Re: Don't synchronously wait for already-in-progress IO in read stream

Peter Geoghegan <pg@bowt.ie>

From: Peter Geoghegan <pg@bowt.ie>

To: Andres Freund <andres@anarazel.de>

Cc: pgsql-hackers@postgresql.org, Tomas Vondra <tv@fuzzy.cz>, Thomas Munro <thomas.munro@gmail.com>

Date: 2025-11-09T19:51:08Z

Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

bufmgr: Improve StartBufferIO interface
- 74eafeab1a57 19 (unreleased) landed
test_aio: Add basic tests for StartReadBuffers()
- 020c02bd9089 19 (unreleased) landed
test_aio: Add read_stream test infrastructure & tests
- 1f6f200cab67 19 (unreleased) landed
bufmgr: Make buffer hit helper
- df09452c3209 19 (unreleased) landed
bufmgr: Pass io_object and io_context through to PinBufferForBlock()
- c2a68e08b13f 19 (unreleased) landed
bufmgr: Restructure AsyncReadBuffers()
- 8a1a1d6ab862 19 (unreleased) landed
aio: Refactor tests in preparation for more tests
- 906a0469728e 19 (unreleased) landed

On Thu, Sep 11, 2025 at 5:46 PM Andres Freund <andres@anarazel.de> wrote:
> The problem leading to that is that we are currently synchronously waiting for
> IO on a buffer when AsyncReadBuffers() encounters a buffer already undergoing
> IO. If a block is read twice, that means we won't actually have enough IOs in
> flight to have good performance. What's worse, currently the wait is not
> attributed to IO wait (since we're waiting in WaitIO, rather than waiting for
> IO).

This patch no longer cleanly applies. Can you post a new version?

Thanks
-- 
Peter Geoghegan