PostgreSQL WAL basics: generation, retention, and safe checks

Before PostgreSQL writes a changed data page, it first records enough information in the write-ahead log to redo the change. Sequential WAL writes support crash recovery, continuous archiving, point-in-time recovery, and replication.

Three questions must stay separate: how many WAL bytes the workload generates, how many segment files PostgreSQL currently retains, and how much archived or provider-managed storage those records occupy.

pg_stat_wal returns one cluster-wide row. Its counters are cumulative since stats_reset, so take two timestamped observations and compare compatible deltas to estimate a rate.

SELECT
  wal_records,
  wal_fpi,
  wal_bytes,
  wal_buffers_full,
  stats_reset
FROM pg_stat_wal;

A slot's restart_lsn identifies the oldest WAL it might still require. This read-only query exposes slot state without calculating a provider-specific disk value.

SELECT
  slot_name,
  slot_type,
  database,
  active,
  restart_lsn,
  confirmed_flush_lsn,
  wal_status,
  safe_wal_size
FROM pg_replication_slots
ORDER BY active, slot_name
LIMIT 50;

Supabase describes WAL as a separate component of disk usage. Neon routes WAL through Safekeepers and derives pages through its storage service, so assumptions based on access to a traditional server filesystem are not portable. Follow the provider's supported controls and definitions instead of translating a PostgreSQL counter directly into cost.

• Do not reset WAL statistics just to make a graph easier to read; preserve the observation window.
• Do not assume an inactive slot is abandoned or safe to drop.
• Do not treat logical database size as the size of WAL or archived backups.
• Do not change retention, checkpoint, or archiving settings from one snapshot.

When pg_stat_statements exposes WAL counters, the analyzer can rank normalized query fingerprints by their cumulative WAL bytes and records. That can identify statement-level contributors within the extension's statistics window without collecting SQL text.

DatabaseMeter does not currently collect cluster-wide pg_stat_wal totals or replication-slot retention. Its logical database-size trend also does not represent WAL directory or archive size, so confirm retention with PostgreSQL and provider-native evidence.