Feedback-ID: id4a34324:Fastmail
Date: Fri, 27 Mar 2026 13:29:03 -0400
From: Andres Freund <andres@anarazel.de>
To: Melanie Plageman <melanieplageman@gmail.com>
Cc: Nazir Bilal Yavuz <byavuz81@gmail.com>,
	Thomas Munro <thomas.munro@gmail.com>, pgsql-hackers@postgresql.org,
 Peter Geoghegan <pg@bowt.ie>,
	Tomas Vondra <tv@fuzzy.cz>
Subject: Re: Don't synchronously wait for already-in-progress IO in read
 stream
Message-ID: <us6m5jdjxmveyan77tdg2tysjls4yu3tsohpvjjedz6pnjkyn7@2g6b2asvgug3>
References: <u73un3xeljr4fiidzwi4ikcr6vm7oqugn4fo5vqpstjio6anl2@hph6fvdiiria>
 <CAAKRu_bCZfWpmfmfUiQjDgFK12T8FUox-hV9YFDy6XHTh=i0ew@mail.gmail.com>
 <prkj4fv7dvyew4lbad2g5l346nn6u66inndwx7nzzap5c3bdmh@fds54c2hx3yd>
 <42rdu4q44kvsq53fz5qgzuawqpaytvnemsnquynlfch5mqhc2m@6ytnlgivtzro>
 <CAAKRu_YiaJgbmHGTFHRJEP9+AtVCt+gHTd1rmXm05agT_1Mx2A@mail.gmail.com>
 <rk56uxuqxi3udzu5oqc7qft5u7nncau5d2avlbui6obd7ybols@wlu2euvawj7t>
 <CAAKRu_bxajARZ0sXg_P=ZN76f1GofBmojE6HbQd7chjjhzWnRg@mail.gmail.com>
 <tjhnl6hiooo4aj6tktyehf4pn4yxcaqndlqebgpmynw4tmticq@odfcssnjf2qu>
 <CAAKRu_a8htJqvgMAMtA54zX35EZKhx3j7nLDXhksbciCxiNaJQ@mail.gmail.com>
 <wmqrglrv4eq2m2kzgbjbtmmuctklg7q4ylwo22rks7r7mlbf3w@dn6txz6znmtr>
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="4bovv3espbe3a3ho"
Content-Disposition: inline
In-Reply-To: <wmqrglrv4eq2m2kzgbjbtmmuctklg7q4ylwo22rks7r7mlbf3w@dn6txz6znmtr>
Archived-At: 
 <https://www.postgresql.org/message-id/us6m5jdjxmveyan77tdg2tysjls4yu3tsohpvjjedz6pnjkyn7%402g6b2asvgug3>
Precedence: bulk


--4bovv3espbe3a3ho
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

Hi,

I think I forgot to update the thread in my last message to note that I
had committed some of the preliminary changes.


On 2026-03-26 20:12:30 -0400, Andres Freund wrote:
> One test used did_io=(t|f). That was actually only needed once "aio: Don't
> wait for already in-progress IO" is in, as we might join the foreign IO. I
> chose to hide that by making that part of the query "did_io and not
> foreign_io", so we would detect if we were to falsely start IO ourselves.

I ended up not liking did_io, as that seems misleading when we just needed to
wait for a foreign IO.  I instead named it io_reqd.


> Still need to extend the test as part of the "don't wait" commit, to actually
> ensure that we reach the path for joining foreign IO.

That's done now.  I've added verification that we don't wrongly recognize
in-progress-ios without a wref as a foreign IO and an injection point based
test that verifies that we do see the foreign IO.

I've also done a bunch of cleanup in the commits. A few typos in commit
messages and the actual code changes and a few larger changes in the test code
& infrastructure. Mostly as part of allowing the aforementioned testing
(read_buffers() now only waits at the end, to make some of the tests
possible), but also just making the modified code a bit cleaner.

Greetings,

Andres Freund

--4bovv3espbe3a3ho
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v9-0001-test_aio-Add-basic-tests-for-StartReadBuffers.patch"