Testing a fix of MDEV-15326 revealed that if the call to trx->lock.table_locks.clear(); is skipped at transaction commit when trx->lock.trx_locks is empty, occasionally a transaction would be left with nonempty trx->lock.table_locks list. This proves that the table_locks is not a subset of trx_locks, like it is expected to be.

We will need a test case for this, so that the bug can be found and fixed. To repeat the assertion failures, revert the following part of the MDEV-15326 fix:

diff --git a/storage/innobase/trx/trx0trx.cc b/storage/innobase/trx/trx0trx.cc

index 1dcca7c1f72..b06e15d4f37 100644

--- a/storage/innobase/trx/trx0trx.cc

+++ b/storage/innobase/trx/trx0trx.cc

@@ -561,8 +561,6 @@ inline void trx_t::release_locks()

   if (UT_LIST_GET_LEN(lock.trx_locks))

     lock_trx_release_locks(this);

-  else

-    lock.table_locks.clear(); /* Work around MDEV-20483 */

 }

 /********************************************************************//**

This should lead to occasional assertion failures on trx->lock.table_locks.empty() on transaction commit.