Alert definitions

Plain-English reference for every monitoring warning that is live in production (Prometheus tier0.yml). Each row links to its fix.

(the failure model) and the report at monitoring/threshold_tuning/threshold_report.md.

AlertSevWatchesFires whenWhat it meansFix
InstanceDowncriticala scrape target (prometheus, node_exporter, postgres_exporter, sui_node)unreachable for 2 mina host or exporter is down → we're flying blind on it↗
PostgresDowncriticalPostgreSQL (pg_up)down for 2 minthe database is unreachable → the API can't serve↗
SuiIndexerDowncriticalthe sui-indexer servicestopped >2 min (host still up)chain ingestion has halted → MEV data goes stale↗
SuiNodeFallingBehindcriticalSui node checkpoint lag>30 min behind (tuned from 10 min)the node isn't executing checkpoints → indexer gets stale data. Usual cause: WireGuard tunnel down↗
HostMemoryAlmostExhaustedcriticalhost free memory (MemAvailable)< 500 MB for 2 minOOM-kill risk to Postgres / API / sshd↗
BackupStalecriticalage of last successful pg_dump→S3> 24 h (tuned from 26 h)a daily backup run was missed → no recent restore point↗
PartitionArchiveStalecriticalage of last partition archive> 9 dayscold-partition offload stalled → DB + dumps grow unbounded↗
DiskAlmostFullwarningany DB-host filesystem< 10 % free for 5 mindisk filling → Postgres write failure at 0 %↗
PgDumpDurationDriftwarning7-day median pg_dump runtime> 4 h (tuned from 90 min)capacity drift — backups still succeed, but the dump is growing toward the 14 h timeout; schedule partitioning↗

Notes

(the old values fired on normal operation — chiefly PgDumpDurationDrift, which accounted for ~99 % of alert noise).

retired — they referenced metrics no longer scraped after the Vultr migration and could never fire. See alert-relationships.md.