PostgreSQL Architecture

PostgreSQL Cluster / Instance is combination of Memory and Background process, cluster is nothing but collection / Group of databases managed by single instance of the PostgreSQL cluster 

Memory Components

Shared Memory 

Shared Buffer ==> This is the primary cache of PostgreSQL, where data blocks writes/reads from the disk. 

WAL Buffer ==> Buffer for Write ahead logging (WAL) data before it is written to disk.

CLOG Buffer ==> Store commit log information to manage transaction states

Temp Buffers ==> Used for storing temporary tables. Default 8MB

Work Memory ==> Used for sorting operations , Bitmap operation, hash join and merge join operations. Default 4MB 

Maintenance Work Memory ==> Used for vacuum and index operations. Default 64MB

Catalog Cache and Execution Memory ==> Memory used for caching system catalogs and                 executing SQL queries

Backend Process

Postmaster Process

    When we start PostgreSQL cluster ,the postmaster process is the first to initialize 

    Responsibilities of Postmaster

            Accepting incoming connection 

            Forks new backend process

            Starts background process

            Handles server restart and crash recovery 

BG Writer ==> Periodically Writes Dirty buffers/pages from shared buffer to the disk

WAL Writer ==> Writes contents of the WAL buffers to WAL files

Checkpoint ==> Writes dirty buffers to disk when a checkpoint occurs , this reduces the recovery time after a crash. 

Stats Collector ==> Collect DBMS usage statistics information such as session activity information(pg_stat_activity) and table usage statistics information (pg_stat_all_tables)

Gathers and maintains statistical data about the database, which the query planner uses to optimize query execution plans 

Archiver ==> When in archive log mode , copies the WAL files to the archive log location 

Sys Logger  ==> Logs error messages and a variety of information to log files

Auto vacuum Launcher ==> Fork the auto vacuum worker when vacuum is required(Automatically reclaims storage occupied by dead tuples( Rows that have updated or deleted), to prevent table bloat and maintains query performance.

Logical Replication Launcher ==> Manages the logical replication workers that handle replication of changes to other databases.

WAL Sender  ==> Sending WAL records to standby servers, runs on the primary server

WAL Receiver ==> Receive WAL records and apply on standby server , runs on the standby server