Hi, Aleksey! On Nov 06, Aleksey Midenkov wrote:

revision-id: 503e494ec6e103b23dc55f8fc8ae68ecec8b8316 (versioning-1.0.6-1-g503e494ec6e) parent(s): 6120ae4acafb23badecc623e33797949e3f6fbbb author: Aleksey Midenkov <midenok@gmail.com> committer: Aleksey Midenkov <midenok@gmail.com> timestamp: 2018-10-22 11:50:10 +0300 message:

MDEV-16241 Assertion `inited==RND' failed in handler::ha_rnd_end()

Discrepancy in open indexes due to overwritten `read_partitions` upon `ha_open()` in `ha_partition::clone()`.

[Fixes tempesta-tech/mariadb#551]

diff --git a/mysql-test/main/partition_innodb.result b/mysql-test/main/partition_innodb.result index d27abc07984..c63cbc4601d 100644 --- a/mysql-test/main/partition_innodb.result +++ b/mysql-test/main/partition_innodb.result @@ -925,3 +925,14 @@ test_jfg test_jfg11 test_jfg test_jfg12#P#p1000 test_jfg test_jfg12#P#pmax DROP DATABASE test_jfg; +# +# MDEV-16241 Assertion `inited==RND' failed in handler::ha_rnd_end() +# +CREATE TABLE t1 (pk INT PRIMARY KEY, x INT, y INT, z INT, KEY (x), KEY (y, z)) +WITH SYSTEM VERSIONING +PARTITION BY SYSTEM_TIME (PARTITION p1 HISTORY, PARTITION pn CURRENT); +INSERT INTO t1 VALUES (1, 7, 8, 9), (2, NULL, NULL, NULL), (3, NULL, NULL, NULL); +SELECT COUNT(*) FROM t1 WHERE x IS NULL AND y IS NULL AND z IS NULL; +COUNT(*) +2 +DROP TABLE t1;

please, add a non-system-versioning test too. It's in one of the comments in the bug report.

diff --git a/sql/ha_partition.cc b/sql/ha_partition.cc index 40380ffeec5..901f812a0e6 100644 --- a/sql/ha_partition.cc +++ b/sql/ha_partition.cc @@ -3725,6 +3725,9 @@ void ha_partition::rebind_psi() handler *ha_partition::clone(const char *name, MEM_ROOT *mem_root) { ha_partition *new_handler; + MY_BITMAP read_partitions; + MY_BITMAP lock_partitions; + int error;

DBUG_ENTER("ha_partition::clone"); new_handler= new (mem_root) ha_partition(ht, table_share, m_part_info, @@ -3748,9 +3751,21 @@ handler *ha_partition::clone(const char *name, MEM_ROOT *mem_root) ALIGN_SIZE(m_ref_length)*2))) goto err;

- if (new_handler->ha_open(table, name, + DBUG_ASSERT(m_tot_parts == m_part_info->read_partitions.n_bits); + DBUG_ASSERT(m_tot_parts == m_part_info->lock_partitions.n_bits); + my_bitmap_init(&read_partitions, NULL, m_tot_parts, FALSE); + my_bitmap_init(&lock_partitions, NULL, m_tot_parts, FALSE); + bitmap_copy(&read_partitions, &m_part_info->read_partitions); + bitmap_copy(&lock_partitions, &m_part_info->lock_partitions); + error= new_handler->ha_open(table, name, table->db_stat, - HA_OPEN_IGNORE_IF_LOCKED | HA_OPEN_NO_PSI_CALL)) + HA_OPEN_IGNORE_IF_LOCKED | HA_OPEN_NO_PSI_CALL); + bitmap_copy(&m_part_info->read_partitions, &read_partitions); + bitmap_copy(&m_part_info->lock_partitions, &lock_partitions); + my_bitmap_free(&read_partitions); + my_bitmap_free(&lock_partitions); + + if (error)

I don't think this is needed. ha_partition::open() is aware of clones, it has `if (m_is_clone_of)` to do some stuff only for clones and some only for fresh opens. Perhaps you just need to move the bitmap-resetting code to be done only for fresh opens? Regards, Sergei Chief Architect MariaDB and security@mariadb.org