Commit c751e00f authored by Ignacio's avatar Ignacio
Browse files

Merge branch 'dev' into 'master' 

Elimina config de prometheus

See merge request redmic-project/metric/swarmprom-ci!17
parents d647a553 61004465

.env

+1 −0
Original line number Diff line number Diff line 
PUBLIC_HOSTNAME=localhost

DOCKER_GWBRIDGE_IP=172.18.0.1

JOBS=pushgateway:9091

.gitlab-ci.yml

+50 −0
Original line number Diff line number Diff line 
stages:

  - check_rules

  - deploy



image: registry.gitlab.com/redmic-project/docker/docker-deploy
@@ -14,6 +15,17 @@ variables: 
  STACK: metric

  SERVICES_TO_CHECK: metric_dockerd-exporter metric_cadvisor metric_grafana metric_alertmanager metric_unsee metric_node-exporter metric_prometheus



check-rule-dev:

  stage: check_rules

  script:

    - >

      docker run -v $(pwd)/prometheus/rules:/tmp dnanexus/promtool:1.0

      check rules /tmp/swarm_task.rules.yml

  environment:

    name: dev

  only:

    - dev



deploy-dev:

  stage: deploy

  variables:
@@ -24,17 +36,35 @@ deploy-dev: 
    DOCKER_GWBRIDGE_IP: ${DEV_DOCKER_GWBRIDGE_IP}

    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 COMPOSE_FILE=${COMPOSE_FILE} ADMIN_USER=${ADMIN_USER} ADMIN_PASSWORD=${ADMIN_PASSWORD}

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

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

  environment:

    name: dev

  only:

    - dev

  when: manual



check-rule-supporting-branch:

  stage: check_rules

  script:

    - >

      docker run -v $(pwd)/prometheus/rules:/tmp dnanexus/promtool:1.0

      check rules /tmp/swarm_task.rules.yml

  environment:

    name: dev

  only:

    - branches

  except:

    - master

    - dev



deploy-supporting-branch:

  stage: deploy

  variables:
@@ -45,11 +75,15 @@ deploy-supporting-branch: 
    DOCKER_GWBRIDGE_IP: ${DEV_DOCKER_GWBRIDGE_IP}

    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 COMPOSE_FILE=${COMPOSE_FILE} ADMIN_USER=${ADMIN_USER} ADMIN_PASSWORD=${ADMIN_PASSWORD}

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

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

  environment:

    name: dev

  only:
@@ -59,6 +93,18 @@ deploy-supporting-branch: 
    - dev

  when: manual





check-rule-pro:

  stage: check_rules

  script:

    - >

      docker run -v $(pwd)/prometheus/rules:/tmp dnanexus/promtool:1.0

      check rules /tmp/swarm_task.rules.yml

  environment:

    name: pro

  only:

    - master



deploy-pro:

  stage: deploy

  variables:
@@ -69,11 +115,15 @@ deploy-pro: 
    DOCKER_GWBRIDGE_IP: ${PRO_DOCKER_GWBRIDGE_IP}

    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 COMPOSE_FILE=${COMPOSE_FILE} ADMIN_USER=${ADMIN_USER} ADMIN_PASSWORD=${ADMIN_PASSWORD}

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

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

  environment:

    name: pro

  only:

docker-compose.dev.yml

+2 −147
Original line number Diff line number Diff line 
version: '3.5'



services:

  dockerd-exporter:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  cadvisor:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  grafana:

    ports:

      - target: 3000

        published: 18004

        mode: host

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "3000"

        traefik.docker.network: traefik-net

        traefik.frontend.rule: Host:grafana.${PUBLIC_HOSTNAME}

        traefik.backend: grafana

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  alertmanager:

    ports:

      - target: 9093

        published: 9093

        mode: host

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9093"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:alertmanager.${PUBLIC_HOSTNAME}

        traefik.backend: alertmanager

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  unsee:

    ports:

      - target: 8080

        published: 9094

        mode: host

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "8080"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:unsee.${PUBLIC_HOSTNAME}

        traefik.backend: unsee

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  node-exporter:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  prometheus:

    ports:

      - target: 9090

        published: 9090

        mode: host

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9090"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:prometheus.${PUBLIC_HOSTNAME}

        traefik.backend: prometheus

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 512M

        reservations:

          memory: 128M



volumes:

  prometheus:

    name: prometheus-vol

  pushgateway:

    name: pushgateway-vol



  grafana:

    name: grafana-vol

docker-compose.prod.yml

+3 −123
Original line number Diff line number Diff line 
version: '3.5'



services:

  dockerd-exporter:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  cadvisor:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  grafana:

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "3000"

        traefik.docker.network: traefik-net

        traefik.frontend.rule: Host:grafana.${PUBLIC_HOSTNAME}

        traefik.backend: grafana

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  alertmanager:

  pushgateway:

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9093"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:alertmanager.${PUBLIC_HOSTNAME}

        traefik.backend: alertmanager

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  unsee:

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "8080"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:unsee.${PUBLIC_HOSTNAME}

        traefik.backend: unsee

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  node-exporter:

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  prometheus:

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9090"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:prometheus.${PUBLIC_HOSTNAME}

        traefik.backend: prometheus

      placement:

        constraints:

          - node.role == worker

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 512M

        reservations:

          memory: 128M



volumes:

  prometheus:

    name: prometheus-vol

    driver: "cloudstor:aws"

    driver_opts:

      backing: relocatable

      size: 5

      ebstype: gp2

  pushgateway:

    name: pushgateway-vol



  grafana:

    name: grafana-vol

docker-compose.tmpl.yml

+112 −20
Original line number Diff line number Diff line
@@ -12,6 +12,17 @@ services: 
    configs:

      - source: dockerd_config

        target: /etc/caddy/Caddyfile

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  cadvisor:

    image: google/cadvisor
@@ -24,6 +35,17 @@ services: 
      - /var/run:/var/run

      - /sys:/sys:ro

      - /var/lib/docker/:/var/lib/docker:ro

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  grafana:

    image: stefanprodan/swarmprom-grafana:5.0.1
@@ -31,6 +53,7 @@ services: 
      - GF_SECURITY_ADMIN_USER=${ADMIN_USER:-admin}

      - GF_SECURITY_ADMIN_PASSWORD=${ADMIN_PASSWORD:-admin}

      - GF_USERS_ALLOW_SIGN_UP=false

      - GF_INSTALL_PLUGINS=grafana-clock-panel,grafana-piechart-panel

      #- GF_SERVER_ROOT_URL=${GF_SERVER_ROOT_URL:-localhost}

      #- GF_SMTP_ENABLED=${GF_SMTP_ENABLED:-false}

      #- GF_SMTP_FROM_ADDRESS=${GF_SMTP_FROM_ADDRESS:-grafana@test.com}
@@ -45,6 +68,23 @@ services: 
      - grafana:/var/lib/grafana

      - grafana-etc:/etc/grafana

      - grafana-log:/var/log/grafana

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "3000"

        traefik.docker.network: traefik-net

        traefik.frontend.rule: Host:grafana.${PUBLIC_HOSTNAME}

        traefik.backend: grafana

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  alertmanager:

    image: stefanprodan/swarmprom-alertmanager:v0.14.0
@@ -60,6 +100,27 @@ services: 
      - traefik-net

    volumes:

      - alertmanager:/alertmanager

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9093"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:alertmanager.${PUBLIC_HOSTNAME}

        traefik.backend: alertmanager

      placement:

        constraints:

          - node.role == manager

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  unsee:

    image: cloudflare/unsee:v0.8.0
@@ -68,6 +129,24 @@ services: 
      - traefik-net

    environment:

      - "ALERTMANAGER_URIS=default:http://alertmanager:9093"

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "8080"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:unsee.${PUBLIC_HOSTNAME}

        traefik.backend: unsee

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  node-exporter:

    image: stefanprodan/swarmprom-node-exporter:v0.15.2
@@ -89,26 +168,45 @@ services: 
      - /sys:/host/sys:ro

      - /:/rootfs:ro

      - /etc/hostname:/etc/nodename

    deploy:

      mode: global

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



  prometheus:

    image: stefanprodan/swarmprom-prometheus:v2.2.0-rc.0

  pushgateway:

    image: prom/pushgateway

    command:

      - '--config.file=/etc/prometheus/prometheus.yml'

      - '--web.console.libraries=/etc/prometheus/console_libraries'

      - '--web.console.templates=/etc/prometheus/consoles'

      - '--storage.tsdb.path=/prometheus'

      - '--storage.tsdb.retention=30d'

    user: "0:0"

      - '--persistence.file=/db/metrics'

    networks:

      - metric-net

      - traefik-net

    volumes:

      - prometheus:/prometheus

    configs:

      - source: node_rules

        target: /etc/prometheus/swarm_node.rules.yml

      - source: task_rules

        target: /etc/prometheus/swarm_task.rules.yml

      - pushgateway:/db

    deploy:

      mode: replicated

      replicas: 1

      labels:

        traefik.port: "9091"

        traefik.docker.network: traefik-net

        traefik.frontend.auth.basic: ${UI_AUTH}

        traefik.frontend.rule: Host:pushgateway.${PUBLIC_HOSTNAME}

        traefik.backend: pushgateway

      restart_policy:

        delay: 1m

        window: 3m

      resources:

        limits:

          cpus: '0.5'

          memory: 128M

        reservations:

          memory: 64M



networks:

  metric-net:
@@ -120,9 +218,3 @@ networks: 
configs:

  dockerd_config:

    file: ./swarmprom/dockerd-exporter/Caddyfile
 


  node_rules:

    file: ./swarmprom/prometheus/rules/swarm_node.rules.yml



  task_rules:

    file: ./swarmprom/prometheus/rules/swarm_task.rules.yml