Transactions & connectionsstaff🔒 Pro concept

Subtransaction SLRU overflow (the 64-subxid cliff)

Simple terms

A backend can hold 64 subtransactions in fast shared memory. Open the 65th and it overflows, and the part that catches people is who pays for it. Not that backend. From that point on, other sessions' visibility checks may have to go to disk for something that used to be a memory read, and the tax lands across the whole cluster. Nobody has to type SAVEPOINT for this, either. Every PL/pgSQL block with an EXCEPTION clause quietly opens a subtransaction of its own.

You might be asked

A backend can cache only 64 active subtransaction XIDs. What exactly happens on the 65th, why does it silently slow down every OTHER session's visibility checks across the whole cluster (not just the backend that overflowed), and how would you prove and detect it in production? Where do subtransactions get created without anyone writing SAVEPOINT?

TopicSubtransactions / snapshots / pg_subtrans SLRU / scaling
PostgreSQL14, 15, 16, 17, 18
Tools usedpg_stat_slru, pg_stat_reset_slru(), pg_stat_force_next_flush(), EXPLAIN (ANALYZE, BUFFERS), two concurrent psql sessions (session.sh FIFO harness), source navigation @REL_17_10, PG 17.10
Last reviewed2026-06-19

Pro concept

Full answer and evidence depth sit behind Pro

You have the question and a plain-English lead. Pro unlocks the short answer, what the docs say, what the code does, the labeled lab evidence or reproduction protocol, how to use it under pressure, and the references.

ProFull answer + evidence depth

Unlock the full breakdown for Subtransaction SLRU overflow (the 64-subxid cliff)

You have the interview question and a plain-English lead. Pro opens the short answer, the manual walkthrough, the source decode, and a 55-line evidence section that states whether it is raw output, a captured run summary, or a protocol to run yourself, plus how to use it under pressure.
  • How you'd answer it, the full senior-level short answer
  • What the docs say, the manual's actual wording, with the citations
  • What the code does, the mechanism decoded from PostgreSQL's own source at a pinned tag
  • Proof from a real run, labeled as raw output, a captured run summary, or a run-it-yourself protocol
  • Using it under pressure, the situation, the call you'd make, what goes wrong, and what people get wrong
  • Version notes, what changed across PostgreSQL releases
  • 5 interviewer follow-ups for this concept

Card required. Cancel before day 7 and you are not charged.

Compare plans

How this was verified

The open teaser is the question and a plain-English lead. The short answer, the manual walkthrough, the source decode, the lab evidence, and how to use it under pressure unlock with Pro. Evidence is labeled as raw output, a captured run summary, or a run-it-yourself protocol.

Connected

Where this concept connects

How this concept links across the library, the interview questions that test it, its plain-English glossary definition, and the guided pathways it belongs to. Open the full map to explore further.

Open in the interactive map →