Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-16142

Merge new release of InnoDB MySQL 5.7.22 to 10.2

    Details

      Description

      The following changes between MySQL 5.7.21 and MySQL 5.7.22 touch InnoDB files.

      (empty commit message) Bug 27122803 - BACKPORT FIX FOR BUG 25899959 TO MYSQL-5.7
      This fixes a bug in ALTER TABLE that adds an indexed virtual column. A test case is included.

      BUG#27189701 INNODB: EARLY RELEASE GAP LOCKS FOR RC ISOLATION
      This is not a bug fix, but a performance improvement to release gap locks for transactions that are in the XA PREPARE state. This looks like a simple optimization that could be useful with distributed transactions.

      However, compared to MySQL 5.7, MariaDB supports innodb_lock_schedule_algorithm=vats, and it is the default setting. If we wanted to take this patch from MySQL, we should look at the 8.0 version of it, which introduced LOCK_CATS_THRESHOLD. Furthermore, we should ensure that this does not break the locking of SPATIAL INDEX.

      Bug #26731689 FK ON TABLE WITH GENERATED COLS: ASSERTION POS < N_DEF
      A bug fix for FOREIGN KEY and VIRTUAL columns (MDEV-5800), with test case.

      Bug #27212129 [ERROR] INNODB: A RECORD LOCK WAIT HAPPENS IN A DICTIONARY OPERATION. INDEX `ID_
      This is a regression from a MySQL 5.7.21 bug fix related to native InnoDB partitioning, which MariaDB does not have. It looks like the fix is renaming statistics for all partitions in a single transaction, using a peculiar mechanism that involves quite a bit of memory allocation. This change does is not applicable to MariaDB.

      Bug #26805833 INNODB COMPLAINS OF SYNTAX ERROR, BUT DOES NOT SAY WHICH OPTION
      This adds an error message for refusing startup.

      Bug #25076416 VIRTUAL COLUMN IS NOT CONSIDERED WHILE FETCHING THE AUTOINC COLUMN
      I removed the buggy function in MDEV-6076 (MariaDB 10.2.4) and properly handled virtual columns.

      Bug #27041445 SERVER ABORTS IF FTS_DOC_ID EXCEEDS FTS_DOC_ID_MAX_STEP
      This has been merged from 5.6.40 to 10.0.

      Bug #26935001 ALTER TABLE AUTO_INCREMENT TRIES TO READ INDEX FROM DISCARDED TABLESPACE
      I filed and fixed this as MDEV-16145. The code was refactored as part of MDEV-6076, so the patch does not apply directly. Oracle did not publish a test case, but creating one was trivial based on the bug title.

      Bug#26105334 WARNINGS WHEN BUILD WITH GCC 7.1.1.
      I fixed many GCC 7 warnings (including this -Wimplicit-fallthrough) several months earlier. This is not applicable to MariaDB.

      Bug #26958695 INNODB NESTED STORED FIELD WITH CONSTRAINT KEY PRODUCE BROKEN TABLE
      There is no test case. This is fixing something with FOREIGN KEY and virtual columns.
      According to Thirunarayanan Balathandayuthapani we do not need this fix in MariaDB. I will add his test case instead.

      Bug#27399897 PUNCH HOLE IS MISSED IN FIRST COMPRESSION OF PAGES.
      This applies to the MySQL 5.7 reimplementation of the MariaDB 10.1 page_compression. The code is different, and there is no test case. Furthermore, MariaDB does not have any field skip_punch_hole.

      Bug #27304661 MYSQL CRASH DOING SYNC INDEX ] [FATAL] INNODB: SEMAPHORE WAIT HAS LASTED > 600
      This claims to fix a hang with FULLTEXT INDEX, and there is a test case.

      Bug #27141613 ASSERTION: TRX0REC.CC:319:COL->IS_VIRTUAL() / CRASH IN TRX_UNDO_READ_V_COLS
      This appears to fix an inconsistency in the error handling related to CREATE INDEX on virtual columns (MDEV-5800). There is no test case, but it seems possible to create one based on the description.

      BUG#27272806 ASSERT DURING REPLICAITON:: LOCK == __NULL BUG#27294066 INNODB: FAILING ASSERTION: !(TYPE_MODE & 1024)
      This looks like a regression from the XA PREPARE performance improvement, which we will not merge to MariaDB for now.

      Bug #26300119 ASSERTION IN PREBUILT->TRX->ISOLATION_LEVEL == TRX_ISO_READ_UNCOMMITTED
      This is fixing one more regression from the MySQL 5.6 change that we decided not to merge, in MDEV-11751. There is a test case, so we can easily check that MariaDB is not affected.
      The fix essentially relaxes a non-debug assertion and adds some DEBUG_SYNC instrumentation.
      The added DEBUG_SYNC in the file row0sel.cc is in the code that we did not merge, so the test will just hang.
      We can skip this patch altogether.

      FOLLOW-UP FIX FOR BUG#27141613
      This fixes a regression caused by the above CREATE INDEX bug fix. The commit comment refers to a private test name.

      Bug#26696448 REPLAY LOG RECORD CAUSE MYSQLD CRASH DURING ONLINE DDL
      This has been merged from 5.6.40 to 10.0. There is no test case.
      It is possible that a similar bug exists in row_log_apply_ops() but has not been fixed. Maybe we should run RQG tests of ADD INDEX with larger tables? Or we should manually create a test where the critical log size is reached.

      Bug #27158030 INNODB ONLINE ALTER CRASHES WITH CONCURRENT DML
      This fixes one more problem with online rebuilding ALTER TABLE when indexed virtual columns exist.
      In MariaDB 10.2.11 (MDEV-13795/MDEV-14332) I took a more conservative approach and disabled the online ALTER altogether for this case. The patch does not apply to MariaDB, because the code was removed by my patch. There is no test case.

      (empty commit message)
      This has been merged from 5.6.40 to 10.0.

      Bug #23593654 CRASH IN BUF_BLOCK_FROM_AHI WHEN LARGE PAGES AND AHI ARE ENABLED
      This fixes a bug in buffer pool resizing.

      Bug#27361662 ERROR NOT THROWN ON CREATING TEMPORARY TABLE IN INNODB_FILE_PER_TABLE TABLESPACE
      This does not apply to MariaDB. Any TABLESPACE attribute would be ignored, and MariaDB 10.2 would always create temporary tables in the shared temporary tablespace.

      Bug#26960215 : EXPORT/IMPORT IS CRASHING WITH TABLE CREATED IN REDUNDANT ROW FORMAT
      This is duplicating MDEV-13899 that was fixed in MariaDB 10.0.33, 10.1.28, 10.2.10, 10.3.2.

      Bug #26334149 - MYSQL CRASHES WHEN FULL TEXT INDEXES IBD FILES ARE ORPHANED DUE TO RENAME TABLE
      This has been merged from 5.6.40 to 10.0.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                marko Marko Mäkelä
                Reporter:
                marko Marko Mäkelä
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: