Pre-MDL versions had direct relationship between LOCK_open and
LOCK_global_system_variables, e.g.:
intern_sys_var_ptr // locks LOCK_global_system_variable
mysql_sys_var_char
create_options_are_valid
ha_innobase::create
handler::ha_create
ha_create_table
rea_create_table
mysql_create_table_no_lock // locks LOCK_open
mysql_create_table
With MDL this relationship was removed, but mutex order was still
recorded. In fact there is indirect relationship between LOCK_open
and LOCK_global_system_variables via rwlocks in reverse order.
Removed LOCK_open and LOCK_global_system_variables order recording,
instead assert that LOCK_open is never held in intern_sys_var_ptr().
This solves only one of many problems detected with MDEV-5089.
- There was List<N>::nth_element() but it didn't work because linker removed it.
- Now, removal by linker is prevented for important values of T, and function is renamed.
(This is attempt at fix#2) (re-commit with fixed typo)
- Moved the testcase from partition_test to partition_innodb.test where it can really work.
- Made ordered index scans over ha_partition tables to satisfy ROR property for
the case where underlying table uses extended keys.
The problem was that a big record was allocated on the stack, which casued stack to run out.
Fixed by using my_safe_alloca() instead of my_alloca() when allocating records.
Now only records <= 16384 are allocated on the stack.
mysql-test/r/stack-crash.result:
Added test case
mysql-test/t/stack-crash.test:
Added test case
storage/maria/ma_blockrec.c:
Use my_safe_alloca() instead of my_alloca()
storage/maria/ma_dynrec.c:
Use my_safe_alloca() instead of my_alloca()
storage/maria/maria_def.h:
Added MARIA_MAX_RECORD_ON_STACK
storage/maria/maria_pack.c:
Use my_safe_alloca() instead of my_alloca()
- MariaDB-5.5 part of the fix: since we can't easily fix query optimization for I_S tables,
run the affected-tablespaces query with semijoin=off. It happens to have a good query plan
with that setting.
That error 'Can't open the pid file' leads to mysqld crash signal 11 in mysql_audit_general() called
with the 'thd' parameter set to NULL. That wasn't checked when the thd->db and thd->db_length
were accessed. Fixed by checking for the NULL thd.
file storage.
Analysis: posix_fallocate was called using 0 as offset and len as
desired size. This is not optimal for SSDs.
Fix: Call posix_fallocate with correct offset i.e. current file size
and extend the file from there len bytes.
on debug build when innodb_use_fallocate=1
Analysis: There was merge error that caused additional call to
fil_node_complete_io.
Fixed by removing incorrect call and fixed calculation of extended
file size.
In ::position() federated needs to know an element before the
current ("data_cursor") in the single-linked list. Replace list
traversal for every ::position() call (which is O(n^2)) with remembering
the current element before it's advanced by mysql_fetch_row().
storage/federatedx/federatedx_io_mysql.cc:
mdev:5698
On fedora:
1. provide/obsoleve mariadb-* packages
2. MariaDB-common conflicts with mariadb-libs (on filesystem level),
but does not provide or obsolete it.
An attempt to introduce libmysqlclient.so symbol versioning that is
compatible both with Debian and Fedora all versions: put all symbols into
libmysqlclient_18 version node (as on Debian), but also put aliases of
old symbols into libmysqlclient_16 version node (as on Fedora).
Also use a linker script to create aliases of exported symbols, not
rpm_support.cc source file.
It helps to interpret valgrind/safemalloc memory-related warnings that are
printed when a plugin is unloaded (and thus cannot resolve addresses automatically)
actually consumed it. With this patch evaluated query buffer is freed
along with query buffer.
The problem was uncovered by udf_debug_sync.test when it was run with
--embedded.
MDEV-5555: Incorrect index_merge on BTREE indices
- In ha_partition, make ordered index reads return rows in rowid order
when index columns are the same.
