X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/be9687cc230d12b9ddc9246840861aaeae8749ea..65393f3e8c6636734c32efc34cc33a7fbcac529e:/tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls

diff --git a/tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls b/tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls
index dd5594dd88..6dc90c840b 100644
--- a/tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls
+++ b/tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls
@@ -3,17 +3,18 @@
 #
 # SPDX-License-Identifier: AGPL-3.0
 
+{%- set controller_nodes = "__CONTROLLER_NODES__".split(',') %}
+{%- set enable_balancer = ("__ENABLE_BALANCER__"|to_bool) %}
+
 ### PROMETHEUS
 prometheus:
   wanted:
     component:
       - prometheus
       - alertmanager
-      - blackbox_exporter
+      - node_exporter
   pkg:
     use_upstream_repo: true
-    use_upstream_archive: true
-
     component:
       prometheus:
         config:
@@ -34,18 +35,47 @@ prometheus:
                   cluster: __CLUSTER__
 
             ## Arvados unique jobs
+            - job_name: arvados_ws
+              bearer_token: __MANAGEMENT_TOKEN__
+              scheme: https
+              static_configs:
+                - targets: ['ws.__DOMAIN__:443']
+                  labels:
+                    instance: ws.__CLUSTER__
+                    cluster: __CLUSTER__
+            - job_name: arvados_controller
+              bearer_token: __MANAGEMENT_TOKEN__
+              {%- if enable_balancer %}
+              scheme: http
+              {%- else %}
+              scheme: https
+              {%- endif %}
+              static_configs:
+                {%- if enable_balancer %}
+                  {%- for controller in controller_nodes %}
+                - targets: ['{{ controller }}']
+                  labels:
+                    instance: {{ controller.split('.')[0] }}.__CLUSTER__
+                    cluster: __CLUSTER__
+                  {%- endfor %}
+                {%- else %}
+                - targets: ['__DOMAIN__:443']
+                  labels:
+                    instance: controller.__CLUSTER__
+                    cluster: __CLUSTER__
+                {%- endif %}
             - job_name: keep_web
               bearer_token: __MANAGEMENT_TOKEN__
               scheme: https
               static_configs:
-                - targets: ['keep.__CLUSTER__.__DOMAIN__:443']
+                - targets: ['keep.__DOMAIN__:443']
                   labels:
                     instance: keep-web.__CLUSTER__
                     cluster: __CLUSTER__
             - job_name: keep_balance
               bearer_token: __MANAGEMENT_TOKEN__
               static_configs:
-                - targets: ['__CONTROLLER_INT_IP__:9005']
+                - targets: ['__KEEPBALANCE_INT_IP__:9005']
                   labels:
                     instance: keep-balance.__CLUSTER__
                     cluster: __CLUSTER__
@@ -56,14 +86,10 @@ prometheus:
                   labels:
                     instance: keep0.__CLUSTER__
                     cluster: __CLUSTER__
-                - targets: ['__KEEPSTORE1_INT_IP__:25107']
-                  labels:
-                    instance: keep1.__CLUSTER__
-                    cluster: __CLUSTER__
             - job_name: arvados_dispatch_cloud
               bearer_token: __MANAGEMENT_TOKEN__
               static_configs:
-                - targets: ['__CONTROLLER_INT_IP__:9006']
+                - targets: ['__DISPATCHER_INT_IP__:9006']
                   labels:
                     instance: arvados-dispatch-cloud.__CLUSTER__
                     cluster: __CLUSTER__
@@ -78,3 +104,18 @@ prometheus:
                   labels:
                     instance: database.__CLUSTER__
                     cluster: __CLUSTER__
+
+            # Nodes
+            {%- set node_list = "__NODELIST__".split(',') %}
+            {%- set nodes = [] %}
+            {%- for node in node_list %}
+              {%- set _ = nodes.append(node.split('.')[0]) %}
+            {%- endfor %}
+            - job_name: node
+              static_configs:
+                {% for node in nodes %}
+                - targets: [ "{{ node }}.__DOMAIN__:9100" ]
+                  labels:
+                    instance: "{{ node }}.__CLUSTER__"
+                    cluster: __CLUSTER__
+                {% endfor %}