Add InfluxDB Relay alerts
Change-Id: I9299d81a592fdfee33f3f4e1d5ca6300d18d94cc
diff --git a/influxdb/map.jinja b/influxdb/map.jinja
index 47a9c00..a2a181c 100644
--- a/influxdb/map.jinja
+++ b/influxdb/map.jinja
@@ -41,6 +41,8 @@
'http_errors_percentage': 5,
'failed_points_percentage': 5,
'dropped_points_percentage': 5,
+ 'max_relay_buffer_percentage': 70,
+ 'relay_failed_requests_percentage': 5,
},
}, grain='os_family', merge=salt['pillar.get']('influxdb:monitoring')) %}
diff --git a/influxdb/meta/prometheus.yml b/influxdb/meta/prometheus.yml
index ea66b51..c266dfc 100644
--- a/influxdb/meta/prometheus.yml
+++ b/influxdb/meta/prometheus.yml
@@ -70,6 +70,40 @@
annotations:
summary: 'Influxdb too many dropped writes'
description: '{{ printf `%.1f` $value }}% of written points have been dropped on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_points_written_dropped_threshold }}).'
+{%- if relay.get('enabled', False) and relay.telemetry is defined and relay.telemetry.get('enabled') %}
+ {%- set buffer_sizes = [] %}
+ {%- for name, listen in relay.listen.iteritems()|sort %}
+ {%- for backend_name, backend in listen.output.iteritems()|sort %}
+ {%- do buffer_sizes.append(backend.get('buffer_size_mb', 0)|float) %}
+ {%- endfor %}
+ {%- endfor %}
+ {%- set buffer_sizes = buffer_sizes|sort %}
+ {%- set buffer_size = buffer_sizes[-1] * 1024 * 1024 %}
+ {%- if buffer_size > 0 %}
+ InfluxdbRelayBufferNearFull:
+ {%- set influx_relay_buffer_size_threshold = monitoring.max_relay_buffer_percentage %}
+ if: >-
+ influxdb_relay_backend_buffer_bytes > {{ buffer_size }} * {{ influx_relay_buffer_size_threshold }} / 100
+ {% raw %}
+ labels:
+ severity: warning
+ service: influxdb-relay
+ annotations:
+ summary: 'InfluxDB Relay buffer almost full'
+ description: 'The buffer size for the {{ $labels.instance }}/{{ $labels.backend }} backend is getting full (current value={{ $value }} bytes, threshold={%- endraw %}{{ buffer_size * influx_relay_buffer_size_threshold / 100 }}).'
+ {%- endif %}
+ InfluxdbRelayFailedRequests:
+ {%- set influx_relay_failed_requests_threshold = monitoring.relay_failed_requests_percentage %}
+ if: >-
+ rate(influxdb_relay_failed_requests_total[5m]) / rate(influxdb_relay_requests_total[5m]) * 100 > {{ influx_relay_failed_requests_threshold }}
+ {% raw %}
+ labels:
+ severity: warning
+ service: influxdb-relay
+ annotations:
+ summary: 'InfluxDB Relay too many failed requests'
+ description: '{{ printf `%.1f` $value }}% of requests have been dropped on {{ $labels.instance }} (threshold={%- endraw %}{{ influx_relay_failed_requests_threshold }}).'
+
{%- endif %}
{%- if relay.get('enabled') and relay.telemetry.get('enabled') %}
@@ -96,3 +130,4 @@
{%- endif %}
{%- endif %}
+{%- endif %}
\ No newline at end of file