binlog.mdev_29803 fails: mariadb-binlog exec failure (errno 38) reading local binlog files after flush binary logs

binlog.mdev_29803 fails in buildbot on release branches 10.11, 11.4 and 11.8, on the local-file mariadb-binlog invocation that runs right after "flush binary logs;" in the test.

Build Run Info

Failure text

Captured output at the point of failure, identical across all matched release-branch occurrences:

reset master;

create table t1 (a int);

create table t2 (a int);

drop table t1, t2;

flush binary logs;

drop table if exists t1, t2;

Warnings:

Note	1051	Unknown table 'test.t1,test.t2'

flush binary logs;

The same test, same window, on non-release (PR) branches captured a fuller error for what is evidently the same failure:

mysqltest: At line 32: exec of 'mariadb-binlog --defaults-file=... --local-load=... master-bin.000001 master-bin.000002 > /dev/null 2>&1' failed, error: 256, status: 1, errno: 38

Output from before failure:

flush binary logs;

Analysis

All 9 matching occurrences in the window (across 10.11, 11.4 and 11.8) occur on a single builder, amd64-msan-clang-22, pointing at this specific MSAN test/build path rather than a broad per-branch regression. errno 38 is ENOSYS ("Function not implemented"), an unusual failure mode for exec'ing the mariadb-binlog client binary itself rather than a GTID/content mismatch inside the test.

This test file is the regression test for MDEV-29803 (Closed/Fixed - an out-of-order GTID bug in mysqlbinlog, failing with errno 11, fixed in 10.11.19/11.4.13/11.8.9/12.3.3/13.0.2). Today's failure has a different errno and no GTID-ordering message, so this looks like a distinct, new failure recurring in the same test file rather than a reopening of the original bug - noted here as related context, not a duplicate.

Created via the mtr-failure-bug skill.