Parameters

Write-Ahead Log

40 PostgreSQL 18.6 parameters. Each one has its own animation and explanation.

Archive Recovery

archive_cleanup_command: Command run at each restartpoint on a standby, to prune archived WAL it no longer needs.
recovery_end_command: Shell command run once when archive recovery or standby mode ends.
restore_command: Shell command that fetches an archived WAL segment during recovery (%f, %p).

archive_command: Shell command that copies a finished WAL segment to the archive (%p, %f).
archive_library: Archive WAL with a loadable module instead of a shell command (PostgreSQL 15+).
archive_mode: Turn on WAL archiving (on), or also archive on a standby (always).
archive_timeout: Force a WAL segment switch after this long, so quiet periods still get archived.

checkpoint_completion_target: Fraction of the checkpoint interval used to spread the writes (default 0.9).
checkpoint_flush_after: Ask the OS to begin checkpoint writeback in smaller batches; the default is platform-dependent.
checkpoint_timeout: Longest time between automatic checkpoints (default 5 min).
checkpoint_warning: Log a warning if WAL-triggered checkpoints come closer together than this.
max_wal_size: Soft limit on pg_wal size; checkpoints are started early enough to stay below it.
min_wal_size: Keep at least this much WAL around for reuse, so bursts do not create new files.

recovery_prefetch: Read ahead in WAL during recovery and prefetch data blocks it will need.
wal_decode_buffer_size: How far ahead in WAL recovery may decode to find blocks to prefetch.

recovery_target: Only value: 'immediate', i.e. stop as soon as the restored backup is consistent.
recovery_target_action: What to do once the recovery target is reached: pause, promote or shutdown.
recovery_target_inclusive: Stop just after (on) or just before (off) the time/xid/LSN target.
recovery_target_lsn: Replay WAL up to this exact WAL position (LSN).
recovery_target_name: Replay WAL up to a named restore point created with pg_create_restore_point().
recovery_target_time: Replay WAL up to this timestamp (point-in-time recovery).
recovery_target_timeline: Which timeline to follow during recovery: 'latest' (default), 'current' or an ID.
recovery_target_xid: Replay WAL up to the commit of this transaction ID.

commit_delay: Microseconds a committing session waits so others can share the same WAL flush.
commit_siblings: Minimum other active transactions before commit_delay is applied.
fsync: Never turn off in production: without fsync a crash can corrupt the whole cluster.
full_page_writes: Log a whole page image on first change after a checkpoint, to repair torn pages.
synchronous_commit: What COMMIT waits for: nothing, local disk, or standby write/flush/replay.
wal_buffers: Shared memory for WAL not yet written out. -1 = 1/32 of shared_buffers (max 16MB).
wal_compression: Compress full-page images in WAL with pglz, lz4 or zstd.
wal_init_zero: Fill new WAL segment files with zeroes before use (off can help on copy-on-write FS).
wal_level: How much information goes into WAL: crash-only, replicas, or logical decoding.
wal_log_hints: WAL-log pages changed only by hint bits; needed for pg_rewind without checksums.
wal_recycle: Reuse old WAL segment files by renaming them instead of deleting and recreating.
wal_skip_threshold: With wal_level=minimal, new relations smaller than this are WAL-logged, bigger are fsynced.
wal_sync_method: Which system call forces WAL to disk at commit (fdatasync on Linux by default).
wal_writer_delay: How often the WAL writer wakes to write and flush WAL in the background.
wal_writer_flush_after: How much WAL the WAL writer may write before it must also flush (fsync).

summarize_wal: Run the WAL summarizer, needed for incremental backups (PostgreSQL 17+).
wal_summary_keep_time: How long WAL summary files are kept before automatic deletion (default 10 days).