Postgres Memory Management

1 post · Mar 2026 – Aug 2026

What this trail covers

Postgres memory management focuses on understanding the difference between cache and process memory (RSS). High cache usage is good, indicating data is readily available to the CPU. High RSS, however, signifies memory pressure and potential OOM risks. The post details how to investigate performance degradation by correlating Cluster Metrics, Query Insights, and EXPLAIN (ANALYZE, BUFFERS, MEMORY) to identify the root cause of RSS growth, emphasizing efficient connection pooling as a key strategy to reduce RSS.

All PlanetScale trails

Aug 2026The dangers of Postgres subtransactionsThis post introduces the concept of subtransaction cache overflow in PostgreSQL. It explains that when a transaction accumulates more than `PGPROC_MAX_CACHED_SUBXIDS` subtransactions (either via `SAVEPOINT` or PL/pgSQL exception blocks), the subtransaction cache overflows. This forces PostgreSQL to perform costly lookups in `pg_subtrans` for every scanned tuple within the affected snapshot's range. These lookups require SLRU read locks and potentially slow disk access, leading to cluster-wide performance degradation and preventing new read replicas from serving queries. The post also details how replicas stay in sync via WAL replay and the role of `RUNNING_XACTS` records in maintaining MVCC snapshots.
Jul 2026Every UPDATE Leaves a Ghost: MVCC, Bloat, and VACUUM in PostgreSQLThis post provides a deep dive into PostgreSQL's Multi-Version Concurrency Control (MVCC) implementation, explaining how tuples, transaction IDs (xmin, xmax), and snapshots work to enable readers and writers to operate concurrently without blocking each other. It details how updates result in a DELETE and INSERT operation, marking the old tuple with an `xmax` and creating a new version. The post also introduces subtransactions and their impact on transaction IDs and visibility, highlighting how excessive subtransactions can lead to cache overflows and performance degradation by forcing lookups in `pg_subtrans`.
Mar 2026High memory usage in Postgres is good, actuallyThis post clarifies that high memory usage in Postgres is often beneficial due to caching mechanisms (shared_buffers and OS page cache). It distinguishes between reclaimable cache memory and non-reclaimable RSS (Resident Set Size). High RSS is identified as the primary cause of memory pressure and OOM events. The post outlines a debugging strategy for performance degradation by correlating metrics from Cluster Metrics, Query Insights, and EXPLAIN (ANALYZE, BUFFERS, MEMORY) to pinpoint the source of RSS growth, and highlights efficient connection pooling (e.g., PgBouncer in transaction mode) as a method to reduce RSS.
Advertisement