0.0
+          Alfresco/alfresco-build-tools/.github/actions/dbp-charts/verify-compose@v8.1.0
         with:
           compose_file_path: docker-compose/community-compose.yaml
+          postman_path: test/postman/docker-compose
+          postman_json: acs-test-docker-compose-collection.json
diff --git a/.github/workflows/docker-compose-enterprise.yml b/.github/workflows/docker-compose-enterprise.yml
@@ -53,11 +53,14 @@ jobs:
         && github.event.pull_request.head.user.login == 'Alfresco'
       )
     steps:
+      - uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938  # v4.2.0
       - name: Verify docker-compose
         uses: >-
-          Alfresco/alfresco-build-tools/.github/actions/dbp-charts/verify-compose@v7.0.0
+          Alfresco/alfresco-build-tools/.github/actions/dbp-charts/verify-compose@v8.1.0
         timeout-minutes: 10
         with:
+          postman_path: test/postman/docker-compose
+          postman_json: acs-test-docker-compose-collection.json
           compose_file_path: docker-compose/${{ matrix.compose_file }}
           quay_username: ${{ secrets.QUAY_USERNAME }}
           quay_password: ${{ secrets.QUAY_PASSWORD }}
diff --git a/docker-compose/7.1.N-compose.yaml b/docker-compose/7.1.N-compose.yaml
@@ -56,11 +56,17 @@ services:
         -XX:MinRAMPercentage=50
         -XX:MaxRAMPercentage=80
         -XX:MaxRAM=1900m
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/alfresco/api/-default-/public/alfresco/versions/1/probes/-ready-"]
+      interval: 30s
+      timeout: 3s
+      retries: 5
+      start_period: 1m
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.alfresco.rule=PathPrefix(`/`)"
       - "traefik.http.services.alfresco.loadbalancer.server.port=8080"
-      - "traefik.http.routers.solrapideny.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/api/solr/`)"
+      - "traefik.http.routers.solrapideny.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/api/solr/.*$`)"
       - "traefik.http.middlewares.acsfakeauth.basicauth.users=fake:"
       - "traefik.http.routers.solrapideny.middlewares=acsfakeauth@docker"
       - "traefik.http.routers.alfrescomicrometer.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/prometheus`)"
@@ -80,8 +86,17 @@ services:
         http://shared-file-store:8099/alfresco/api/-default-/private/sfs/versions/1/file
     ports:
       - "8095:8095"
-    links:
-      - activemq
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8095/actuator/health"]
+      interval: 10s
+      timeout: 2s
+      retries: 3
+      start_period: 20s
+    depends_on:
+      activemq:
+        condition: service_healthy
+      shared-file-store:
+        condition: service_healthy
   transform-core-aio:
     image: alfresco/alfresco-transform-core-aio:3.1.2
     mem_limit: 1536m
@@ -95,8 +110,17 @@ services:
         http://shared-file-store:8099/alfresco/api/-default-/private/sfs/versions/1/file
     ports:
       - "8090:8090"
-    links:
-      - activemq
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8090/ready"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 10s
+    depends_on:
+      activemq:
+        condition: service_healthy
+      shared-file-store:
+        condition: service_healthy
   shared-file-store:
     image: quay.io/alfresco/alfresco-shared-file-store:2.1.2
     mem_limit: 512m
@@ -109,6 +133,12 @@ services:
       scheduler.cleanup.interval: 86400000
     ports:
       - "8099:8099"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8099/ready"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 10s
     volumes:
       - shared-file-store-volume:/tmp/Alfresco/sfs
   share:
@@ -127,11 +157,20 @@ services:
         -Dalfresco.port=8080
         -Dalfresco.context=alfresco
         -Dalfresco.protocol=http
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/share"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 15s
+    depends_on:
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.share.rule=PathPrefix(`/share`)"
       - "traefik.http.services.share.loadbalancer.server.port=8080"
-      - "traefik.http.routers.proxiedsolrapideny.rule=PathRegexp(`^/share/proxy/alfresco(-(noauth|feed|api))?/api/solr/.*$`)"
+      - "traefik.http.routers.proxiedsolrapideny.rule=PathRegexp(`^/share/proxy/alfresco(-(noauth|feed|api))?/api/solr/`)"
       - "traefik.http.middlewares.sharefakeauth.basicauth.users=fake:"
       - "traefik.http.routers.proxiedsolrapideny.middlewares=sharefakeauth@docker"
   postgres:
@@ -144,6 +183,12 @@ services:
     command: postgres -c max_connections=300 -c log_min_messages=LOG
     ports:
       - "5432:5432"
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -d $$POSTGRES_DB -U $$POSTGRES_USER"]
+      interval: 10s
+      timeout: 3s
+      retries: 3
+      start_period: 5s
   solr6:
     image: quay.io/alfresco/search-services:2.0.2.2
     mem_limit: 2g
@@ -168,21 +213,36 @@ services:
       - "5672:5672" # AMQP
       - "61616:61616" # OpenWire
       - "61613:61613" # STOMP
+    healthcheck:
+      test: ["CMD", "/opt/activemq/bin/activemq", "query", "--objname", "type=Broker,brokerName=*,service=Health", "|", "grep", "Good"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 5s
   digital-workspace:
     image: quay.io/alfresco/alfresco-digital-workspace:4.4.1
     mem_limit: 128m
     environment:
+      APP_CONFIG_PROVIDER: "ECM"
       APP_CONFIG_AUTH_TYPE: "BASIC"
       BASE_PATH: ./
+      APP_CONFIG_PLUGIN_PROCESS_SERVICE: "false"
+      APP_CONFIG_PLUGIN_MICROSOFT_ONLINE: "false"
       APP_BASE_SHARE_URL: "http://localhost:8080/workspace/#/preview/s"
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.adw.rule=PathPrefix(`/workspace`)"
-      - "traefik.http.middlewares.adwroot.stripprefix.prefixes=/workspace"
       - "traefik.http.middlewares.adwforceslash.redirectregex.regex=^(.*/workspace)$$"
       - "traefik.http.middlewares.adwforceslash.redirectregex.replacement=$${1}/"
+      - "traefik.http.middlewares.adwroot.stripprefix.prefixes=/workspace"
       - "traefik.http.middlewares.adwchain.chain.middlewares=adwforceslash,adwroot"
       - "traefik.http.routers.adw.middlewares=adwchain@docker"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
+      interval: 10s
+      timeout: 1s
+      retries: 3
+      start_period: 1s
   proxy:
     image: traefik:3.1
     mem_limit: 128m
@@ -193,9 +253,17 @@ services:
       - "--entryPoints.traefik.address=:8888"
       - "--accesslog=true"
       - "--providers.docker.exposedByDefault=false"
+      - "--ping.entrypoint=web"
+      - "--ping=true"
     ports:
       - "8080:8080"
       - "8888:8888"
+    healthcheck:
+      test: ["CMD", "traefik", "healthcheck", "--ping"]
+      interval: 20s
+      timeout: 10s
+      retries: 3
+      start_period: 10s
     security_opt:
       - label=disable # Required for accessing the Docker socket on Selinux enabled systems
     volumes:
@@ -218,6 +286,17 @@ services:
         -XX:MaxRAM=1g
     ports:
       - "9090:9090"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:9090/alfresco/healthcheck"]
+      interval: 30s
+      timeout: 2s
+      retries: 3
+      start_period: 20s
+    depends_on:
+      postgres:
+        condition: service_healthy
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.syncservice.rule=PathPrefix(`/syncservice`)"

diff --git a/docker-compose/7.2.N-compose.yaml b/docker-compose/7.2.N-compose.yaml
@@ -57,11 +57,17 @@ services:
         -XX:MinRAMPercentage=50
         -XX:MaxRAMPercentage=80
         -XX:MaxRAM=1900m
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/alfresco/api/-default-/public/alfresco/versions/1/probes/-ready-"]
+      interval: 30s
+      timeout: 3s
+      retries: 5
+      start_period: 1m
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.alfresco.rule=PathPrefix(`/`)"
       - "traefik.http.services.alfresco.loadbalancer.server.port=8080"
-      - "traefik.http.routers.solrapideny.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/api/solr/`)"
+      - "traefik.http.routers.solrapideny.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/api/solr/.*$`)"
       - "traefik.http.middlewares.acsfakeauth.basicauth.users=fake:"
       - "traefik.http.routers.solrapideny.middlewares=acsfakeauth@docker"
       - "traefik.http.routers.alfrescomicrometer.rule=PathRegexp(`^/alfresco/(wc)?s(ervice)?/prometheus`)"
@@ -81,8 +87,17 @@ services:
         http://shared-file-store:8099/alfresco/api/-default-/private/sfs/versions/1/file
     ports:
       - "8095:8095"
-    links:
-      - activemq
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8095/actuator/health"]
+      interval: 10s
+      timeout: 2s
+      retries: 3
+      start_period: 20s
+    depends_on:
+      activemq:
+        condition: service_healthy
+      shared-file-store:
+        condition: service_healthy
   transform-core-aio:
     image: alfresco/alfresco-transform-core-aio:3.1.2
     mem_limit: 1536m
@@ -96,8 +111,17 @@ services:
         http://shared-file-store:8099/alfresco/api/-default-/private/sfs/versions/1/file
     ports:
       - "8090:8090"
-    links:
-      - activemq
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8090/ready"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 10s
+    depends_on:
+      activemq:
+        condition: service_healthy
+      shared-file-store:
+        condition: service_healthy
   shared-file-store:
     image: quay.io/alfresco/alfresco-shared-file-store:2.1.2
     mem_limit: 512m
@@ -110,6 +134,12 @@ services:
       scheduler.cleanup.interval: 86400000
     ports:
       - "8099:8099"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8099/ready"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 10s
     volumes:
       - shared-file-store-volume:/tmp/Alfresco/sfs
   share:
@@ -128,6 +158,15 @@ services:
         -Dalfresco.port=8080
         -Dalfresco.context=alfresco
         -Dalfresco.protocol=http
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/share"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 15s
+    depends_on:
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.share.rule=PathPrefix(`/share`)"
@@ -145,6 +184,12 @@ services:
     command: postgres -c max_connections=300 -c log_min_messages=LOG
     ports:
       - "5432:5432"
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -d $$POSTGRES_DB -U $$POSTGRES_USER"]
+      interval: 10s
+      timeout: 3s
+      retries: 3
+      start_period: 5s
   solr6:
     image: quay.io/alfresco/search-services:2.0.12
     mem_limit: 2g
@@ -172,17 +217,31 @@ services:
       - "5672:5672" # AMQP
       - "61616:61616" # OpenWire
       - "61613:61613" # STOMP
+    healthcheck:
+      test: ["CMD", "/opt/activemq/bin/activemq", "query", "--objname", "type=Broker,brokerName=*,service=Health", "|", "grep", "Good"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 5s
   digital-workspace:
     image: quay.io/alfresco/alfresco-digital-workspace:4.4.1
     mem_limit: 128m
     environment:
+      APP_CONFIG_PROVIDER: "ECM"
       APP_CONFIG_AUTH_TYPE: "BASIC"
       BASE_PATH: ./
+      APP_CONFIG_PLUGIN_PROCESS_SERVICE: "false"
+      APP_CONFIG_PLUGIN_MICROSOFT_ONLINE: "false"
       APP_BASE_SHARE_URL: "http://localhost:8080/workspace/#/preview/s"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
+      interval: 10s
+      timeout: 1s
+      retries: 3
+      start_period: 1s
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.adw.rule=PathPrefix(`/workspace`)"
-      - "traefik.http.middlewares.adfroot.stripprefix.prefixes=/workspace"
       - "traefik.http.middlewares.adwforceslash.redirectregex.regex=^(.*/workspace)$$"
       - "traefik.http.middlewares.adwforceslash.redirectregex.replacement=$${1}/"
       - "traefik.http.middlewares.adwroot.stripprefix.prefixes=/workspace"
@@ -195,6 +254,12 @@ services:
       APP_CONFIG_PROVIDER: "ECM"
       APP_CONFIG_AUTH_TYPE: "BASIC"
       BASE_PATH: ./
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
+      interval: 10s
+      timeout: 1s
+      retries: 3
+      start_period: 1s
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.acc.rule=PathPrefix(`/control-center`) || PathPrefix(`/admin`)"
@@ -213,9 +278,17 @@ services:
       - "--entryPoints.traefik.address=:8888"
       - "--accesslog=true"
       - "--providers.docker.exposedByDefault=false"
+      - "--ping.entrypoint=web"
+      - "--ping=true"
     ports:
       - "8080:8080"
       - "8888:8888"
+    healthcheck:
+      test: ["CMD", "traefik", "healthcheck", "--ping"]
+      interval: 20s
+      timeout: 10s
+      retries: 3
+      start_period: 10s
     security_opt:
       - label=disable # Required for accessing the Docker socket on Selinux enabled systems
     volumes:
@@ -238,6 +311,17 @@ services:
         -XX:MaxRAM=1g
     ports:
       - "9090:9090"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:9090/alfresco/healthcheck"]
+      interval: 30s
+      timeout: 2s
+      retries: 3
+      start_period: 20s
+    depends_on:
+      postgres:
+        condition: service_healthy
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.syncservice.rule=PathPrefix(`/syncservice`)"