Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-11751

Merge new release of InnoDB MySQL 5.7.18 to 10.2

Details

    • 10.2.6-2, 10.2.6-3, 10.1.23

    Attachments

      Issue Links

        Activity

          This is now running on bb-10.2-marko.
          I omitted a number of commits. Some commits may be considered for inclusion later.

          1. Bug#23264552 – XA: ASSERT `M_STATUS == DA_ERROR' (some of the changed files do dot exist in MariaDB, and we are probably missing some XA tests, such as those for the XA disconnect bug)
          2. Bug#23481444 OPTIMISER CALL ROW_SEARCH_MVCC() AND READ THE INDEX APPLIED BY UNCOMMITTED ROWS is an optimization that requires some changes to the SQL layer, which differs from MariaDB. We could optimize this further, by performing the end-range check also when switching pages, to terminate earlier when the index is filled with records that do not exist in the read view. Also, the number of follow-up commits suggests that this could be risky in a GA release: 1, 2, 3, 4, 5, 6.
          3. Bug #14025581 FILE IO INSTRUMENTATION DISABLED IN PERFORMANCE SCHEMA FOR INNODB. This is a large patch with little benefit. Our I/O code is somewhat different, too. Follow-up patches: 1, 2, 3.
          4. BUG#24686908 INNODB SHOULD HAVE A MECHANISM TO CHECK IF ALL PFS KEYS ARE REGISTERED WITH PFS Again, MariaDB maybe does not care that much about performance_schema, to make it worth the effort to merge this 1900-line patch.
          5. Bug #25297593 BUILD ON NATIVE FEDORA 25 FAILS DUE TO DEPRECATED LZ4_COMPRESS_LIMITEDOUTPUT The patch is only applicable to the Oracle reimplementation of the MariaDB 10.1 page_compression. Maybe we should consider something similar for that?
          6. Bug#24329079 Crash with InnoDB Encryption, 5.7.13, FusionIO, innodb_flush_method=O_DIRECT This only applies to the Oracle reimplementation of MariaDB 10.1 encryption. Our I/O subsystem is very different. But maybe we should run some tests on FusionIO and other SSD.
          7. Bug#25551311 BACKPORT BUG #23517560 REMOVE SPACE_ID RESTRICTION FOR UNDO TABLESPACES The purpose of this is to allow dedicated undo tablespaces to be created without reinitializing the whole InnoDB instance. We do not test innodb_undo_tablespaces much, and I am a bit worried about possible regressions. I think that the proper solution would be in MDEV-11633 and its subtasks.
          marko Marko Mäkelä added a comment - This is now running on bb-10.2-marko . I omitted a number of commits. Some commits may be considered for inclusion later. Bug#23264552 – XA: ASSERT `M_STATUS == DA_ERROR' (some of the changed files do dot exist in MariaDB, and we are probably missing some XA tests, such as those for the XA disconnect bug ) Bug#23481444 OPTIMISER CALL ROW_SEARCH_MVCC() AND READ THE INDEX APPLIED BY UNCOMMITTED ROWS is an optimization that requires some changes to the SQL layer, which differs from MariaDB. We could optimize this further, by performing the end-range check also when switching pages, to terminate earlier when the index is filled with records that do not exist in the read view. Also, the number of follow-up commits suggests that this could be risky in a GA release: 1 , 2 , 3 , 4 , 5 , 6 . Bug #14025581 FILE IO INSTRUMENTATION DISABLED IN PERFORMANCE SCHEMA FOR INNODB . This is a large patch with little benefit. Our I/O code is somewhat different, too. Follow-up patches: 1 , 2 , 3 . BUG#24686908 INNODB SHOULD HAVE A MECHANISM TO CHECK IF ALL PFS KEYS ARE REGISTERED WITH PFS Again, MariaDB maybe does not care that much about performance_schema, to make it worth the effort to merge this 1900-line patch. Bug #25297593 BUILD ON NATIVE FEDORA 25 FAILS DUE TO DEPRECATED LZ4_COMPRESS_LIMITEDOUTPUT The patch is only applicable to the Oracle reimplementation of the MariaDB 10.1 page_compression. Maybe we should consider something similar for that? Bug#24329079 Crash with InnoDB Encryption, 5.7.13, FusionIO, innodb_flush_method=O_DIRECT This only applies to the Oracle reimplementation of MariaDB 10.1 encryption. Our I/O subsystem is very different. But maybe we should run some tests on FusionIO and other SSD. Bug#25551311 BACKPORT BUG #23517560 REMOVE SPACE_ID RESTRICTION FOR UNDO TABLESPACES The purpose of this is to allow dedicated undo tablespaces to be created without reinitializing the whole InnoDB instance. We do not test innodb_undo_tablespaces much, and I am a bit worried about possible regressions. I think that the proper solution would be in MDEV-11633 and its subtasks.
          jplindst Jan Lindström (Inactive) added a comment - - edited

          About 5. in 10.2 we already use lz4_compress_default() on fil0pagecompress.cc if system has lz4 library supporting it. Opened https://jira.mariadb.org/browse/MDEV-12593 for 10.1.

          jplindst Jan Lindström (Inactive) added a comment - - edited About 5. in 10.2 we already use lz4_compress_default() on fil0pagecompress.cc if system has lz4 library supporting it. Opened https://jira.mariadb.org/browse/MDEV-12593 for 10.1.

          The changes 2, 3, 7 were also introduced in MySQL 5.6.36, so it looks like we will be getting them when merging that to 10.0 and then via 10.1 to 10.2.
          Change 1 should definitely be done. Maybe change 4 as well.

          marko Marko Mäkelä added a comment - The changes 2, 3, 7 were also introduced in MySQL 5.6.36, so it looks like we will be getting them when merging that to 10.0 and then via 10.1 to 10.2. Change 1 should definitely be done. Maybe change 4 as well.

          I forgot to bump the InnoDB version from 5.7.14 to 5.7.18 when doing the merge. So, MariaDB 10.2.6 will report the InnoDB version as 5.7.14.

          Change 2 was omitted when merging MySQL 5.6.36 to MariaDB 10.0, because it would require a handler API change.
          Changes 3 and 7 were merged to 10.0, and from there to 10.1 and 10.2. So, they will appear in MariaDB 10.2.7.
          MariaDB 10.2.7 will report InnoDB version as 5.7.18.

          marko Marko Mäkelä added a comment - I forgot to bump the InnoDB version from 5.7.14 to 5.7.18 when doing the merge. So, MariaDB 10.2.6 will report the InnoDB version as 5.7.14. Change 2 was omitted when merging MySQL 5.6.36 to MariaDB 10.0, because it would require a handler API change. Changes 3 and 7 were merged to 10.0, and from there to 10.1 and 10.2. So, they will appear in MariaDB 10.2.7. MariaDB 10.2.7 will report InnoDB version as 5.7.18.

          People

            marko Marko Mäkelä
            jplindst Jan Lindström (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            4 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved:

              Git Integration

                Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.