Samuel Berthe c37ef8f50c fix: review and fix 74 database & broker alert rules (#504) ... * fix: review and fix 74 database & broker alert rules Comprehensive review of all database and broker alerts covering 16 services. Typos & descriptions (8 fixes): - PGBouncer: "a a server" → "a server" - RabbitMQ: "instace" → "instance", "RabbmitMQ" → "RabbitMQ", "unactive" → "inactive" - Cassandra: write failure said "Read failures", "bad hacker" → "authentication failures" - Solr: replication errors said "failed updates" - Meilisearch: "index is empty" said "instance is down" Duplicates removed (5 fixes): - PostgreSQL: 2 rules using wrong exporter metric (postgresql_errors_total) - ClickHouse: "High Network Traffic" (thread counts) duplicated byte-rate rule - NATS: 2 rules with low thresholds duplicated better rules Broken queries (20 fixes): - Patroni: patroni_master → patroni_primary (renamed in v3) - MongoDB: rate() on gauge → direct ratio for connection queries - MongoDB: removed WiredTiger-incompatible virtual memory rule - Cassandra instaclustr: avg() on counter → rate()[5m] - Cassandra criteo: increase() on JMX rate metric → direct threshold - ClickHouse: increase() on gauge → direct threshold - NATS: rate() on gauge → direct comparison, removed 4 config-value rules - SQL Server: increase() on gauge → direct threshold - Pulsar: moved comparison outside sum() (4 rules) - Hadoop: inverted comparison < 0.2 → > 0.8, counters → increase()[1h] Severity adjustments (7 fixes): - Redis: backup threshold 24h → 48h, rejected connections → warning > 5 - RabbitMQ: no consumer for: 5m with comment - Elasticsearch: unassigned shards added for: 2m - CouchDB: process restarted critical → info - Kafka: consumer group lag → warning, threshold 10000, better description - Hadoop: HBase heap low critical → warning Missing for duration (18 fixes): - Added for: 1m to service-down alerts across MySQL, PostgreSQL, SQL Server, Patroni, Redis, MongoDB, RabbitMQ, Elasticsearch, Cassandra, Zookeeper with restart-tolerance comments Division by zero guards (9 fixes): - Added denominator > 0 guards to ratio queries in PostgreSQL, RabbitMQ, Elasticsearch, ClickHouse, CouchDB, NATS Query design improvements (5 fixes): - Cassandra: removed unnecessary sum() and redundant avg_over_time() - ClickHouse: ZooKeeper avg() → per-instance check - PostgreSQL: sum() → sum by (instance) for SSL and locks - PGBouncer: 30s range window → 2m Hardcoded labels (2 fixes): - ClickHouse: added comment about job="clickhouse" - Cassandra criteo: removed hardcoded service="cas" * fix: address PR review comments - Cassandra connection timeouts: wrap rate() in sum by() (rate() by() is invalid PromQL) - Elasticsearch query latency: add division-by-zero guard - Redis backup: "backuped" → "backed up"		2026-03-16 01:27:18 +01:00
..
rules.yml	fix: review and fix 74 database & broker alert rules (#504)	2026-03-16 01:27:18 +01:00