Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.5, 10.6, 10.7(EOL), 10.8(EOL)
Description
The log resizing test occasionally fails like this:
|
10.8 972b45642afee364cd140f17ae46c00cd0af6e11 |
innodb.log_file_size '4k,innodb' w5 [ fail ]
|
…
|
2022-02-14 17:45:38 0 [Note] InnoDB: Starting final batch to recover 4 pages from redo log.
|
2022-02-14 17:45:38 0 [Note] InnoDB: Resizing redo log from 12.000MiB to 5.000MiB; LSN=24674766
|
mariadbd: /home/buildbot/maria-slave/fulltest-debug-big-only/build/storage/innobase/srv/srv0start.cc:852: lsn_t srv_prepare_to_delete_redo_log_file(): Assertion `!buf_pool.any_io_pending()' failed.
|
I was able to reproduce this failure locally by running this test 120 times in parallel with itself (3 times the number of hardware threads reported by nproc).
In the core dump, the assertion would have held. Therefore, it looks like we must actually protect the reads in this debug assertion with buf_pool.mutex in order to avoid false alarms. With the following patch, I am not able to reproduce any crash. The function is only being called by debug assertions (which are not present in release builds).
diff --git a/storage/innobase/include/buf0buf.h b/storage/innobase/include/buf0buf.h
|
index 89987b05fdb..fec8dcb4e2a 100644
|
--- a/storage/innobase/include/buf0buf.h
|
+++ b/storage/innobase/include/buf0buf.h
|
@@ -2042,9 +2042,14 @@ class buf_pool_t
|
buf_tmp_buffer_t *io_buf_reserve() { return io_buf.reserve(); }
|
|
/** @return whether any I/O is pending */
|
- bool any_io_pending() const
|
+ bool any_io_pending()
|
{
|
- return n_pend_reads || n_flush_LRU() || n_flush_list();
|
+ if (n_pend_reads)
|
+ return true;
|
+ mysql_mutex_lock(&mutex);
|
+ const bool any_pending{n_flush_LRU_ || n_flush_list_};
|
+ mysql_mutex_unlock(&mutex);
|
+ return any_pending;
|
}
|
/** @return total amount of pending I/O */
|
ulint io_pending() const |