RDS for PostgreSQL memory

RDS for PostgreSQL memory is divided into shared and local.

Shared memory in RDS for PostgreSQL

RDS for PostgreSQL allocates shared memory when the instance starts. Shared memory is divided into multiple subareas. Following, you can find a description of the most important ones.

Shared buffers

The shared buffer pool is an RDS for PostgreSQL memory area that holds all pages that are or were being used by application connections. A page is the memory version of a disk block. The shared buffer pool caches the data blocks read from disk. The pool reduces the need to reread data from disk, making the database operate more efficiently.

Every table and index is stored as an array of pages of a fixed size. Each block contains multiple tuples, which correspond to rows. A tuple can be stored in any page.

The shared buffer pool has finite memory. If a new request requires a page that isn't in memory, and no more memory exists, RDS for PostgreSQL evicts a less frequently used page to accommodate the request. The eviction policy is implemented by a clock sweep algorithm.

The shared_buffers parameter determines how much memory the server dedicates to caching data.

Write ahead log (WAL) buffers

A write-ahead log (WAL) buffer holds transaction data that RDS for PostgreSQL later writes to persistent storage. Using the WAL mechanism, RDS for PostgreSQL can do the following:

When a client changes data, RDS for PostgreSQL writes the changes to the WAL buffer. When the client issues a COMMIT, the WAL writer process writes transaction data to the WAL file.

The wal_level parameter determines how much information is written to the WAL.

Local memory in RDS for PostgreSQL

Every backend process allocates local memory for query processing.

Work memory area

The work memory area holds temporary data for queries that performs sorts and hashes. For example, a query with an ORDER BY clause performs a sort. Queries use hash tables in hash joins and aggregations.

The work_mem parameter the amount of memory to be used by internal sort operations and hash tables before writing to temporary disk files. The default value is 4 MB. Multiple sessions can run simultaneously, and each session can run maintenance operations in parallel. For this reason, the total work memory used can be multiples of the work_mem setting.

Maintenance work memory area

The maintenance work memory area caches data for maintenance operations. These operations include vacuuming, creating an index, and adding foreign keys.

The maintenance_work_mem parameter specifies the maximum amount of memory to be used by maintenance operations. The default value is 64 MB. A database session can only run one maintenance operation at a time.

Temporary buffer area

The temporary buffer area caches temporary tables for each database session.

Each session allocates temporary buffers as needed up to the limit you specify. When the session ends, the server clears the buffers.

The temp_buffers parameter sets the maximum number of temporary buffers used by each session. Before the first use of temporary tables within a session, you can change the temp_buffers value.