mirror of
https://github.com/samber/awesome-prometheus-alerts.git
synced 2026-06-21 00:47:18 +08:00
2258835c30
* fix(data): comprehensive PromQL review across all ~937 rules
Query fixes:
- Replace rate()/increase() with deriv()/delta() on gauge metrics exposed
as untyped by exporters (node_vmstat_oom_kill, mysql_global_status_*,
systemd_socket_refused_connections_total)
- Fix Ceph OSD latency metric name: ceph_osd_perf_apply_latency_seconds
→ ceph_osd_apply_latency_ms (Ceph MGR Prometheus module)
- Fix NATS subscriptions metric: gnatsd_connz_subscriptions (per-conn)
→ gnatsd_varz_subscriptions (server total)
- Fix Caddy reverse proxy down query: count()==0 → direct gauge == 0
- Fix RabbitMQ total connections metric: connectionsTotal → connections
- Fix Cilium ClusterMesh/KVStoreMesh: deriv() on failure gauge → direct
gauge comparison (deriv > 0 misses stable non-zero failure states)
- Fix cert-manager ACME metric name: certmanager_http_acme_client_request_count
→ certmanager_acme_client_request_count (renamed in v1.19+)
- Fix Thanos Query gRPC filter: grpc_code!="OK" → explicit error codes
- Fix Flink duplicate comments: field (YAML last-write-wins bug)
- Add datid!="0" filter to PostgreSQL dead locks query
- Fix PostgreSQL high rollback rate: restructure division-by-zero guard
and move ratio calculation outside sum()
- Add division-by-zero guards: Container Low CPU, Hadoop ResourceManager
memory, Hadoop HBase heap, Vault cluster health
- Add for: 1m to Blackbox probe failed/HTTP failure and Ceph State/
OSD Down/PG unavailable
Threshold fixes:
- Replace > 0 with meaningful thresholds on rate()/increase() queries
across: Alertmanager, eBPF decoder errors, systemd refused connections,
Memcached, Cassandra (Instaclustr + Criteo), ClickHouse distributed
inserts, CouchDB log entries, HAProxy healthcheck failures, RabbitMQ
unroutable messages, Spinnaker, Cilium, Mimir TSDB/alertmanager,
OTel Collector receiver refused metrics
- Fix Elasticsearch High Indexing Latency threshold: 0.0005s → 0.01s
(0.5ms was below normal operating range; 10ms is more realistic)
Description fixes:
- Fix MySQL slow queries: remove duplicate "mysql" word
- Fix SMART device description: remove trailing stray ")" (6 rules)
- Fix host disk IO description: remove duplicate "Check storage for issues."
- Fix EDAC correctable errors: "last 5 minutes" → "last 1 minute"
- Fix EDAC uncorrectable errors: remove time-window claim (raw counter)
- Fix Mimir store-gateway sync description: said "10 minutes" but
threshold is 1800s (30 minutes)
- Fix Vault description false "%" suffix on count values
- Improve descriptions across RabbitMQ, Zookeeper, Kafka, Pulsar, Envoy,
Istio rules to include {{ $labels }} and {{ $value }} template vars
- Downgrade Cassandra key cache hit rate: critical → warning
Comments:
- Add note on node_vmstat_oom_kill gauge type (delta vs increase)
- Add note on systemd_socket_refused_connections_total gauge type
- Add note on mysql_global_status_* gauge type (delta/deriv vs rate)
- Add note on pg_txid_current requiring a custom postgres_exporter query
- Add note on pg_stat_ssl_compression availability (PG 9.5-13 only)
- Add note on cert-manager legacy metric name for users on v1.18 and older
- Add threshold rationale for Elasticsearch, Cassandra, CouchDB rules
- Add note on NATS leaf node spurious fires when leaf nodes not configured
* fix(data): PromQL type fixes, job filter cleanup, query correctness review
- Replace rate()/increase() with deriv()/delta() on gauge metrics:
node_vmstat_pgmajfault, cassandra_stats (criteo exporter),
gitlab_ci_pipeline_failure_reasons, flink_taskmanager_job_task_numRecordsIn
- Fix histogram_quantile on non-_bucket metric: cilium_policy_implementation_delay
- Fix Thanos bucket replicate latency: use _count instead of _bucket for guard clause
- Fix Thanos query latency: use _count instead of _bucket for guard clause
- Restore job filter in Thanos objstore guard clauses (compact + store)
- Remove redundant job= filters from unique metrics: ~30 Thanos rules,
kube_persistentvolume_status_phase, otelcol_process_runtime_*
- Fix high-cardinality Istio latency grouping (drop source labels from by())
- Add division-by-zero guard to host context switch ratio
- Raise noisy ClickHouse thresholds: RejectedInserts > 2, DelayedInserts > 10
- Remove redundant for: 1m from HAProxy check failure rules
- Add job rename comments to up{job=...} rules (Hadoop, OpenStack, SNMP, OTel)
- Remove external mixin references from comments
- Fix Tempo dropped spans metric name: add missing _total suffix
- Fix Thanos bucket replicate run latency: add missing le label in by()
|
||
|---|---|---|
| .. | ||
| rules.yml | ||