Merge branch 'fix-prometheus'

BBVA · Mar 4, 2019 · 2faad60 · 2faad60
2 parents 839925e + eeee46d
commit 2faad60
Show file tree

Hide file tree

Showing 3 changed files with 40 additions and 15 deletions.
diff --git a/deploy/aws/provision/files/grafana/dashboards/Host.json b/deploy/aws/provision/files/grafana/dashboards/Host.json
@@ -17,7 +17,7 @@
   "gnetId": 179,
   "graphTooltip": 1,
   "id": 2,
-  "iteration": 1550157092295,
+  "iteration": 1551713243950,
   "links": [],
   "panels": [
     {
@@ -437,7 +437,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "sum(node_filesystem_size_bytes{job=~\"Qed0-HostMetrics\"})",
+          "expr": "sum(node_filesystem_size_bytes{job=~\"Qed.*\"})/sum(qed_instances_count)",
           "format": "time_series",
           "hide": false,
           "interval": "2m",
@@ -542,14 +542,21 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "sum(node_memory_MemTotal_bytes{job=~\"Qed0-HostMetrics\"})",
+          "expr": "sum(node_memory_MemTotal_bytes{job=~\"Qed.*\"})/sum(qed_instances_count)",
           "format": "time_series",
           "hide": false,
           "interval": "2m",
           "intervalFactor": 2,
-          "legendFormat": "Total",
+          "legendFormat": "Max-QED",
           "refId": "A"
         },
+        {
+          "expr": "sum(node_memory_MemTotal_bytes{job=~\"Agent.*\"})/(sum(qed_auditor_instances_count)+sum(qed_monitor_instances_count)+sum(qed_publisher_instances_count))",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Max-Agent",
+          "refId": "C"
+        },
         {
           "expr": "sum(node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) by (job)",
           "format": "time_series",
@@ -564,7 +571,7 @@
       "timeFrom": null,
       "timeRegions": [],
       "timeShift": null,
-      "title": "Node Mermory",
+      "title": "Node Memory",
       "tooltip": {
         "shared": true,
         "sort": 0,
@@ -813,5 +820,5 @@
   "timezone": "browser",
   "title": "Hosts Monitoring",
   "uid": "64nrElFmk",
-  "version": 250
-}
+  "version": 1
+}
diff --git a/deploy/aws/provision/main.yml b/deploy/aws/provision/main.yml
@@ -47,6 +47,8 @@
   tasks:
     - name: Gather facts
       setup:
+  tags:
+    - prometheus
 
 - hosts: type_aws_instance
   remote_user: ec2-user
@@ -88,6 +90,8 @@
     - vars.yml
   tasks:
     - include: tasks/prometheus/main.yml
+  tags:
+    - prometheus
 
 - hosts: role_qed
   remote_user: ec2-user
@@ -161,4 +165,4 @@
         - /var/qed/exporter-start.sh
   tags:
     - start-prometheus
-
+    - prometheus
diff --git a/deploy/aws/provision/templates/prometheus-config.yml.j2 b/deploy/aws/provision/templates/prometheus-config.yml.j2
@@ -23,37 +23,47 @@ scrape_configs:
     static_configs:
       - targets: ['localhost:9090']
 {% for host in groups['role_qed'] %}
-  - job_name: 'Qed{{loop.index}}-HostMetrics'
+  - job_name: 'Qed{{loop.index}}-Metrics'
     scrape_interval: 10s
     static_configs:
-      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
-{% endfor %}
-{% for host in groups['role_qed'] %}
-  - job_name: 'Qed{{loop.index}}-QedMetrics'
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:8600']
+  - job_name: 'Qed{{loop.index}}-HostMetrics'
     scrape_interval: 10s
     static_configs:
-      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:8600']
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
 {% endfor %}
 {% for host in groups['role_publisher'] %}
   - job_name: 'Agent-Publisher{{loop.index}}-Metrics'
     scrape_interval: 10s
     static_configs:
       - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:18300']
+  - job_name: 'Agent-Publisher{{loop.index}}-HostMetrics'
+    scrape_interval: 10s
+    static_configs:
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
 {% endfor %}
 {% for host in groups['role_monitor'] %}
   - job_name: 'Agent-Monitor{{loop.index}}-Metrics'
     scrape_interval: 10s
     static_configs:
       - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:18200']
+  - job_name: 'Agent-Monitor{{loop.index}}-HostMetrics'
+    scrape_interval: 10s
+    static_configs:
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
 {% endfor %}
 {% for host in groups['role_auditor'] %}
   - job_name: 'Agent-Auditor{{loop.index}}-Metrics'
     scrape_interval: 10s
     static_configs:
       - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:18100']
+  - job_name: 'Agent-Auditor{{loop.index}}-HostMetrics'
+    scrape_interval: 10s
+    static_configs:
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
 {% endfor %}
 {% for host in groups['role_riot'] %}
-  - job_name: 'Riot{{loop.index}}-Metrics'
+  - job_name: 'Riot{{loop.index}}-HostMetrics'
     scrape_interval: 10s
     static_configs:
       - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
@@ -63,4 +73,8 @@ scrape_configs:
     scrape_interval: 10s
     static_configs:
       - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:18888']
+  - job_name: 'Storage{{loop.index}}-HostMetrics'
+    scrape_interval: 10s
+    static_configs:
+      - targets: ['{{ hostvars[host]['ansible_eth0']['ipv4']['address'] }}:9100']
 {% endfor %}