PostgreSQL in Practice

PostgreSQL Architecture

PostgreSQL 18 uses a process-per-connection server around shared memory, durable relation files, catalogs, and WAL. That architecture makes session count, cache behavior, checkpoints, and background maintenance application-visible operational concerns.

The postmaster accepts connections and supervises child processes.
Shared buffers mediate database-page access.
Catalogs are ordinary, transactionally visible relations.
Relation forks separate concerns.
WAL decouples commit durability from heap-page flushing.
Auxiliary processes protect liveness and operability.

Connection, cache, storage, and durability flow

Layer and consequence
Layer	Responsibility	Operational consequence
Backend	parse, plan, execute, session state	pool sessions; avoid unbounded connection admission
Shared buffers	shared page cache and dirty state	inspect workload and I/O before changing size
Relation files/forks	heap, index, FSM, VM, unlogged initialization	use SQL/catalog tools, never manipulate files
WAL/checkpoint	durability and restart frontier	observe WAL rate, checkpoint I/O, flush latency
Workers	vacuum, archive, replication, parallel work	reserve capacity and alert on stalled work

Inspect the running PostgreSQL 18 instance

SELECT version();
SELECT name, setting, unit, source, sourcefile
FROM pg_settings
WHERE name IN ('shared_buffers','max_connections','checkpoint_timeout');

SELECT backend_type, count(*)
FROM pg_stat_activity
GROUP BY backend_type ORDER BY backend_type;