Add alert for node exporter text file scrape errors

[chef.git] / cookbooks / prometheus / templates / default / prometheus.yml.erb

diff --git a/cookbooks/prometheus/templates/default/prometheus.yml.erb b/cookbooks/prometheus/templates/default/prometheus.yml.erb
index 6e28106cb8bd35966e7ef181545a9dd1c780e270..02ade7d8e225d9e94dd7b2df6bbd6bbc3f9929d8 100644 (file)
--- a/cookbooks/prometheus/templates/default/prometheus.yml.erb
+++ b/cookbooks/prometheus/templates/default/prometheus.yml.erb
@@ -31,11 +31,13 @@ scrape_configs:
     scrape_interval: 15m
     metrics_path: /probe
     static_configs:
-      - targets:
 <% @certificates.values.sort_by { |c| c[:domains].first }.each do |certificate| -%>
+      - targets:
 <% certificate[:nodes].sort_by { |h| h[:name] }.each do |host| -%>
           - <%= certificate[:domains].first %>/<%= host[:name] %>:443
 <% end -%>
+        labels:
+          domain: <%= certificate[:domains].first %>
 <% end -%>
     relabel_configs:
       - source_labels: [__address__]
@@ -51,12 +53,28 @@ scrape_configs:
         replacement: 127.0.0.1:9219
 <% @jobs.sort.each do |name, targets| -%>
   - job_name: <%= name %>
+<% if targets.first[:scrape_interval] -%>
+    scrape_interval: <%= targets.first[:scrape_interval] %>
+<% end -%>
+<% if targets.first[:scrape_timeout] -%>
+    scrape_timeout: <%= targets.first[:scrape_timeout] %>
+<% end -%>
+<% if targets.first[:sni] -%>
+    tls_config:
+      server_name: <%= targets.first[:sni] %>
+    relabel_configs:
+      - target_label: __scheme__
+        replacement: https
+<% end -%>
     static_configs:
 <% targets.each do |target| -%>
       - targets:
           - "<%= target[:address] %>"
         labels:
           instance: <%= target[:instance] %>
+<% target[:labels].sort.each do |name, value| -%>
+          <%= name %>: <%= value %>
+<% end -%>
 <% end -%>
     metric_relabel_configs:
 <% targets.each do |target| -%>
@@ -67,18 +85,52 @@ scrape_configs:
 <% end -%>
 <% end -%>
 <% end -%>
-
-remote_write:
-  - url: "http://localhost:9201/write"
-    write_relabel_configs:
-      - source_labels: [__name__]
-        regex: "go_.*"
-        action: drop
-      - source_labels: [__name__]
-        regex: "promhttp_.*"
-        action: drop
-    queue_config:
-      capacity: 10000
-      min_shards: 4
-      batch_send_deadline: 30s
-      max_backoff: 1s
+  - job_name: junos
+    scrape_interval: 5m
+    scrape_timeout: 4m
+    static_configs:
+<% @junos_targets.sort_by { |t| t[:instance] }.each do |target| -%>
+      - targets:
+          - "<%= target[:target] %>/<%= target[:address] %>"
+        labels:
+          instance: <%= target[:instance] %>
+<% target[:labels].sort.each do |name, value| -%>
+          <%= name %>: <%= value %>
+<% end -%>
+<% end -%>
+    relabel_configs:
+      - source_labels: [__address__]
+        regex: "([^/]+)/.*"
+        target_label: __param_target
+      - source_labels: [__address__]
+        regex: "[^/]+/(.*)"
+        target_label: __address__
+  - job_name: snmp
+    scrape_interval: 5m
+    scrape_timeout: 2m
+    metrics_path: /snmp
+    static_configs:
+<% @snmp_targets.sort_by { |t| t[:instance] }.each do |target| -%>
+      - targets:
+<% target[:modules].each do |module_name| -%>
+          - "<%= target[:target] %>/<%= module_name %>/<%= target[:address] %>"
+<% end -%>
+        labels:
+          instance: <%= target[:instance] %>
+<% target[:labels].sort.each do |name, value| -%>
+          <%= name %>: <%= value %>
+<% end -%>
+<% end -%>
+    relabel_configs:
+      - source_labels: [__address__]
+        regex: "([^/]+)/[^/]+/.*"
+        target_label: __param_target
+      - source_labels: [__address__]
+        regex: "[^/]+/([^/]+)/.*"
+        target_label: __param_module
+      - source_labels: [__address__]
+        regex: "[^/]+/([^/]+)/.*"
+        target_label: module
+      - source_labels: [__address__]
+        regex: "[^/]+/[^/]+/(.*)"
+        target_label: __address__