1
0
Fork 0
mirror of https://github.com/monitoring-mixins/website.git synced 2024-12-14 11:37:31 +00:00
monitoring-mixins-website/assets/varnish/alerts.yaml

68 lines
2.8 KiB
YAML
Raw Normal View History

groups:
- name: varnish-cache
rules:
- alert: VarnishCacheLowCacheHitRate
annotations:
description: The Cache hit rate is {{ printf "%.0f" $value }} percent over the
last 5 minutes on {{$labels.instance}}, which is below the threshold of 80
percent.
summary: Cache is not answering a sufficient percentage of read requests.
expr: |
increase(varnish_main_cache_hit[10m]) / (clamp_min((increase(varnish_main_cache_hit[10m]) + increase(varnish_main_cache_miss[10m])), 1)) * 100 < 80 and (increase(varnish_main_cache_hit[10m]) + increase(varnish_main_cache_miss[10m]) > 0)
for: 10m
labels:
severity: warning
- alert: VarnishCacheHighMemoryUsage
annotations:
description: Current Memory Usage is {{ printf "%.0f" $value }} percent on {{$labels.instance}},
which is above the threshold of 90 percent.
summary: Varnish Cache is running low on available memory.
expr: |
(varnish_sma_g_bytes{type="s0"} / (varnish_sma_g_bytes{type="s0"} + varnish_sma_g_space{type="s0"})) * 100 > 90
for: 5m
labels:
severity: warning
- alert: VarnishCacheHighCacheEvictionRate
annotations:
description: The Cache has evicted {{ printf "%.0f" $value }} objects over the
last 5 minutes on {{$labels.instance}}, which is above the threshold of 0.
summary: The cache is evicting too many objects.
expr: |
increase(varnish_main_n_lru_nuked[5m]) > 0
for: 5m
labels:
severity: critical
- alert: VarnishCacheHighSaturation
annotations:
description: The thread queue length is {{ printf "%.0f" $value }} over the
last 5 minutes on {{$labels.instance}}, which is above the threshold of 0.
summary: There are too many threads in queue, Varnish is saturated and responses
are slowed.
expr: |
varnish_main_thread_queue_len > 0
for: 5m
labels:
severity: warning
- alert: VarnishCacheSessionsDropping
annotations:
description: The amount of sessions dropped is {{ printf "%.0f" $value }} over
the last 5 minutes on {{$labels.instance}}, which is above the threshold of
0.
summary: Incoming requests are being dropped due to a lack of free worker threads.
expr: |
increase(varnish_main_sessions{type="dropped"}[5m]) > 0
for: 5m
labels:
severity: critical
- alert: VarnishCacheBackendUnhealthy
annotations:
description: The amount of unhealthy backend statuses detected is {{ printf
"%.0f" $value }} over the last 5 minutes on {{$labels.instance}}, which is
above the threshold of 0.
summary: Backend has been marked as unhealthy due to slow 200 responses.
expr: |
increase(varnish_main_backend_unhealthy[5m]) > 0
for: 5m
labels:
severity: critical