PostgreSQL architecture, drawn

Quick reference

• Tier 1 · Client: Application (APP) · Connection pooler (POOL)
• Tier 2 · Processes: Postmaster (POST) · Backend process (BACK) · WAL writer (WAL) · Checkpointer (CKPT)
• Tier 3 · Storage: Heap files (HEAP) · WAL segments (WAL) · WAL archive (ARCH)

Request flow: Application → Connection pooler → Postmaster → Backend process → WAL writer → WAL segments → WAL archive; Backend process → Heap files; Checkpointer → Heap files

Explained

Client The application opens a connection, and a pooler keeps a small set of server connections warm. PostgreSQL forks a backend process per connection, so each idle connection still costs memory; the pooler exists to stop that cost growing with every application worker.

Processes The postmaster listens on the port and forks a backend for each client. The backend parses, plans and executes the query, reads pages through shared buffers, and writes WAL buffers before the change reaches the heap. The WAL writer flushes those buffers into WAL segments. The checkpointer writes dirty shared buffers back to the data files and updates the control file.

Storage Heap files hold table pages in 8 kB blocks, and indexes live in their own files. WAL segments hold the redo stream, 16 MB each by default, and the WAL archive keeps completed segments for point-in-time recovery and for rebuilding a standby. The archive is written by archive_command, not by the checkpointer.

In one line

Check the archiver

SELECT * FROM pg_stat_archiver;