Merge "Fix alerts"
diff --git a/opencontrail/meta/prometheus.yml b/opencontrail/meta/prometheus.yml
index 6f86598..ed997a5 100644
--- a/opencontrail/meta/prometheus.yml
+++ b/opencontrail/meta/prometheus.yml
@@ -59,9 +59,9 @@
enabled: true
{%- if packages is defined %}
packages:
- {% for pkg in packages %}
+ {%- for pkg in packages %}
- {{ pkg }}
- {% endfor %}
+ {%- endfor %}
{%- endif %}
{%- if template is defined %}
template: {{ template }}
@@ -90,7 +90,7 @@
{%- if control_processes is defined %}
{%- for contrail_api in control_apis %}
{%- set words = contrail_api.split('.') %}
- {% for word in words %}{%- if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
+ {% for word in words %}{% if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
if: >-
http_response_status{service=~"{{ contrail_api }}"} == 0
{%- raw %}
@@ -113,13 +113,15 @@
severity: down
service: {{ contrail_process }}
annotations:
- summary: '{{ contrail_process }} service is down'
- description: '{{ contrail_process }} service is down on node {% raw %}{{ $labels.host }}{% endraw %}'
+{%- raw %}
+ summary: '{{ $labels.service }} service is down'
+ description: '{{ $labels.service }} service is down on node {{ $labels.host }}'
+{%- endraw %}
{%- endfor %}
{%- raw %}
ContrailBGPSessionsNoneUp:
if: >-
- max(contrail_bgp_session_up_count) == 0
+ max(contrail_bgp_session_up_count) by (host) == 0
for: 2m
labels:
severity: warning
@@ -129,7 +131,7 @@
description: 'There are no active BGP sessions on node {{ $labels.host }}'
ContrailBGPSessionsSomeDown:
if: >-
- min(contrail_bgp_session_down_count) > 0
+ min(contrail_bgp_session_down_count) by (host) > 0
for: 2m
labels:
severity: warning
@@ -139,7 +141,7 @@
description: 'There are inactive BGP sessions on node {{ $labels.host }}'
ContrailBGPSessionsNone:
if: >-
- max(contrail_bgp_session_count) == 0
+ max(contrail_bgp_session_count) by (host) == 0
for: 2m
labels:
severity: warning
@@ -149,7 +151,7 @@
description: 'There are no BGP sessions on node {{ $labels.host }}'
ContrailXMPPSessionsNoneUp:
if: >-
- max(contrail_xmpp_session_up_count) == 0
+ max(contrail_xmpp_session_up_count) by (host) == 0
for: 2m
labels:
severity: warning
@@ -159,7 +161,7 @@
description: 'There are no active XMPP sessions on node {{ $labels.host }}'
ContrailXMPPSessionsSomeDown:
if: >-
- min(contrail_xmpp_session_down_count) > 0
+ min(contrail_xmpp_session_down_count) by (host) > 0
for: 2m
labels:
severity: warning
@@ -169,7 +171,7 @@
description: 'There are inactive XMPP sessions on node {{ $labels.host }}'
ContrailXMPPSessionsNone:
if: >-
- max(contrail_xmpp_session_count) == 0
+ max(contrail_xmpp_session_count) by (host) == 0
for: 2m
labels:
severity: warning
@@ -181,7 +183,7 @@
if: >-
{%- endraw %}
{%- set xmpp_toohigh_threshold = monitoring.xmpp_sessions_too_high_threshold %}
- min(contrail_xmpp_session_count) >= {{ xmpp_toohigh_threshold }}
+ min(contrail_xmpp_session_count) by (host) >= {{ xmpp_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -189,7 +191,7 @@
service: contrail-control
annotations:
summary: 'Too many XMPP sessions'
- description: 'There are too many XMPP sessions on node {{ $labels.host }} (current value={{ $value }}, threshold={%- endraw %}{{ xmpp_toohigh_threshold }})'
+ description: 'There are too many XMPP sessions on node {{ $labels.host }} (current value={{ $value }}, threshold={% endraw %}{{ xmpp_toohigh_threshold }})'
ContrailXMPPSessionsTooManyVariations:
if: >-
{%- set xmpp_variation_threshold = monitoring.xmpp_sessions_variation_threshold %}
@@ -200,13 +202,13 @@
service: contrail-control
annotations:
summary: 'Number of XMPP sessions changed between checks is too high'
- description: 'There are too many XMPP sessions changes on node {{ $labels.host }} (current value={{ $value }}, threshold={%- endraw %}{{ xmpp_variation_threshold }})'
+ description: 'There are too many XMPP sessions changes on node {{ $labels.host }} (current value={{ $value }}, threshold={% endraw %}{{ xmpp_variation_threshold }})'
{%- endif %}
{%- if collector_processes is defined %}
{%- for contrail_api in collector_apis %}
{%- set words = contrail_api.split('.') %}
- {% for word in words %}{%- if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
+ {% for word in words %}{% if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
if: >-
http_response_status{service=~"{{ contrail_api }}"} == 0
{%- raw %}
@@ -229,8 +231,10 @@
severity: down
service: {{ contrail_process }}
annotations:
- summary: '{{ contrail_process }} service is down'
- description: '{{ contrail_process }} service is down on node {% raw %}{{ $labels.host }}{% endraw %}'
+{%- raw %}
+ summary: '{{ $labels.service }} service is down'
+ description: '{{ $labels.service }} service is down on node {{ $labels.host }}'
+{%- endraw %}
{%- endfor %}
{%- endif %}
@@ -238,7 +242,7 @@
{%- raw %}
ContrailVrouterXMPPSessionsNone:
if: >-
- max(contrail_vrouter_xmpp) == 0
+ max(contrail_vrouter_xmpp) by (host) == 0
for: 2m
labels:
severity: warning
@@ -250,7 +254,7 @@
if: >-
{%- endraw %}
{%- set vrouter_xmpp_toohigh_threshold = monitoring.vrouter_xmpp_sessions_too_high_threshold %}
- min(contrail_vrouter_xmpp) >= {{ vrouter_xmpp_toohigh_threshold }}
+ min(contrail_vrouter_xmpp) by (host) >= {{ vrouter_xmpp_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -273,7 +277,7 @@
{%- raw %}
ContrailVrouterDNSXMPPSessionsNone:
if: >-
- max(contrail_vrouter_dns_xmpp) == 0
+ max(contrail_vrouter_dns_xmpp) by (host) == 0
for: 2m
labels:
severity: warning
@@ -285,7 +289,7 @@
if: >-
{%- endraw %}
{%- set vrouter_dns_xmpp_toohigh_threshold = monitoring.vrouter_dns_xmpp_sessions_too_high_threshold %}
- min(contrail_vrouter_dns_xmpp) >= {{ vrouter_dns_xmpp_toohigh_threshold }}
+ min(contrail_vrouter_dns_xmpp) by (host) >= {{ vrouter_dns_xmpp_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -308,7 +312,7 @@
ContrailVrouterLLSSessionsTooMany:
if: >-
{%- set vrouter_lls_toohigh_threshold = monitoring.vrouter_lls_too_high_threshold %}
- min(contrail_vrouter_lls) >= {{ vrouter_lls_toohigh_threshold }}
+ min(contrail_vrouter_lls) by (host) >= {{ vrouter_lls_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -331,7 +335,7 @@
ContrailFlowsActiveTooMany:
if: >-
{%- set vrouter_flows_active_toohigh_threshold = monitoring.vrouter_flows_active_too_high_threshold %}
- min(contrail_vrouter_flows_active) >= {{ vrouter_flows_active_toohigh_threshold }}
+ min(contrail_vrouter_flows_active) by (host) >= {{ vrouter_flows_active_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -343,7 +347,7 @@
ContrailFlowsDiscardTooMany:
if: >-
{%- set vrouter_flows_discard_toohigh_threshold = monitoring.vrouter_flows_discard_too_high_threshold %}
- min(contrail_vrouter_flows_discard) >= {{ vrouter_flows_discard_toohigh_threshold }}
+ min(contrail_vrouter_flows_discard) by (host) >= {{ vrouter_flows_discard_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -355,7 +359,7 @@
ContrailFlowsDropTooMany:
if: >-
{%- set vrouter_flows_flow_action_drop_toohigh_threshold = monitoring.vrouter_flows_flow_action_drop_too_high_threshold %}
- min(contrail_vrouter_flows_flow_action_drop) >= {{ vrouter_flows_flow_action_drop_toohigh_threshold }}
+ min(contrail_vrouter_flows_flow_action_drop) by (host) >= {{ vrouter_flows_flow_action_drop_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -367,7 +371,7 @@
ContrailFlowsFragErrTooMany:
if: >-
{%- set vrouter_flows_frag_err_toohigh_threshold = monitoring.vrouter_flows_frag_err_too_high_threshold %}
- min(contrail_vrouter_flows_frag_err) >= {{ vrouter_flows_frag_err_toohigh_threshold }}
+ min(contrail_vrouter_flows_frag_err) by (host) >= {{ vrouter_flows_frag_err_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -379,7 +383,7 @@
ContrailFlowsInvalidNHTooMany:
if: >-
{%- set vrouter_flows_invalid_nh_toohigh_threshold = monitoring.vrouter_flows_invalid_nh_too_high_threshold %}
- min(contrail_vrouter_flows_invalid_nh) >= {{ vrouter_flows_invalid_nh_toohigh_threshold }}
+ min(contrail_vrouter_flows_invalid_nh) by (host) >= {{ vrouter_flows_invalid_nh_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -391,7 +395,7 @@
ContrailFlowsInvalidITFTooMany:
if: >-
{%- set vrouter_flows_composite_invalid_interface_toohigh_threshold = monitoring.vrouter_flows_composite_invalid_interface_too_high_threshold %}
- min(contrail_vrouter_flows_composite_invalid_interface) >= {{ vrouter_flows_composite_invalid_interface_toohigh_threshold }}
+ min(contrail_vrouter_flows_composite_invalid_interface) by (host) >= {{ vrouter_flows_composite_invalid_interface_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -403,7 +407,7 @@
ContrailFlowsInvalidLabelTooMany:
if: >-
{%- set vrouter_flows_invalid_label_toohigh_threshold = monitoring.vrouter_flows_invalid_label_too_high_threshold %}
- min(contrail_vrouter_flows_invalid_label) >= {{ vrouter_flows_invalid_label_toohigh_threshold }}
+ min(contrail_vrouter_flows_invalid_label) by (host) >= {{ vrouter_flows_invalid_label_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -415,7 +419,7 @@
ContrailFlowsQueueLimitExceededTooMany:
if: >-
{%- set vrouter_flows_flow_queue_limit_exceeded_toohigh_threshold = monitoring.vrouter_flows_flow_queue_limit_exceeded_too_high_threshold %}
- min(contrail_vrouter_flows_flow_queue_limit_exceeded) >= {{ vrouter_flows_flow_queue_limit_exceeded_toohigh_threshold }}
+ min(contrail_vrouter_flows_flow_queue_limit_exceeded) by (host) >= {{ vrouter_flows_flow_queue_limit_exceeded_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -427,7 +431,7 @@
ContrailFlowsTableFullTooMany:
if: >-
{%- set vrouter_flows_flow_table_full_toohigh_threshold = monitoring.vrouter_flows_flow_table_full_too_high_threshold %}
- min(contrail_vrouter_flows_flow_table_full) >= {{ vrouter_flows_flow_table_full_toohigh_threshold }}
+ min(contrail_vrouter_flows_flow_table_full) by (host) >= {{ vrouter_flows_flow_table_full_toohigh_threshold }}
{%- raw %}
for: 2m
labels:
@@ -438,7 +442,7 @@
description: 'There are too many vRouter flows with table full on node {{ $labels.host }} (current value={{ $value }}, threshold={%- endraw %}{{ vrouter_flows_flow_table_full_toohigh_threshold }})'
{%- for contrail_api in compute_apis %}
{%- set words = contrail_api.split('.') %}
- {% for word in words %}{%- if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
+ {% for word in words %}{% if word != 'api' %}{{ word | capitalize }}{% endif %}{% endfor %}APIDown:
if: >-
http_response_status{service=~"{{ contrail_api }}"} == 0
{%- raw %}
@@ -461,8 +465,10 @@
severity: down
service: {{ contrail_process }}
annotations:
- summary: '{{ contrail_process }} service is down'
- description: '{{ contrail_process }} service is down on node {% raw %}{{ $labels.host }}{% endraw %}'
+{%- raw %}
+ summary: '{{ $labels.service }} service is down'
+ description: '{{ $labels.service }} service is down on node {{ $labels.host }}'
+{%- endraw %}
{%- endfor %}
{%- endif %}
@@ -476,8 +482,10 @@
severity: down
service: {{ contrail_process }}
annotations:
- summary: '{{ contrail_process }} service is down'
- description: '{{ contrail_process }} service is down on node {% raw %}{{ $labels.host }}{% endraw %}'
+{%- raw %}
+ summary: '{{ $labels.service }} service is down'
+ description: '{{ $labels.service }} service is down on node {{ $labels.host }}'
+{%- endraw %}
{%- endfor %}
{%- endif %}
@@ -491,8 +499,10 @@
severity: down
service: {{ contrail_process }}
annotations:
- summary: '{{ contrail_process }} service is down'
- description: '{{ contrail_process }} service is down on node {% raw %}{{ $labels.host }}{% endraw %}'
+{%- raw %}
+ summary: '{{ $labels.service }} service is down'
+ description: '{{ $labels.service }} service is down on node { $labels.host }}'
+{%- endraw %}
{%- endfor %}
{%- endif %}