Upstream
a87bb96 introduced a new warning, visible at least on all ppc64el and
s390x builds which makes the post-build mariadb-test-test fail:
[Warning] InnoDB: native AIO failed: falling back to
innodb_use_native_aio=OFF
The case of this is the ci environment only where there is insufficent
fs.aio-max-nr configured. A such we are adding a suppression to the
mtr only. The warning is valuable for users. Upstream MDEV-37411 has
added a io_setup as a warning, which will fail in the CI environment,
so we're pre-emptively suppressing this warning too.
Unlike upstream MariaDB, in Debian we removed libaio in for Linux on
MariaDB 10.6+ in
612630c6 and completely in
1d648d6f. This was re-added
by upstream added a system variable innodb_linux_aio to control, and
also provided automated fallback. As such we're readding it back too.
With liburing still seccomp filterer in container environments having
a libaio fallback is more valuable than the threads implementation of
last resort.
Edited & Updated: Daniel Black <daniel@mariabdb.org>
Forwarded: https://jira.mariadb.org/browse/MDEV-37411 (rejected)
Upstream rejected this with the conclusion, is that it will start working in
Debian/Salsa/Ubuntu/Launchpad once the Linux kernel is new enough to handle the
I/O calls and fallbacks properly, or once the CI system increases the value of
/proc/sys/fs/aio-max-nr to 1 million or so, to allow multiple parallel MariaDB
servers (as started by mariadb-test-run) each have enough quota for native AIO
to work.
Gbp-Pq: Name MDEV-37411-suppress-new-warning-about-native-aio.patch
qr/error .*connecting to master/,
qr/InnoDB: Dumping buffer pool.*/,
qr/InnoDB: Buffer pool.*/,
+ qr/io_setup(%d) returned.*/,
+ qr/InnoDB: native AIO failed: falling back to innodb_use_native_aio=OFF/,
qr/InnoDB: Could not free any blocks in the buffer pool!/,
qr/InnoDB: innodb_open_files .* should not be greater than/,
qr/InnoDB: Trying to delete tablespace.*but there are.*pending/,