Commit c5bf2064 authored by Ignacio's avatar Ignacio
Browse files

Merge branch 'dev' into 'master' 

Añade alertas para la renovación del certificado

See merge request redmic-project/metric/prometheus!1
parents d7f7a3e5 a2484dfc

.gitlab-ci.yml

+3 −15
Original line number Diff line number Diff line
@@ -73,15 +73,11 @@ deploy-dev: 
    COMPOSE_FILE: docker-compose.tmpl.yml:docker-compose.dev.yml

    UI_AUTH: ${DEV_UI_AUTH}

    PUBLIC_HOSTNAME: ${DEV_PUBLIC_HOSTNAME}

    SLACK_URL: ${SLACK_URL}

    SLACK_CHANNEL: ${DEV_SLACK_CHANNEL}

    SLACK_USER: ${SLACK_USER}

  script:

    - create-nets.sh metric-net

    - >

      deploy.sh IMAGE_NAME=${IMAGE_NAME} IMAGE_TAG=${IMAGE_TAG} COMPOSE_FILE=${COMPOSE_FILE}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME} SLACK_URL=${SLACK_URL}

      SLACK_CHANNEL=${SLACK_CHANNEL} SLACK_USER=${SLACK_USER}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME}

  environment:

    name: dev

    url: https://prometheus.${DEV_PUBLIC_HOSTNAME}
@@ -103,15 +99,11 @@ deploy-supporting-branch: 
    COMPOSE_FILE: docker-compose.tmpl.yml:docker-compose.dev.yml

    UI_AUTH: ${DEV_UI_AUTH}

    PUBLIC_HOSTNAME: ${DEV_PUBLIC_HOSTNAME}

    SLACK_URL: ${SLACK_URL}

    SLACK_CHANNEL: ${DEV_SLACK_CHANNEL}

    SLACK_USER: ${SLACK_USER}

  script:

    - create-nets.sh metric-net

    - >

      deploy.sh IMAGE_NAME=${IMAGE_NAME} IMAGE_TAG=${IMAGE_TAG} COMPOSE_FILE=${COMPOSE_FILE}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME} SLACK_URL=${SLACK_URL}

      SLACK_CHANNEL=${SLACK_CHANNEL} SLACK_USER=${SLACK_USER}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME}

  environment:

    name: dev

    url: https://prometheus.${DEV_PUBLIC_HOSTNAME}
@@ -136,15 +128,11 @@ deploy-pro: 
    COMPOSE_FILE: docker-compose.tmpl.yml:docker-compose.prod.yml

    UI_AUTH: ${PRO_UI_AUTH}

    PUBLIC_HOSTNAME: ${PRO_PUBLIC_HOSTNAME}

    SLACK_URL: ${SLACK_URL}

    SLACK_CHANNEL: ${PRO_SLACK_CHANNEL}

    SLACK_USER: ${SLACK_USER}

  script:

    - create-nets.sh metric-net

    - >

      deploy.sh IMAGE_NAME=${IMAGE_NAME} IMAGE_TAG=${IMAGE_TAG} COMPOSE_FILE=${COMPOSE_FILE}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME} SLACK_URL=${SLACK_URL}

      SLACK_CHANNEL=${SLACK_CHANNEL} SLACK_USER=${SLACK_USER}

      UI_AUTH=${UI_AUTH} PUBLIC_HOSTNAME=${PUBLIC_HOSTNAME}

  environment:

    name: pro

    url: https://prometheus.${PUBLIC_HOSTNAME}

deploy/.env

+1 −1
Original line number Diff line number Diff line 
JOBS="api:8080/api/actuator/prometheus oauth:8081/api/oauth/actuator/prometheus user:8082/api/user/actuator/prometheus vessels-commands:8091/api/vessels/commands/actuator/prometheus vessels-view:8092/api/vessels/view/actuator/prometheus web:3050/metrics"
 
JOBS=api:8080/api/actuator/prometheus oauth:8081/api/oauth/actuator/prometheus user:8082/api/user/actuator/prometheus vessels-commands:8091/api/vessels/commands/actuator/prometheus vessels-view:8092/api/vessels/view/actuator/prometheus web:3050/metrics socket:8089/api/socket/actuator/prometheus tasks:8088/api/tasks/actuator/prometheus ais:8090/api/ais/actuactor/prometheus

deploy/rules/swarm_job.rules.yml

+17 −1
Original line number Diff line number Diff line
@@ -8,3 +8,19 @@ groups: 
    annotations:

      description: Error creating backup for '{{ $labels.label }}' database.

      summary: Error creating backup for '{{ $labels.label }}' database.



  - alert: certificate_renew_not_attemped

    expr: time() - certificates_valid_date_seconds{} > 604800

    labels:

      severity: warning

    annotations:

      description: Error attemping to renew '{{ $labels.label }}' certificate.

      summary: Error attemping to renew '{{ $labels.label }}' certificate.



  - alert: certificate_not_renewed

    expr: time() - certificates_updated_date_seconds{} > 5788800

    labels:

      severity: warning

    annotations:

      description: Error, '{{ $labels.label }}' certificate near expiry.

      summary: Error, '{{ $labels.label }}' certificate near expiry.
 
 No newline at end of file

deploy/rules/swarm_service.rules.yml

+8 −0
Original line number Diff line number Diff line
@@ -10,3 +10,11 @@ groups: 
      description: The service '{{ $labels.label }}' has restarted more 3 times in last hour.

      summary: The service '{{ $labels.label }}' has restarted more 3 times in last hour.



  - alert: service_down

    expr: up == 0

    for: 5m

    labels:

      severity: severity

    annotations:

      description: Instance {{ $labels.instance }} down.

      summary: "{{ $labels.instance }} of job {{ $labels.job }} has been down for more than 5 minutes."
 
 No newline at end of file