feat: added healthcheck on jobs

Closes #60
kalisio · May 15, 2019 · bdcb19e · bdcb19e
1 parent 73ac290
commit bdcb19e
Show file tree

Hide file tree

Showing 11 changed files with 76 additions and 3 deletions.
diff --git a/.env.sample b/.env.sample
@@ -9,6 +9,7 @@ AUTH_USER=your authentication identity to access the services
 AUTH_PASSWORD=your authentication password to access the services. It must be encrypted in SHA1
 SECRET=the secret key to generete JWT. It used by the applications KANO/WEACAST
 DB_URL=mongodb://mongodb:27017/kargo
+SLACK_WEBHOOK_URL=url to post alerts on slack
 
 # S3 configuration
 S3_ACCESS_KEY=your access key id to access S3

diff --git a/deploy/arome-france-high-loader.yml b/deploy/arome-france-high-loader.yml
@@ -5,6 +5,14 @@ services:
     image: ${AROME_FRANCE_HIGH_LOADER_IMAGE}:${AROME_FRANCE_HIGH_LOADER_TAG}
     environment:
       - DB_URL=${WEACAST_LOADERS_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--success-rate", "0.75", --link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     networks:
       - kargo-network
     deploy:

diff --git a/deploy/arome-france-loader.yml b/deploy/arome-france-loader.yml
@@ -5,6 +5,14 @@ services:
     image: ${AROME_FRANCE_LOADER_IMAGE}:${AROME_FRANCE_LOADER_TAG}
     environment:
       - DB_URL=${WEACAST_LOADERS_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--success-rate", "0.75", --link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     networks:
       - kargo-network
     deploy:

diff --git a/deploy/arpege-europe-loader.yml b/deploy/arpege-europe-loader.yml
@@ -5,6 +5,14 @@ services:
     image: ${ARPEGE_EUROPE_LOADER_IMAGE}:${ARPEGE_EUROPE_LOADER_TAG}
     environment:
       - DB_URL=${WEACAST_LOADERS_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--success-rate", "0.75", --link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     networks:
       - kargo-network
     deploy:

diff --git a/deploy/arpege-world-loader.yml b/deploy/arpege-world-loader.yml
@@ -5,6 +5,14 @@ services:
     image: ${ARPEGE_WORLD_LOADER_IMAGE}:${ARPEGE_WORLD_LOADER_TAG}
     environment:
       - DB_URL=${WEACAST_LOADERS_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--success-rate", "0.75", --link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     networks:
       - kargo-network
     deploy:

diff --git a/deploy/mongodb.yml b/deploy/mongodb.yml
@@ -13,7 +13,7 @@ services:
       interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 1m        
+      start_period: 1m
     deploy:
       replicas: 1
       placement:

diff --git a/deploy/openaq.yml b/deploy/openaq.yml
@@ -5,6 +5,14 @@ services:
     image: ${OPENAQ_IMAGE}:${OPENAQ_TAG}
     environment:
       - DB_URL=${OPENAQ_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     deploy:
       replicas: 1
       placement:

diff --git a/deploy/teleray.yml b/deploy/teleray.yml
@@ -7,6 +7,14 @@ services:
       - S3_ACCESS_KEY=${S3_ACCESS_KEY}
       - S3_SECRET_ACCESS_KEY=${S3_SECRET_ACCESS_KEY}
       - S3_BUCKET=${TELERAY_S3_BUCKET}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     deploy:
       replicas: 1
       placement:

diff --git a/deploy/vigicrues-observations.yml b/deploy/vigicrues-observations.yml
@@ -5,6 +5,14 @@ services:
     image: ${VIGICRUES_OBSERVATIONS_IMAGE}:observations-${VIGICRUES_OBSERVATIONS_TAG}
     environment:
       - DB_URL=${VIGICRUES_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     deploy:
       replicas: 1
       placement:

diff --git a/deploy/vigicrues-sections.yml b/deploy/vigicrues-sections.yml
@@ -4,7 +4,15 @@ services:
   sections:
     image: ${VIGICRUES_SECTIONS_IMAGE}:sections-${VIGICRUES_SECTIONS_TAG}
     environment:
-      - DB_URL=${VIGICRUES_DB_URL}     
+      - DB_URL=${VIGICRUES_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     deploy:
       replicas: 1
       placement:

diff --git a/deploy/vigicrues-stations.yml b/deploy/vigicrues-stations.yml
@@ -4,7 +4,15 @@ services:
   stations:
     image: ${VIGICRUES_STATIONS_IMAGE}:stations-${VIGICRUES_STATIONS_TAG}
     environment:
-      - DB_URL=${VIGICRUES_DB_URL}     
+      - DB_URL=${VIGICRUES_DB_URL}
+      - SUBDOMAIN
+      - SLACK_WEBHOOK_URL
+    healthcheck:
+      test: ["CMD", "node", "./healthcheck.js", "--link-template", "\"portainer.<%= SUBDOMAIN %>\""]
+      interval: 1m
+      timeout: 10s
+      retries: 3
+      start_period: 1m
     deploy:
       replicas: 1
       placement: