Horton

Re: [PATCH] OAuth: fix performance bug with stuck multiplexer events

Jacob Champion <jacob.champion@enterprisedb.com>

View thread
From: Jacob Champion <jacob.champion@enterprisedb.com>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Daniel Gustafsson <daniel@yesql.se>, Peter Eisentraut <peter@eisentraut.org>
Date: 2025-08-06T23:55:44Z
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 →

  1. oauth: Always link with -lm for floor()

  2. oauth: Add unit tests for multiplexer handling

  3. oauth: Ensure unused socket registrations are removed

  4. oauth: Remove expired timers from the multiplexer

  5. oauth: Remove stale events from the kqueue multiplexer

  6. oauth: Track total call count during a client flow

Attachments

On Wed, Aug 6, 2025 at 9:13 AM Jacob Champion
<jacob.champion@enterprisedb.com> wrote:
> Maybe "drain" would no longer be the
> verb to use there.

I keep describing this as "combing" the queue when I talk about it in
person, so v3-0001 renames this new operation to comb_multiplexer().
And the CI (plus the more strenuous TLS tests) confirms that the
callback count is still stable with this weaker guarantee, so I've
gotten rid of the event-counting code.

Now that I'm no longer counting events, I can collapse the changes to
register_socket(). I can't revert those changes entirely, because then
we regress the case where Curl switches a socket from IN to OUT (this
is enforced by the new unit tests). But I'm not sure that the existing
comment adequately explained that fix anyway, and I didn't remember to
call it out in my initial email, so I've split it out into v3-0002.
It's much smaller.

The tests (now in 0005) have been adjusted for the new "combing"
behavior, and I've added a case to ensure that multiple stale events
are swept up by a single call to comb_multiplexer().

Thanks!
--Jacob