June 2013 - developers - lists.mariadb.org

[Maria-developers] Group commit comparison between mysql and mariadb
by Sergei Petrunia 11 Jun '13

11 Jun '13

Hi Kristian, I'm wondering, do we have any comparison of MariaDB's binlog group commit implementation vs MySQL's binlog's group commit? I recall Mats writing a blog about the differences last year, but he didn't have the code back then. I'm wondering if Oracle had implemented what he described, or did something else... I understand that in order to do benchmark we need to finish merging InnoDB from MySQL 5.6 into MariaDB 10.0. One can't benchmark group commit per se, it is tied to innodb. Do we still have any clue, before the merge happens? Btw, what should the merge do - take group commit from MariaDB or from MySQL? BR Sergei -- Sergei Petrunia, Software Developer Monty Program AB, http://askmonty.org Blog: http://s.petrunia.net/blog

4 3

Re: [Maria-developers] Buildbot question
by Kristian Nielsen 11 Jun '13

11 Jun '13

Sergey Vojtovich <svoj(a)mariadb.org> writes: > I'm observing pretty odd buildbot behavior with some of my pushes (probably > merges). Bar and Daniel were unable to answer my questions and suggested to > ask you instead. The problem seems to be that Buildbot has become confused about revisions. Look at the top of the page you linked, at the revisions for each column. You will notice that they are not in order. The 3rd column is the highest revision, which is probably why spider suite is enabled there (I assume you pushed something to add the suite). I see your current top revision in the tree is 3760, while the Buildbot page has a newer revision 3767. It also looks like the Buildbot history contains _another_ earlier revision that was also number 3760, maybe that is the cause of the confusion (Buildbot perhaps thinks it is rebuilding the old 3760). You can see your revision sorted as the first columns in this link: http://buildbot.askmonty.org/buildbot/grid?branch=10.0-spider&width=20 This confusion usually happens because the tree was overwritten and revisions became renumbered. This can happen due to bzr push --overwrite, but in your case it is probably because you merged in the opposite direction. When merging a tree A with a tree B and pushing to B, it is generally best to merge A into B (cd to B and bzr merge A). This keeps the existing revision numbers of B intact. In the main trees we have protection that enforces this, but not in feature trees. (I would be nice if we could fix Buildbot to prevent this confusion, but so far noone has taken it upon themselves. At one point I asked the Launchpad/bzr developers to make the revision id (as opposed to revision numbers) available so that Buildbot could avoid using revision numbers completely, but the answer was that "hiding revision ids was a feature" ...). Usually the confusion is sorted out by itself on a following push, or you can use the trick above adding width=N to the link to see the column of the revision you want. Hope this helps, - Kristian.

2 1

[Maria-developers] MariaDB 10.0.3 release prep
by Daniel Bartholomew 11 Jun '13

11 Jun '13

All, I've begun the prep work for the MariaDB 10.0.3 alpha release. The packages are uploaded and the mirrors should pick them up soon (if they haven't already). The draft changelog and release notes are here: - https://kb.askmonty.org/en/mariadb-1003-release-notes/ - https://kb.askmonty.org/en/mariadb-1003-changelog/ Please let me know ASAP if there's anything that needs to be mentioned/highlighted/removed/etc... Assuming all goes well, the plan is to activate the release tomorrow (11 Jun 2013) at around 10am US Eastern Time. Thanks. -- Daniel Bartholomew Google+ - http://gplus.to/dbart Twitter - http://twitter.com/daniel_bart MariaDB: An Enhanced Drop-in Replacement for MySQL Website - http://mariadb.org Twitter - http://twitter.com/mariadb Google+ - http://google.com/+mariadb Facebook - http://fb.com/MariaDB.dbms Knowledgebase - http://kb.askmonty.org

1 0

[Maria-developers] audit plugin help with MDEV-575 (fail2ban like)
by Roberto Spadim 09 Jun '13

09 Jun '13

hi guys i want get users / ip that try to connect to server with bad password/user and retry very frequently (normally a brute force attack) some things i need: 1)get information about connection when it fail or not (maybe an audit plugin could do this?), information that i need: 1.1)protocol (tcp/unix/others) 1.2)user 1.3)remote ip (when tcp only) it must be executed in every login 2) log this information maybe we could use CSV,aria, myisam, to log bad logins and put it in a log table: PROTOCOL, USER, REMOTE IP, DATATIME 3) add columns to user BAD_LOGIN_MAX_TRIES, BAD_LOGIN_LOG_TIME, BAD_LOGIN_MIN_RETRY_TIME 4) add server global bad_login_tcp_log_time, bad_login_tcp_max_errors 5) implement checks before connection: when openned a connection to server via TCP execute something like: SELECT COUNT(*) FROM bad_login_log WHERE remote_ip="remote_ip" AND datetime>NOW() - @@bad_login_tcp_log_time if COUNT(*) > @@bad_login_tcp_max_erros disconnect 6) implement user checks: SELECT COUNT(*),(SELECT BAD_LOGIN_MAX_TRIES FROM users WHERE user="user") FROM bad_login_log WHERE user="user" AND datetime>NOW() - (SELECT BAD_LOGIN_LOG_TIME FROM users WHERE user="user") if count(*) > bad_login_max_tries form that user disconnect that's all -- Roberto Spadim SPAEmpresarial

1 0

[Maria-developers] Help with MDEV-4419, ENUM optimization
by Roberto Spadim 09 Jun '13

09 Jun '13

Hi guys, i want to optimize fields with ENUM() in item comparation the point is: SELECT COUNT(*) FROM table WHERE enum_field!='x' column 'enum_field' is ENUM('x,'y','z') NOT NULL DEFAULT 'x' i want to rewrite the query in a way that we can execute it with index: SELECT COUNT(*)FROM table WHERE enum_field IN ('y','z','') the "" is not allowed values ('a' for example) the feature is 'easy' to understand but i don't know where to implement it will read possible values of enum fields, and rewrite "!=","<","<=",">=",">" operators to IN operator with fields that match the condition of the primary operator some think like: column "operator" const value if(operator in ('!=','<','<=','>=','>' AND column is ENUM){ get all possible values from column copy all values to a return variable execute the "operator" in all values with const values if operator return false remove the value from the return variable change the "operator" and "const value" to : "IN" and return variable comma separated }else{ no optimization leave operation as it is } example: column = enum('x','y','z'), operator = ">" const value = "3" if(column is ENUM and operator = ">"){ possible values => "x","y","z","" (check that "" is used when insert into, value = "a" for example) return variable = "x","y","z", "" for each value.... "x" > "3" ? yes, ASCII 120 > ASCII 51 -> do nothing "y" > "3" ? yes, ASCII 121 > ASCII 51 -> do nothing "z" > "3" ? yes, ASCII 122 > ASCII 51 -> do nothing "" > "3" ? no, "" < ASCII 51 -> remove the value "" from return variable end for now the return variable is : "x","y","z" the "" was removed rewrite the operator to: "column IN ('x','y','z')" } ----------------------- this will optimize a lot: using != in a big table (446577 rows) result = Using where, 446577 rows, type= ALL using IN() in a big table: result = Using where, 28 rows, key = index, type = range in other words, a lot of optimization with ENUM fields, since they are a pseudo "index" of what we can found in the column =) thank guys any help is wellcome to make a patch -- Roberto Spadim SPAEmpresarial

1 0

[Maria-developers] GSoC MariaDB Kerberos Authentication Plugin Proposal
by QIU Shuang 06 Jun '13

06 Jun '13

Hi Sergei, Thank you and the community for accepting my proposal of this year's GSoC: Kerberize MariaDB -- add kerberos authentication support to MariaDB. I'm Shuang Qiu, a second year graduate student from Shanghai Jiao Tong University, Shanghai, China. My field of research is distributed computing, especially distributed storage. I just get in touch with my project mentor Mr. Vaintroub and join the #maria IRC channel at freenode.net, where I can be found as shuang. Thank you again and hope we can have a happy and productive summer! Cheers! Sincerely, Shuang

1 0

Re: [Maria-developers] [Maria-discuss] Clear-text plugin
by Peter Laursen 06 Jun '13

06 Jun '13

"no support for dlopen()" . How does this apply to Windows clients? We always build first on Windows and once we have no errors of any kind we build on Linux if the application exists for that platform (what MONyog does and SQLyog does not) and fix what issues may occur there. This issue has occurred with SQLyog (and as such for a Windows client). I am not coder/developer myself, but on Windows I think we have no other option than statically linking? We could of could compile the API as a .dll, but we don't (because we had a lot of problems with it earlier). -- Peter On Thu, Jun 6, 2013 at 3:25 PM, Sergei Golubchik <serg(a)mariadb.org> wrote: > Hi, Peter! > > On Jun 04, Peter Laursen wrote: > > We have the same problem. We cannot test either. But we have a few user > > reports. We will discuss what we can do. > > > > Is there any parameter or compile-time option we should be aware of? Any > > specific documentation somewhere? Anything? > > Not really. If the cleartext plugin is built and installed, it'll be > used when the server requests it. > > I've just checked that this plugin is present both in rpm and deb > packages. I didn't check binary tarballs, but it should be there too. > > So, there's nothing to configure at run-time or compile-time, everything > should work automatically and transparently. > > Hmm, unless you build a completely static client library with no support > for dlopen(). > > Regards, > Sergei >

2 1

Re: [Maria-developers] [Maria-discuss] Clear-text plugin
by Peter Laursen 04 Jun '13

04 Jun '13

We have the same problem. We cannot test either. But we have a few user reports. We will discuss what we can do. Is there any parameter or compile-time option we should be aware of? Any specific documentation somewhere? Anything? -- Peter On Tue, Jun 4, 2013 at 2:19 PM, Sergei Golubchik <serg(a)mariadb.org> wrote: > Hi, Peter! > > On Jun 04, Peter Laursen wrote: > > For clarity: a 'dummy' parameter doing nothing does not solve our > > problem. We actually want to build clients that can (also) connect to > > Oracle MySQL Enterprise servers configured to make use of Oracle's PAM > > implementation. > > We build and distribute a cleartext plugin. And it can be used if the > server requests. In a sense, in MariaDB a cleartext plugin is always > enabled. So a dummy option is needed only to restore the binary protocol > compatibility. That's my understanding, at least, but I don't have > Oracle's PAM plugin, so I cannot test it. > > But if you have verified that your client with our C connector can not > connect to Oracle MySQL Enterprise server with the Oracle's PAM > implementation - please submit a bug report, and we will fix it. > > Regards, > Sergei > >

1 0

Re: [Maria-developers] [Maria-discuss] Clear-text plugin
by Peter Laursen 04 Jun '13

04 Jun '13

For clarity: a 'dummy' parameter doing nothing does not solve our problem. We actually want to build clients that can (also) connect to Oracle MySQL Enterprise servers configured to make use of Oracle's PAM implementation. -- Peter On Tue, Jun 4, 2013 at 11:05 AM, Peter Laursen <peter_laursen(a)webyog.com>wrote: > yes .. in the Connector/C. We use C/C++ for both SQLyog and MONyog. > > -- Peter > > > On Tue, Jun 4, 2013 at 10:17 AM, Sergei Golubchik <serg(a)mariadb.org>wrote: > >> Hi, Peter! >> >> On Jun 04, Peter Laursen wrote: >> > This is a continuation of >> > https://mariadb.atlassian.net/browse/CONC-25. We requested support for >> > the clear-text plugin (or a workaround of any kind) that would allow a >> > client compiled with the MariaDB connector to use PAM authentication >> > with MySQL Enterprise servers from Oracle. >> >> Another related issue is https://mariadb.atlassian.net/browse/MDEV-4574 >> which should be fixed in 5.5.32, next 5.5 release. >> >> But you, probably, want it in Connector/C, don't you? >> >> Regards, >> Sergei >> >> >

1 0

[Maria-developers] Clear-text plugin
by Peter Laursen 04 Jun '13

04 Jun '13

This is a continuation of https://mariadb.atlassian.net/browse/CONC-25. We requested support for the clear-text plugin (or a workaround of any kind) that would allow a client compiled with the MariaDB connector to use PAM authentication with MySQL Enterprise servers from Oracle. I will not dispute Sergei's opinions about Oracle's implementation. But I believe it misses the point. The question is if the MariaDB connector should be a replacement for the C-IPA from Oracle. And the question is nothing else than that. Not the quality of solutions bundled with Enterprise servers from Oracle. I would like to hear opinions not only from 'core developers' like Sergei, but also from wider circles in MariaDB organization (management, marketing) as well as the user community. For us the problem is like this: We introduced support for the clear-text plugin in SQLyog 11.1. In 11.12 beta2 we replaced Oracles connector with MariaDB (and had to removed it again). We had a few crashes and memory leaks with mariaDB Connector but that is getting in control now (ther were bugs in both the API and in oru code). We probably could patch the MariaDB code without much problems to support the full-text plugin - but we don't have a MySQL Enterprise server to test with and we cannot release without proper testing. We have already started getting complaints from users/customers about that what we introduced in 11.1 has been removed in current development tree. It seems that we have quite a lot of customers using Oracle's Enterprise servers wiht Oracle's PAM authentication. A solution/workaround would very much be appreciated. Regards Peter Laursen Webyog

1 0

Re: [Maria-developers] MDEV-4568 - Port Percona response time distribution as audit plugin
by Sergey Vojtovich 03 Jun '13

03 Jun '13

Hi Sergei, 03.06.2013, в 16:36, Sergei Golubchik <serg(a)mariadb.org> написал(а): > Hi, Sergey! > > On May 31, Sergey Vojtovich wrote: >> revno: 3753 >> revision-id: svoj(a)mariadb.org-20130531131833-zs83x5wntvl1i15f >> parent: wlad(a)montyprogram.com-20130523125248-22m56m01r24t0dmb >> committer: Sergey Vojtovich <svoj(a)mariadb.org> >> branch nick: maria-qrtp >> timestamp: Fri 2013-05-31 17:18:33 +0400 >> message: >> MDEV-4568 - Port Percona response time distribution as audit plugin >> >> sql/sp_head.cc: >> Trigger MYSQL_AUDIT_GENERAL_STATUS event for each statement in stored >> procedures. Not strictly needed for query_response_time plugin, but >> makes it behave more like Percona patch. > > I'm sorry, I couldn't quite understand what you have changed, as > compared to the first patch (and the first was not in emails, so I > couldn't compare them). The difference is rather minor: - renamed query_response_time_exec_time to query_response_time_exec_time_debug - updated plugin author and added GPL header to plugin.cc - fixed I_S plugin description What I didn't touch were query_response_time.cc and .h, but below you agree that we shouldn't touch them. > Looks pretty much the same as the previous one, although you need to run > the tests and to see that they pass (you've changed plugin description > in the code, but the test result wasn't changed accordingly). Hmm… you're right, somehow I forgot to update basic test. Other tests are up to date. > Otherwise, looks ok to push, I suppose we can let it be as close to > Percona's version as possible. I don't think you need to rewrite it, > it's in a plugin now anyway. > > Regards, > Sergei Thanks, Sergey

1 0

[Maria-developers] Partition prune and lock
by Roberto Spadim 02 Jun '13

02 Jun '13

Hi guys i'm studing mariadb code and reading user manuals to know more about features. My last study was about query cache with many ideas in some MDEV and some patchs sent Now i'm reading about partition My questions are 1)when a update or a delete is done on a expecific partition like: Update table set field=2 where partition=1 How lock is done? Lock is done at table level or partition level? 2)Today we have vertical partition, could we implement 'virtual' horizontal partition update and delete in some group of fields (horizontal partition) only lock the horizontal partition being changed and used in where clausule? Any idea and problems on how to develop it? 3) can query cache be partition pruned? instead of removing entire table, just remove the partition that was changed It will consume more memory but will do a better work in cache Maybe a new key like SQL_CACHE_PARTITON or a new variable, query_cache_min_unit=PARTITION or TABLE could change how the query will be cached and pruned Thanks guys

1 1

Re: [Maria-developers] [codership-team] Hanging Galera Cluster, looking for the cause
by Jan Kirchhoff 01 Jun '13

01 Jun '13

I have seen exactly the same issue again, second day in a row. This time at 7:30 in the morning and again, no logs or anything that would give me an idea of what could have caused this. "killall -9 mysqld" and "service mysql start" on the hanging node is the only solution I have so far... I am crossposting this to maria-developers, maybe somebody on that list has an idea where to start? Jan Am 31.05.2013 10:29, schrieb Jan Kirchhoff: > Hi, > > we have been testing MariaDB 5.5.29 with Galera > (wsrep_provider_version 23.2.4(r147)) for a few months now and have had > a freeze of the cluster for the second time now. I didn't report the > first hang about 6 weeks ago as I wanted to be sure it wasn't our fault, > but I can't find any type of network issue or such in our monitoring > system and nothing in the logs that could be related to this problem: > > We have 2 database servers running Ubuntu 12.10 with Kernel 3.5.0 > (names: db8, db9), with a bunch of 15k-disks, some SSDs, 128GB RAM and a > test database of about 600-700GB. The amount of binlogs written every > day is around 500GB, so you get an idea of the workload. > db8 is a slave (traditional replication) of our main database. We have a > garbd running on another server as a third galera instance. > > Everything runs smooth for some weeks, but since this morning at 3:50am > db9 seems to be frozen. The server itself looks good, mysqld is taking > about 50% of one CPU core, everything else looks normal. Nothing in the > logs. Connecting to mysql works, but any query (like "show status" or > normal selects) hang forever. Only "show engine innodb status" returns. > There are almost 3000 processes that piled up like this: > > ---TRANSACTION 2F89D7CB, not started > MySQL thread id 113575, OS thread handle 0x7fab31821700, query id > 45249946 cl4.logical-line.de 192.168.0.16 stats_pflege sleeping > > but none of them ever gets any further. I can issue a "kill 113575" and > the prompt returns, but nothing happens. > > This is what db8's global status says: > wsrep_local_state_uuid 41d7259d-7200-11e2-0800-943dd0e2f659 > wsrep_protocol_version 4 > wsrep_last_committed 179518511 > wsrep_replicated 53590833 > wsrep_replicated_bytes 14768686288759 > wsrep_received 894995 > wsrep_received_bytes 7161712 > wsrep_local_commits 53583808 > wsrep_local_cert_failures 60 > wsrep_local_bf_aborts 0 > wsrep_local_replays 0 > wsrep_local_send_queue 8 > wsrep_local_send_queue_avg 0.000000 > wsrep_local_recv_queue 0 > wsrep_local_recv_queue_avg 0.000000 > wsrep_flow_control_paused 1.000000 > wsrep_flow_control_sent 0 > wsrep_flow_control_recv 0 > wsrep_cert_deps_distance 73.229730 > wsrep_apply_oooe 0.000000 > wsrep_apply_oool 0.000000 > wsrep_apply_window 0.000000 > wsrep_commit_oooe 0.000000 > wsrep_commit_oool 0.000000 > wsrep_commit_window 0.000000 > wsrep_local_state 4 > wsrep_local_state_comment Synced > wsrep_cert_index_size 25606 > wsrep_causal_reads 0 > wsrep_incoming_addresses 192.168.10.30:3306,192.168.10.31:3306, > wsrep_cluster_conf_id 15 > wsrep_cluster_size 3 > wsrep_cluster_state_uuid 41d7259d-7200-11e2-0800-943dd0e2f659 > wsrep_cluster_status Primary > wsrep_connected ON > wsrep_local_index 0 > wsrep_provider_name Galera > wsrep_provider_vendor Codership Oy <info(a)codership.com> > wsrep_provider_version 23.2.4(r147) > wsrep_ready ON > > > > > Here is the innodb status from db9 (the one that hangs). This is the > only output I could get out of db9 at all: > > Type Name Status > InnoDB > ===================================== > 130531 10:02:08 INNODB MONITOR OUTPUT > ===================================== > Per second averages calculated from the last 18 seconds > ----------------- > BACKGROUND THREAD > ----------------- > srv_master_thread loops: 907033 1_second, 906789 sleeps, 89831 > 10_second, 22491 background, 22489 flush > srv_master_thread log flush and writes: 478222 > ---------- > SEMAPHORES > ---------- > OS WAIT ARRAY INFO: reservation count 17017782, signal count 287266881 > Mutex spin waits 98936299, rounds 91970366, OS waits 868715 > RW-shared spins 139527705, rounds 455067309, OS waits 6488041 > RW-excl spins 58122849, rounds 842712424, OS waits 8565312 > Spin rounds per wait: 0.93 mutex, 3.26 RW-shared, 14.50 RW-excl > -------- > FILE I/O > -------- > I/O thread 0 state: waiting for completed aio requests (insert buffer > thread) > I/O thread 1 state: waiting for completed aio requests (log thread) > I/O thread 2 state: waiting for completed aio requests (read thread) > I/O thread 3 state: waiting for completed aio requests (read thread) > I/O thread 4 state: waiting for completed aio requests (read thread) > I/O thread 5 state: waiting for completed aio requests (read thread) > I/O thread 6 state: waiting for completed aio requests (write thread) > I/O thread 7 state: waiting for completed aio requests (write thread) > I/O thread 8 state: waiting for completed aio requests (write thread) > I/O thread 9 state: waiting for completed aio requests (write thread) > Pending normal aio reads: 0 [0, 0, 0, 0] , aio writes: 0 [0, 0, 0, 0] , > ibuf aio reads: 0, log i/o's: 0, sync i/o's: 0 > Pending flushes (fsync) log: 0; buffer pool: 0 > 49405970 OS file reads, 277420410 OS file writes, 6653801 OS fsyncs > 1 pending preads, 0 pending pwrites > 49.44 reads/s, 16384 avg bytes/read, 0.61 writes/s, 0.61 fsyncs/s > ------------------------------------- > INSERT BUFFER AND ADAPTIVE HASH INDEX > ------------------------------------- > Ibuf: size 1, free list len 698, seg size 700, 300226 merges > merged operations: > insert 410879, delete mark 8032461, delete 82075 > discarded operations: > insert 0, delete mark 0, delete 0 > Hash table size 212497717, node heap has 450852 buffer(s) > 16846.79 hash searches/s, 77045.28 non-hash searches/s > --- > LOG > --- > Log sequence number 18522888544071 > Log flushed up to 18522888544071 > Last checkpoint at 18522888544051 > Max checkpoint age 2035833201 > Checkpoint age target 1972213414 > Modified age 20 > Checkpoint age 20 > 0 pending log writes, 0 pending chkp writes > 6895012 log i/o's done, 0.39 log i/o's/second > ---------------------- > BUFFER POOL AND MEMORY > ---------------------- > Total memory allocated 110310195200; in additional pool allocated 0 > Internal hash tables (constant factor + variable factor) > Adaptive hash index 9086745120 \t(1699981736 + 7386763384) > Page hash 106249768 (buffer pool 0 only) > Dictionary cache 426512815 \t(424997488 + 1515327) > File system 429392 \t(82672 + 346720) > Lock system 296805680 \t(295674296 + 1131384) > Recovery system 0 \t(0 + 0) > Dictionary memory allocated 1515327 > Buffer pool size 6553599 > Buffer pool size, bytes 107374166016 > Free buffers 0 > Database pages 6102747 > Old database pages 2252752 > Modified db pages 1 > Pending reads 1 > Pending writes: LRU 0, flush list 0, single page 0 > Pages made young 48135625, not young 0 > 64.83 youngs/s, 0.00 non-youngs/s > Pages read 49405737, created 7213276, written 266046626 > 49.44 reads/s, 0.00 creates/s, 0.11 writes/s > Buffer pool hit rate 1000 / 1000, young-making rate 0 / 1000 not 0 / 1000 > Pages read ahead 0.00/s, evicted without access 40.11/s, Random read > ahead 0.00/s > LRU len: 6102747, unzip_LRU len: 0 > I/O sum[2662]:cur[22], unzip sum[0]:cur[0] > -------------- > ROW OPERATIONS > -------------- > 1 queries inside InnoDB, 0 queries in queue > 2 read views open inside InnoDB > ---OLDEST VIEW--- > Normal read view > Read view low limit trx n:o 2F754555 > Read view up limit trx id 2F754555 > Read view low limit trx id 2F754555 > Read view individually stored trx ids: > ----------------- > Main thread process no. 15888, id 140381271201536, state: sleeping > Number of rows inserted 46059494, updated 2501543763, deleted 28801859, > read 17046037029 > 0.00 inserts/s, 0.00 updates/s, 0.00 deletes/s, 48965.50 reads/s > ------------ > TRANSACTIONS > ------------ > Trx id counter 2F89DD34 > Purge done for trx's n:o < 2F754554 undo n:o < 0 > History list length 140 > LIST OF TRANSACTIONS FOR EACH SESSION: > ---TRANSACTION 0, not started > MySQL thread id 113586, OS thread handle 0x7fab314fe700, query id > 45250432 localhost root sleeping > SHOW engine INNODB STATUS > ---TRANSACTION 2F89DC7B, not started > MySQL thread id 113583, OS thread handle 0x7fab315d9700, query id > 45250367 cl4.logical-line.de 192.168.0.16 stats_pflege sleeping > ---TRANSACTION 2F89DB4F, not started > MySQL thread id 113582, OS thread handle 0x7fab31622700, query id > 45250261 cl4.logical-line.de 192.168.0.16 stats_pflege sleeping > ---TRANSACTION 2F89DA23, not started > MySQL thread id 113578, OS thread handle 0x7fab31746700, query id > 45250159 cl4.logical-line.de 192.168.0.16 stats_pflege sleeping > ---TRANSACTION 0, not started > <many many rows like these, around 3000 although max_connections is 300?! > > > As soon as I trigger the restart of mysql on db9 (service mysql > restart), I see normal log entries like this: > May 31 10:16:59 db9 mysqld: 130531 10:16:59 [Note] WSREP: Stop replication > May 31 10:16:59 db9 mysqld: 130531 10:16:59 [Note] WSREP: Closing send > monitor... > May 31 10:16:59 db9 mysqld: 130531 10:16:59 [Note] WSREP: Closed send > monitor. > May 31 10:16:59 db9 mysqld: 130531 10:16:59 [Note] WSREP: gcomm: > terminating thread > > and db8 is accepting inserts again and catches up with replication. > > meanwhile, db9 says: > May 31 10:17:02 db9 mysqld: 130531 10:17:02 [Note] WSREP: killing local > connection: 113529 > May 31 10:17:02 db9 mysqld: 130531 10:17:02 [Note] WSREP: killing local > connection: 113199 > May 31 10:17:02 db9 mysqld: 130531 10:17:02 [Note] WSREP: killing local > connection: 110221 > > and nothing happens any more. No CPU and no IO load coming from the > mysqld processes. > After 5 minutes I killed the mysqld with "killall -9 mysqld" and then > started it again the normal way. > db9 syncs via IST and everything is fine again. But I have no idea what > caused all this. According to a colleague killing mysql on db9 was the > thing that helped 6 weeks ago, too. > > Thanks for any pointers where we should look further. > > Jan Kirchhoff >

1 0