Add aggregate graph for API errors

[chef.git] / cookbooks / munin / templates / default / munin.conf.erb

diff --git a/cookbooks/munin/templates/default/munin.conf.erb b/cookbooks/munin/templates/default/munin.conf.erb
index 8f4af9b4a81a3c8a6338a2a9316a1e0877d9dc0c..06ee3c917bf2463e201ab38c3377b292b30d6a4e 100644 (file)
--- a/cookbooks/munin/templates/default/munin.conf.erb
+++ b/cookbooks/munin/templates/default/munin.conf.erb
@@ -16,7 +16,9 @@ rrdcached_socket /var/run/rrdcached.sock
 
 # Configure alert targets
 contact.admins.command mail -s "Munin Notification for ${var:host}" admins@openstreetmap.org
+contact.admins.always_send invalid
 contact.null.command cat > /dev/null
+contact.null.always_send invalid
 
 # Send alerts to the admins by default
 contacts admins
@@ -133,6 +135,17 @@ unknown_limit 144
     api_calls_www.map.label Map API calls
     api_calls_www.trkpts.sum <%= Chef::Munin.expand "%%.openstreetmap:api_calls_%%.trkpts", @frontends %>
     api_calls_www.trkpts.label GPX trackpoints calls
+    api_calls_error.graph_title HTTP errors
+    api_calls_error.graph_vlabel Number of errors per ${graph_period}
+    api_calls_error.graph_category api
+    api_calls_error.http500.sum <%= Chef::Munin.expand "%%.openstreetmap:api_calls_status.http500", @frontends %>
+    api_calls_error.http500.label 500 Internal Server Error
+    api_calls_error.http502.sum <%= Chef::Munin.expand "%%.openstreetmap:api_calls_status.http502", @frontends %>
+    api_calls_error.http502.label 502 Bad Gateway
+    api_calls_error.http503.sum <%= Chef::Munin.expand "%%.openstreetmap:api_calls_status.http503", @frontends %>
+    api_calls_error.http503.label 503 Service Unavailable
+    api_calls_error.http509.sum <%= Chef::Munin.expand "%%.openstreetmap:api_calls_status.http509", @frontends %>
+    api_calls_error.http509.label 509 Bandwidth Limit Exceeded 
     api_calls_num.graph_title Requests processed
     api_calls_num.graph_vlabel Number of requests per minute
     api_calls_num.graph_category api