MDEV-29166: reduce locking of innodb trx_sys_t::history_*
Acquire and release locks in trx_sys_t::history_* one by one so
we reduce the risk of deadlock with a process that acquires a
later segement and then aquires and earlier segment.
While doing this means that the accuracy of the returned history
length could be off by a factor, but by the time we release from
these microfunctions, the rseg locks are released, so the history
value is out of date by the time the calling function access it already.
MDEV-29166: reduce locking of innodb rseg for user exposure of innodb_history_list_length
SHOW ENGINE INNODB STATUS and SHOW GLOBAL VARIABLES were blocking on the locks used
to access the history length in MDEV-29141. While the reason for the blockage
was elsewhere, we should make these monitoring commands less blocking as there
is a trx_sys.history_size_approx function that can be used.
SHOW ENGINE INNODB STATUS and SHOW GLOBAL STATUS LIKE
'innodb_history_list_length' and Innodb Monitors can use
trx_sys.history_size_approx().
MDEV-21136 InnoDB's records_in_range estimates can be way off
Get rid of BTR_ESTIMATE and btr_cur_t::path_arr.
Before the fix btr_estimate_n_rows_in_range_low() used two
btr_cur_search_to_nth_level() calls to create two arrays of tree path,
the array per border. And then it tried to estimate the number of rows
diving level-by-level with the array elements. As the path pages are
unlatched during the arrays iterating, the tree could be modified, the
estimation function called itself until the number of attempts exceed.
After the fix the estimation happens during search process. Roughly, the
algorithm is the following. Dive in the left page, then if there are pages
between left and right ones, read a few pages to the right, if the right
page is reached, fetch it and count the exact number of rows, otherwise
count the estimated number of rows, and fetch the right page.
The latching order corresponds to WL#6326 rules, i.e.:
(2.1) [same as (1.1)]: Page latches must be acquired in descending order
of tree level.
(2.2) When acquiring a node pointer page latch at level L, we must hold
the left sibling page latch (at level L) or some ancestor latch
(at level>L).
When we dive to the level down, the parent page is unlatched only after
the the current level page is latched. When we estimate the number of rows
on some level, we latch the left border, then fetch the next page, and
then fetch the next page unlatching the previous page after the current
page is latched until the right border is reached. I.e. the left sibling
is always latched when we acquire page latch on the same level. When we
reach the right border, the current page is unlatched, and then the right
border is latched. Following to (2.2) rule, we can do this because the
right border's parent is latched.
Fix Ninja builds on Gitlab-CI by limiting parallelism
In previous versions it was stated that MDEV-25968 was causing other
jobs in the pipeline to fail if not run with "-j 2" but this bug was not
affecting fedora-ninja. This is still true for the public gitlab runners.
However, running the fedora-ninja job on custom runners with more processors
without the "-j 2" flag will cause the compiler to crash.
When running the build with 2,4,8,16,32 threads, build times were
consistent indicating that the typical bottleneck is I/O and not CPU
cores. Therefore, "-j 2" is not a big drawback and it was chosen in
order to remain consistent with the other builds affected by MDEV-25968.
All new code of the whole pull request, including one or several files
that are either new files or modified ones, are contributed under the
BSD-new license. I am contributing on behalf of my employer Amazon Web
Services, Inc.
MDEV-26294 Duplicate entries in unique index not detected when changing collation
Problem:
=======
ALTER TABLE in InnoDB fails to detect duplicate entries
for the unique index when the character set or collation of
an indexed column is changed in such a way that the character
encoding is compatible with the old table definition.
In this case, any secondary indexes on the changed columns
would be rebuilt (DROP INDEX, ADD INDEX).
Solution:
========
During ALTER TABLE, InnoDB keeps track of columns whose collation
changed, and will fill in the correct metadata when sorting the
index records, or applying changes from concurrent DML.
This metadata will be allocated in the dict_index_t::heap of
the being-created secondary indexes.
The fix was developed by Thirunarayanan Balathandayuthapani
and simplified by me.