FAQs

Reload re-reads postgresql.conf for many parameters. Restart recycles all processes and is required for memory-related GUCs.

Depends on RAM and workload. Hundreds of idle connections are cheap; hundreds of active queries are not. Pool aggressively.

The data directory (PGDATA) holds heap files, WAL, and catalog. Use tablespaces only when you need separate storage paths.

No for backends. Parallel query workers are separate processes forked for a single statement.

On startup, Postgres replays WAL from the last checkpoint, then accepts connections.

Yes. Catalogs describe the cluster; most queries filter to current_database() objects.

DDL that changes user tables updates catalogs in the same transaction.

Optional comment storage tied to catalog OIDs via objoid/classoid.

pg_constraint with contype = 'f' or information_schema.table_constraints.

Slightly more view layers. For heavy introspection, pg_catalog joins are fine.

Hundreds are fine. Complexity comes from grants and migration order, not OID cost.

Schema names are per database. Same name in different DBs is unrelated.

A placeholder for a schema matching the role name; often empty.

Many installs use public. Some teams dedicate an extensions schema.

Yes for types and functions resolved for unqualified calls.

Yes for teams tracking stable releases; validate extensions and workload plans in staging.

Five years from its initial release per community policy; confirm current EOL dates on postgresql.org.

Usually no. Reindex if pg_upgrade reports invalid indexes or after collation updates.

Brief restart disconnects sessions; use pooler failover or maintenance window.

Treat version policy separately; this site documents community PostgreSQL.

Docker Postgres 18 plus pinned migrations; still use timestamptz and constraints.

Yes. Minor skew is OK; major skew hides incompatibilities.

Quarterly quick pass; full ADR review yearly.