- developers - lists.mariadb.org

Re: [Maria-developers] MDEV-5792 - Deadlock between SELECTs from METADATA_LOCK_INFO and another
by Sergei Golubchik 01 May '14

01 May '14

Hi, Sergey! On Apr 23, Sergey Vojtovich wrote: > At lp:maria/10.0 > > ------------------------------------------------------------ > revno: 4155 > revision-id: svoj(a)mariadb.org-20140423073406-o20n7jvwfjinvxxr > parent: psergey(a)askmonty.org-20140418100754-v5b5sdqbavbcznh7 > committer: Sergey Vojtovich <svoj(a)mariadb.org> > branch nick: 10.0-mdev5792 > timestamp: Wed 2014-04-23 11:34:06 +0400 > message: > MDEV-5792 - Deadlock between SELECTs from METADATA_LOCK_INFO and another > I_S table > > mdl_iterate() helper function (which is used by the plugin to iterate mdl > locks) acquired mutexes in reverse order. > > Fixed by iterating MDL locks in two stages: > 1. Iterate locks hash under the protection of hash mutex, store all > lock pointers in a thread local array and increment reference counter > for the lock. > 2. Iterate local array without protection of hash mutex, handle destroyed > locks. > > It somewhat echoes hack in MDL_map_partition::move_from_hash_to_lock_mutex. Looks ok to me. Regards, Sergei

1 0

[Maria-developers] Next steps in improving single-threaded performance
by Kristian Nielsen 01 May '14

01 May '14

I have been analysing CPU bottlenecks in single-threaded sysbench read-only load. I found that icache misses is the main bottleneck, and that profile-guided compiler optimisation (PGO) with GCC gives a large speedup, 25% or more. (More details in my blog posts: http://kristiannielsen.livejournal.com/17676.html http://kristiannielsen.livejournal.com/18168.html ) Now I would like to ask for some discussions/help in how to get this implemented in practice. It involves changing the build process for our binaries: First compile with gcc --coverage, then run some profile workload, then recompile with -fprofile-use. I implemented a simple program to generate some profile load: https://github.com/knielsen/gen_profile_load It runs a bunch of simple insert/select/update/delete, with different combinations of storage engine, binlog format, and client API. It is designed to run inside the build tree and handle starting and stopping the server being tested, so it is pretty close to a working setup. These commands work to generate a binary that is faster due to PGO: mkdir bld cd bld cmake -DWITHOUT_PERFSCHEMA_STORAGE_ENGINE=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_C_FLAGS_RELWITHDEBINFO="-Wno-maybe-uninitialized -g -O3 --coverage" -DCMAKE_CXX_FLAGS_RELWITHDEBINFO="-Wno-maybe-uninitialized -g -O3 --coverage" .. make tests/gen_profile_load cmake -DWITHOUT_PERFSCHEMA_STORAGE_ENGINE=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_C_FLAGS_RELWITHDEBINFO="-Wno-maybe-uninitialized -g -O3 -fprofile-use -fprofile-correction" -DCMAKE_CXX_FLAGS_RELWITHDEBINFO="-Wno-maybe-uninitialized -g -O3 -fprofile-use -fprofile-correction" make So all the pieces really are there, it should be possible to implement it. But we need to find a good way to integrate it into our build system. The best would be to integrate it into our cmake files. The gen_profile_load.c could go into tests/, ideally we would build both a static and dynamically linked version (so we get PGO for both libmysqlclient.a and libmysqlclient.so). Anyone can help me get cmake to do that? And it would be cool if we could get the above procedure to work completely within cmake, so that the user could just do: cmake -DWITH_PGO ... ; make and cmake would itself handle first building with --coverage, then running gen_profile_load.static and gen_profile_load.dynamic, then rebuilding with -fprofile-use. Anyone know if this is possible with cmake, and if so could help implement it? But alternatively, we could integrate a double build, like the commands above, into the buildbot scripts (.deb, .rpm, bintar). Any comments? Here are some more points: - I tested that gen_profile_load gives a good speedup of sysbench read-only (around 30%, so still very significant even though it generates a different and more varied load). - As another test, I removed all SELECT from gen_profile_load, and ran the resulting PGO binary with sysbench read-only. This still gave a fair speedup, despite the PGO load being completely different from the benchmark load. This gives me confidence that the PGO should not cause performance regressions in cases not covered well by gen_profile_load - More tests would be nice, of course. Axel, would you be able to build some binaries following above procedure, and test some different random benchmarks? Anything that is easy to run could be interesting, both to test for improvement, and to check against regressions. - We probably need a recent GCC version to get good results. I used GCC version 4.7.2. Maybe we should install this GCC version in all the VMs we use to build binaries? - Should we do this in 5.5? I think we might want to. The speedup is quite significant, and it seems very safe - no code modifications are involved, only different compiler options. Any thoughts? Volunteeres for helping with the cmake or buildbot parts? - Kristian.

4 14

[Maria-developers] valgrind failure in main.statistics
by Sergey Petrunia 29 Apr '14

29 Apr '14

Hello Igor, Valgrind build in buildbot currently fails in main.statistics test: http://buildbot.askmonty.org/buildbot/builders/work-amd64-valgrind/builds/5… I have tracked it down: the failure appeared after this push by you: revno: 4157 [merge] committer: Igor Babaev <igor(a)askmonty.org> branch nick: maria-10.0 timestamp: Tue 2014-04-22 14:43:13 -0700 message: Merge. ------------------------------------------------------------ revno: 4152.1.1 committer: Igor Babaev <igor(a)askmonty.org> branch nick: maria-10.0-bugs timestamp: Tue 2014-04-22 14:39:57 -0700 message: Fixed the problem of mdev-5947. Back-ported from the mysql 5.6 code line the patch with the following comment: Fix for Bug#11757108 CHANGE IN EXECUTION PLAN FOR COUNT_DISTINCT_GROUP_ON_KEY CAUSES PEFORMANCE REGRESSION I suppose it would take you much less time than anybody else to fix this? The failure is not repeatable on every machine: it doesn't repeat on my machines, but it can be repeated on work.askmonty.org. BR Sergei -- Sergei Petrunia, Software Developer MariaDB | Skype: sergefp | Blog: http://s.petrunia.net/blog

1 0

Re: [Maria-developers] 10.1 planning
by Kristian Nielsen 29 Apr '14

29 Apr '14

Sergei Golubchik <serg(a)mariadb.org> writes: > See below. > What do you think about it? > > On Apr 23, Elena Stepanova wrote: >> >> There is also https://mariadb.atlassian.net/browse/MDEV-4487 >> (Replication from 5.6 with enabled GTID) to be considered. It's >> currently Minor and has 10.0.x as a target version, both of which are >> questionable: recently Colin said on IRC that it should be much more >> than a minor FR, and yet I doubt that somebody will get to it in 10.0. >> Maybe it's more realistic to raise the priority but to re-target it for >> 10.1? I think Monty had some ideas for doing this, IIRC it was discussed at the meeting in Barcelona in January. If what I remember is true, it is actually just that a 5.6 master refuses to replicate if the slave does not say it is configured for MySQL 5.6 GTID. The slave does not need to actually _use_ GTID. So I think Monty's plan was for the MariaDB slave to just lie to the master and say that it's a 5.6 slave with appropriate config. And the MariaDB slave would need to recognise and ignore the new MySQL 5.6 events, such as GTID. - Kristian.

2 2

[Maria-developers] Old autotools cruft in source tree?
by Otto Kekäläinen 28 Apr '14

28 Apr '14

Hello The Debian Lintian QA scanner yields this warning: W outdated-autotools-helper-file (http://lintian.debian.org/tags/outdated-autotools-helper-file.html) extra/jemalloc/config.guess 2012-02-10 extra/jemalloc/config.sub 2012-02-10 storage/tokudb/ft-index/third_party/xz-4.999.9beta/build-aux/config.guess 2009-04-27 storage/tokudb/ft-index/third_party/xz-4.999.9beta/build-aux/config.sub 2009-04-17 Are these still needed in MariaDB or is this simply old cruft from old autotools time that could be removed? -- Check out our blog at http://seravo.fi/blog and follow @ottokekalainen

2 1

[Maria-developers] "SELECT ... LIMIT 0" is now deterministic
by Alexander Barkov 28 Apr '14

28 Apr '14

Hello Ian, A fix for MDEV-6170 fixed another problem as a side effect: "SELECT ... LIMIT 0" is now deterministic for statement based binary log. Queries with "LIMIT 0" do not return ER_BINLOG_UNSAFE_STATEMENT any more. Can you please mention this in the manual? Some discussion is here: https://mariadb.atlassian.net/browse/MDEV-6170 Related articles in the MySQL manual: https://dev.mysql.com/doc/refman/5.6/en/replication-rbr-safe-unsafe.html https://dev.mysql.com/doc/refman/5.6/en/replication-features-limit.html Thanks.

1 0

[Maria-developers] a Query_log_event charset bug in parallel replication
by nanyi607rao 25 Apr '14

25 Apr '14

hi, If character_set in different Query_log_events changed, worker threads may apply them with wrong character_set. the codes leading this problem is in Query_log_event::do_apply_event, that is: if (charset_inited) { if (rli->cached_charset_compare(charset)) { /* Verify that we support the charsets found in the event. */ if (!(thd->variables.character_set_client= get_charset(uint2korr(charset), MYF(MY_WME))) || !(thd->variables.collation_connection= get_charset(uint2korr(charset+2), MYF(MY_WME))) || !(thd->variables.collation_server= get_charset(uint2korr(charset+4), MYF(MY_WME)))) { There is a charset[6] in rli, which cached last Query_log_event's charset in serial replication. But in parallel replication it would lead mistake, because every worker thread can read and set rli->charset[6], so rli->charset[6] isn't any worker threads' last Query_log_event charset. but rli->charset[6] can affect every worker threads' thd->variables.character_set_* setting. 2014-04-17 nanyi607rao

2 7

[Maria-developers] About the query performance issue reported on IRC (Percona 5.5 vs MariaDB 10.0)
by Elena Stepanova 24 Apr '14

24 Apr '14

Hi all, Over the last couple of days, some of you have seen or even participated in a "thread" on #maria IRC channel regarding the performance problem on query execution that james_woods observed on MariaDB 10.0 instance comparing to a Percona 5.5 instance. This email is for those who are curious about the outcome/current status. Summary ------- The original complaint sounded like this: "After switching to MariaDB 10 I see 5-10 very long running queries - which are not using indices as they do on Percona 5.5." The data in both instances is exactly the same. The HW is essentially different, but it's better on the machine with MariaDB 10. After some back-and-forth, the complaint got split into three main parts: - the query is considerably slower on MariaDB 10.0 instance than on Percona 5.5 instance; - the query plans show very different row counts, although the data is identical; - Percona uses indices when MariaDB doesn't. Sergei Petrunia investigated the alleged issues with query plans and found that there was no problem: - the row counts that Percona 5.5 reported for derived tables were smaller because EXPLAIN in Percona 5.5 had actually executed the query. - the difference in index usage was an effect of table elimination. A table was removed, hence no index use for it was shown in the explain. He also broke down the query and got a simple flat one, without subqueries, which still revealed ~4x difference in execution time. Upon his request, I ran some tests locally to see if I could reproduce such a difference between Percona 5.5 and MariaDB 10.0 and found that I couldn't (of course I don't have the HW that james_woods uses, the but the difference this big was likely to be seen anywhere). Meanwhile, james_woods installed Percona 5.5 on the machine where MariaDB 10.0 was running, and didn't get any performance gain -- that is, the difference was not between Percona 5.5 and MariaDB 10.0, but was caused by something else (HW, environment, ...). As of now, we don't know what it is. (end of summary). Sergei asked me to share the numbers I got while running the local tests. I will also put here the initial queries and explains, for the reference. It's mainly for the record, there is nothing particularly interesting in here; if you stop reading now, you won't lose much. Queries and plans ----------------- The initial offending query: SELECT SUM(visits) FROM ( SELECT IFNULL(ref.probename_unique_visits,'tr_trafficother2_uv') AS probename, SUM(x.value) AS visits, IFNULL(ref.category,'Other') AS category FROM ( SELECT aggregated_data._getSource(source, source2, source3, fb_source, request_ids) AS source, COUNT(DISTINCT u.fbuserid) AS value FROM g9_tracker.app_logins u WHERE u.created_date = '2014-04-22' GROUP BY 1 ) x LEFT JOIN aggregated_data.ref_parameter_mappings ref ON (ref.value_of_ref_parameter = x.source) GROUP BY 1,3 ) y; EXPLAINs (that's where row counts were off; note EXPLAIN execution time): MariaDB 10.0: +------+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +------+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ | 1 | PRIMARY | <derived2> | ALL | NULL | NULL | NULL | NULL | 3394453 | | | 2 | DERIVED | <derived3> | ALL | NULL | NULL | NULL | NULL | 3394453 | Using temporary; Using filesort | | 2 | DERIVED | ref | ref | value_of_ref_parameter_UNIQUE | value_of_ref_parameter_UNIQUE | 603 | x.source | 1 | Using where | | 3 | DERIVED | u | ALL | NULL | NULL | NULL | NULL | 3394453 | Using where; Using filesort | +------+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ 4 rows in set (0,03 sec) Percona 5.5: +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ | 1 | PRIMARY | <derived2> | ALL | NULL | NULL | NULL | NULL | 15 | | | 2 | DERIVED | <derived3> | ALL | NULL | NULL | NULL | NULL | 75 | Using temporary; Using filesort | | 2 | DERIVED | ref | ref | value_of_ref_parameter_UNIQUE | value_of_ref_parameter_UNIQUE | 603 | x.source | 1 | | | 3 | DERIVED | u | ALL | NULL | NULL | NULL | NULL | 3669609 | Using filesort | +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+---------------------------------+ 4 rows in set (1 min 40,20 sec) Example of the indices difference (caused by table elimination): Query: SELECT COUNT(*) FROM ( SELECT aggregated_data._getSource(source, source2, source3, fb_source, request_ids) AS source, COUNT(DISTINCT u.fbuserid) AS value FROM g9_tracker.app_logins u WHERE u.created_date = '2014-04-22' GROUP BY 1 ) x LEFT JOIN aggregated_data.ref_parameter_mappings ref ON (ref.value_of_ref_parameter = x.source); EXPLAINs: MariaDB 10.0: +------+-------------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +------+-------------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+ | 1 | PRIMARY | <derived2> | ALL | NULL | NULL | NULL | NULL | 3394453 | 100.00 | | | 2 | DERIVED | u | ALL | NULL | NULL | NULL | NULL | 3394453 | 100.00 | Using where; Using filesort | +------+-------------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+ 2 rows in set, 1 warning (0,02 sec) Percona 5.5: +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+----------+----------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+----------+----------------+ | 1 | PRIMARY | <derived2> | ALL | NULL | NULL | NULL | NULL | 75 | 100.00 | | | 1 | PRIMARY | ref | ref | value_of_ref_parameter_UNIQUE | value_of_ref_parameter_UNIQUE | 603 | x.source | 1 | 100.00 | Using index | | 2 | DERIVED | u | ALL | NULL | NULL | NULL | NULL | 3669609 | 100.00 | Using filesort | +----+-------------+------------+------+-------------------------------+-------------------------------+---------+----------+---------+----------+----------------+ 3 rows in set, 1 warning (1 min 43,62 sec) Numbers ------- The query that Sergei suggested to run in local tests: SELECT SUM(varchar_non_indexed_column_with_non_empty_different_values='foo') FROM table_with_3_mio_rows I ran tests on Percona 5.5/5.6, MariaDB 5.5/10.0, MySQL 5.5/5.6. Only the relevant results are here, I'll skip the rest. For every run the caches were cleaned, the server was started, and then the queries were executed, 3 times each. Initially servers were run with the default options (not server default, but packaging default, how they are installed on Wheezy). The difference between 5.5 and 5.6/10.0 was astonishing, but it was explained by Performance Schema. It went like this: 5.5 select sum(c2='foo') from t_innodb; 1 row in set (8.17 sec) 1 row in set (0.99 sec) 1 row in set (0.95 sec) select sum(c2='foo') from t_myisam; 1 row in set (3.04 sec) 1 row in set (0.56 sec) 1 row in set (0.55 sec) 5.6 / 10.0 select sum(c2='foo') from t_innodb; 1 row in set (24.24 sec) 1 row in set (21.65 sec) 1 row in set (23.43 sec) select sum(c2='foo') from t_myisam; 1 row in set (18.83 sec) 1 row in set (17.23 sec) 1 row in set (16.05 sec) Pretty much the same picture for MariaDB, Percona and MySQL. It would explain everything, but it turned out that james_woods already had Performance Schema turned off. (Note: we *really* should consider switching off Performance schema, at least in configs provided upon the new package installation). Then I ran another test, with the PS turned off, and otherwise default options (now really default, without any pre-installed config). There was still a difference between 5.5 and 5.6/10.0, although not that big, and for InnoDB only: Percona 5.5 select sum(c2='foo') from t_innodb; 1 row in set (5.97 sec) 1 row in set (0.79 sec) 1 row in set (0.79 sec) select sum(c2='foo') from t_myisam; 1 row in set (3.68 sec) 1 row in set (0.56 sec) 1 row in set (0.56 sec) Percona 5.6 select sum(c2='foo') from t_innodb; 1 row in set (7.33 sec) 1 row in set (2.32 sec) 1 row in set (1.48 sec) select sum(c2='foo') from t_myisam; 1 row in set (2.90 sec) 1 row in set (0.59 sec) 1 row in set (0.59 sec) MariaDB 5.5 select sum(c2='foo') from t_innodb; 1 row in set (8.17 sec) 1 row in set (0.99 sec) 1 row in set (0.95 sec) select sum(c2='foo') from t_myisam; 1 row in set (3.04 sec) 1 row in set (0.56 sec) 1 row in set (0.55 sec) MariaDB 10.0 select sum(c2='foo') from t_innodb; 1 row in set (7.34 sec) 1 row in set (2.07 sec) 1 row in set (2.07 sec) select sum(c2='foo') from t_myisam; 1 row in set (3.49 sec) 1 row in set (0.58 sec) 1 row in set (0.57 sec) MySQL 5.6 select sum(c2='foo') from t_innodb; 1 row in set (7.88 sec) 1 row in set (1.95 sec) 1 row in set (2.11 sec) select sum(c2='foo') from t_myisam; 1 row in set (4.11 sec) 1 row in set (0.55 sec) 1 row in set (0.57 sec) So, on 5.6/10.0 with default options the query from the "warm" InnoDB table was approximately 2x slower than on 5.5. Then I took non-default options from james_woods and applied them, as much as I could, considering the HW difference, to my modest VM. In fact, only innodb_buffer_pool_size mattered. After doubling it on 5.6/10.0, I got pretty much the same results as on 5.5. Percona 5.5 (default options) select sum(c2='foo') from t1; 1 row in set (5.97 sec) 1 row in set (0.79 sec) 1 row in set (0.79 sec) Percona 5.6 (default options + no PS + increased buffer pool) select sum(c2='foo') from t1; 1 row in set (8.33 sec) 1 row in set (0.83 sec) 1 row in set (0.82 sec) MariaDB 5.5 (default options) select sum(c2='foo') from t1; 1 row in set (8.17 sec) 1 row in set (0.99 sec) 1 row in set (0.95 sec) MariaDB 10.0 (default options + no PS + increased buffer pool) select sum(c2='foo') from t1; 1 row in set (7.85 sec) 1 row in set (0.86 sec) 1 row in set (0.86 sec) There are still some minor differences, but my VM is not suitable for fine-grained performance testing. If somebody wants to investigate it further, it should be done in a proper way; in any case I'm sure it can be tuned and made to perform better on 5.6/10.0, nobody who cares about differences this small actually runs the server with default options. Regards, /E

1 0

[Maria-developers] Difference between 0xHHHH and X'HHHH' hex literals
by Alexander Barkov 24 Apr '14

24 Apr '14

Hello Ian, I tried to explain in https://mariadb.atlassian.net/browse/MDEV-6092 to the complainer why it's not a bug. Can you please document the difference between the two hex notations and then close as "not a bug"? Thanks.

1 0

Re: [Maria-developers] bb01.mariadb.net
by Kristian Nielsen 23 Apr '14

23 Apr '14

Daniel Bartholomew <dbart(a)mariadb.com> writes: > Serg and Knielsen: I've got the box mostly configured I think, but > could be (very) wrong, so I wanted to give you a chance to make sure > everything looks good on it before I go ahead and add it to the It looks like it already mostly works. I noticed an error in a build though: https://buildbot.askmonty.org/buildbot/builders/kvm-tarbake-jaunty-x86/buil… The "kernel" log has this: ----------------------------------------------------------------------- /dev/sda1 contains a file system with errors, check forced. /dev/sda1: Entry 'pack-names' in /home/buildbot/.bzr/repository (3415) has an incorrect filetype (was 1, should be 2). /dev/sda1: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY. (i.e., without -a or -p options) ----------------------------------------------------------------------- One likely reason is a failed rsync step. This is the tarbake build; unlike most other virtual machines, this one is actually modified during a build. So it's quite possible that the rsync would go wrong if a tarbake build happened to run in the meantime. So just re-sync that particular VM, and that might solve the problem. Thanks for setting this up! - Kristian.

2 1