OPSEXP-2253 Add healthchecks and conditional service startup based on service health in docker compose

.github/workflows/docker-compose-community.yml

@@ -22,6 +22,8 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - 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

.github/workflows/docker-compose-enterprise.yml

@@ -55,9 +55,11 @@ jobs:
     steps:
       - 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 }}

docker-compose/7.1.N-compose.yaml

@@ -56,6 +56,12 @@ 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(`/`)"
@@ -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: 10s
+      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:
@@ -127,6 +157,15 @@ 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`)"
@@ -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
@@ -168,6 +213,12 @@ 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
@@ -183,6 +234,12 @@ services:
       - "traefik.http.middlewares.adwforceslash.redirectregex.replacement=$${1}/"
       - "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 +250,17 @@ 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 +283,17 @@ services:
         -XX:MaxRAM=1g
     ports:
       - "9090:9090"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:9090/alfresco/healthcheck"]
+      interval: 30s
+      timeout: 2s
+      retries: 5
+      start_period: 40s
+    depends_on:
+      postgres:
+        condition: service_healthy
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.syncservice.rule=PathPrefix(`/syncservice`)"

docker-compose/7.2.N-compose.yaml

@@ -57,6 +57,12 @@ 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(`/`)"
@@ -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
@@ -112,6 +136,12 @@ services:
       - "8099:8099"
     volumes:
       - shared-file-store-volume:/tmp/Alfresco/sfs
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8099/ready"]
+      interval: 20s
+      timeout: 2s
+      retries: 3
+      start_period: 10s
   share:
     image: quay.io/alfresco/alfresco-share:7.2.2.5
     mem_limit: 1g
@@ -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
@@ -172,13 +217,25 @@ 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_AUTH_TYPE: "BASIC"
       BASE_PATH: ./
       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`)"
@@ -195,6 +252,12 @@ 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 +276,17 @@ 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 +309,17 @@ services:
         -XX:MaxRAM=1g
     ports:
       - "9090:9090"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:9090/alfresco/healthcheck"]
+      interval: 30s
+      timeout: 2s
+      retries: 5
+      start_period: 40s
+    depends_on:
+      postgres:
+        condition: service_healthy
+      alfresco:
+        condition: service_healthy
     labels:
       - "traefik.enable=true"
       - "traefik.http.routers.syncservice.rule=PathPrefix(`/syncservice`)"