Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Duplicate
-
10.5.1
-
None
Description
Running full test suite on a relative late 10.5.1 tree gave me the following
(was not able to repeat this)
encryption.innochecksum '8k,cbc,innodb,strict_full_crc32' w7 [ fail ]
Program terminated with signal SIGABRT, Aborted.
#0 0x00007f28357cdf14 in pthread_kill () from /lib64/libpthread.so.0
[Current thread is 1 (Thread 0x7f2835bc7880 (LWP 26235))]
#0 0x00007f28357cdf14 in pthread_kill () from /lib64/libpthread.so.0
#1 0x00000000015fccc8 in my_write_core (sig=6) at /my/maria-10.5/mysys/stacktrace.c:518
#2 0x0000000000bed102 in handle_fatal_signal (sig=6) at /my/maria-10.5/sql/signal_handler.cc:343
#3 <signal handler called>
#4 0x00007f28357cc90d in pthread_cond_wait@@GLIBC_2.3.2 () from /lib64/libpthread.so.0
#5 0x00007f2834052b4c in std::condition_variable::wait(std::unique_lock<std::mutex>&) () from /usr/lib64/libstdc++.so.6
#6 0x0000000001468c46 in tpool::waitable_task::wait (this=0x25c4320 <purge_coordinator_task>) at /my/maria-10.5/tpool/task.cc:64
#7 0x000000000122757f in srv_shutdown_purge_tasks () at /my/maria-10.5/storage/innobase/srv/srv0srv.cc:2232
#8 0x0000000001227a00 in srv_purge_shutdown () at /my/maria-10.5/storage/innobase/srv/srv0srv.cc:2302
#9 0x0000000001230903 in innodb_preshutdown () at /my/maria-10.5/storage/innobase/srv/srv0start.cc:2304
#10 0x0000000000bf147a in plugin_pre_shutdown (plugin=0x4d495a8) at /my/maria-10.5/sql/handler.cc:885
#11 0x00000000008cd1b4 in plugin_foreach_with_mask (thd=0x0, func=0xbf1439 <plugin_pre_shutdown(THD*, plugin_ref, void*)>, type=1, state_mask=10, arg=0x0) at /my/maria-10.5/sql/sql_plugin.cc:2452
#12 0x0000000000bf14a6 in ha_pre_shutdown () at /my/maria-10.5/sql/handler.cc:89
Attachments
Issue Links
- relates to
-
-
- Closed
-
-
MDEV-16678 Use MDL for innodb background threads instead of dict_operation_lock
-
- Closed
-
Activity
As noted in MDEV-21344, both Valgrind and MemorySanitizer report that memcmp() is being invoked on uninitialized data in dict_acquire_mdl_shared().
I am convinced that the MDEV-21344 fix also fixed the shutdown hangs. I have been running ./mtr --big-test on a 10.5-branch at least 5 times, maybe more than 10 times, on a branch that includes the fix, and did not see the hang once. Before the fix, I would observe a shutdown hang at least once in 5 or 10 runs.
We are aware of occasional InnoDB shutdown hangs in 10.5 ever since
MDEV-16678was pushed. thiru should try to repeat and fix this, and perhaps start by looking at the MemorySanitizer reports (MDEV-20377) for the memcmp() calls in dict_acquire_mdl_shared().