- developers - lists.mariadb.org

[Maria-developers] Review request: mdev-4651
by Alexander Barkov 13 Jun '13

13 Jun '13

Hi Serg, Please review a patch for "mdev-4651 Crash in my_decimal2decimal in a ORDER BY query". Thanks.

1 0

[Maria-developers] Review request: mdev-4635
by Alexander Barkov 13 Jun '13

13 Jun '13

Hi Serg, please review a fix for MDEV-4635 Thanks.

1 0

Re: [Maria-developers] Review request: mdev-4634
by Alexander Barkov 13 Jun '13

13 Jun '13

Hi Sergei, Thanks for review! On 06/11/2013 09:33 PM, Sergei Golubchik wrote: > Hi, Alexander! > > This looks like a lot of changes for what you explained the reason of > the bug was. There will be lots of "whys" below... The fix itself was quite small. However, an attempt to write a reasonable test case revealed a few other bugs. I just tried to fixing everything in a single patch. Okey. Let's do everything separately. These are separate reports for the problems found during writing a test case for 4634: https://mariadb.atlassian.net/browse/MDEV-4652 https://mariadb.atlassian.net/browse/MDEV-4653 https://mariadb.atlassian.net/browse/MDEV-4654 Here's a new one, found today: https://mariadb.atlassian.net/browse/MDEV-4651 > > On Jun 11, Alexander Barkov wrote: >> === modified file 'sql/item.cc' >> --- sql/item.cc 2013-03-17 06:41:22 +0000 >> +++ sql/item.cc 2013-06-11 11:52:43 +0000 >> @@ -252,7 +252,8 @@ String *Item::val_string_from_decimal(St >> String *Item::val_string_from_date(String *str) >> { >> MYSQL_TIME ltime; >> - if (get_date(&ltime, TIME_FUZZY_DATE) || >> + uint time_flag= field_type() == MYSQL_TYPE_TIME ? TIME_TIME_ONLY : 0; >> + if (get_date(&ltime, TIME_FUZZY_DATE | time_flag) || > > Why? > >> str->alloc(MAX_DATE_STRING_REP_LENGTH)) >> { >> null_value= 1; >> >> === modified file 'sql/item_func.cc' >> --- sql/item_func.cc 2013-04-06 13:14:46 +0000 >> +++ sql/item_func.cc 2013-06-11 10:54:50 +0000 >> @@ -2456,7 +2456,8 @@ bool Item_func_min_max::get_date(MYSQL_T >> { >> Item **arg= args + i; >> bool is_null; >> - longlong res= get_datetime_value(thd, &arg, 0, compare_as_dates, &is_null); >> + longlong res= get_datetime_value(thd, &arg, 0, compare_as_dates, >> + &is_null, fuzzy_date); > > why is that? to handle CONVERT_TZ(GREATEST(...)) ? > That is to pass NO_ZERO_DATE down? > > I'm not sure it's a good idea. Strictly speaking, only the greatest > value should be NO_ZERO_DATE, not every single one. > > I'd suggest to add check_date in Item_func_min_max::get_date() instead. > >> /* Check if we need to stop (because of error or KILL) and stop the loop */ >> if (thd->is_error() || args[i]->null_value) >> @@ -2469,6 +2470,11 @@ bool Item_func_min_max::get_date(MYSQL_T >> min_max= res; >> } >> unpack_time(min_max, ltime); >> + >> + if (!(fuzzy_date & TIME_TIME_ONLY) && >> + ((null_value= check_date_with_warn(ltime, fuzzy_date)))) >> + return true; > > Yes, that's what I mean. If you check the return value anyway, there's > no need to check it inside get_datetime_value(). > >> if (compare_as_dates->field_type() == MYSQL_TYPE_DATE) >> { >> ltime->time_type= MYSQL_TIMESTAMP_DATE; >> >> === modified file 'sql/item_timefunc.cc' >> --- sql/item_timefunc.cc 2013-03-20 15:13:00 +0000 >> +++ sql/item_timefunc.cc 2013-06-11 10:57:09 +0000 >> @@ -1923,8 +1921,10 @@ void Item_date_add_interval::fix_length_ >> bool Item_date_add_interval::get_date(MYSQL_TIME *ltime, uint fuzzy_date) >> { >> INTERVAL interval; >> - >> - if (args[0]->get_date(ltime, TIME_NO_ZERO_DATE | TIME_FUZZY_DATE | TIME_NO_ZERO_IN_DATE) || >> + if (args[0]->get_date(ltime, >> + TIME_NO_ZERO_DATE | TIME_FUZZY_DATE | TIME_NO_ZERO_IN_DATE | >> + ((args[0]->field_type() == MYSQL_TYPE_TIME) ? >> + TIME_TIME_ONLY : 0)) || > > why is that? > >> get_interval_value(args[1], int_type, &value, &interval)) >> return (null_value=1); >> >> @@ -2290,6 +2290,12 @@ bool Item_time_typecast::get_date(MYSQL_ >> return 1; >> if (decimals < TIME_SECOND_PART_DIGITS) >> ltime->second_part= sec_part_truncate(ltime->second_part, decimals); >> + if (!(fuzzy_date & TIME_TIME_ONLY)) >> + { >> + if (ltime->time_type == MYSQL_TIMESTAMP_TIME) >> + ltime->time_type= MYSQL_TIMESTAMP_DATETIME; > > why? > >> + return ((null_value= check_date_with_warn(ltime, fuzzy_date))); >> + } >> /* >> MYSQL_TIMESTAMP_TIME value can have non-zero day part, >> which we should not lose. >> === modified file 'sql/item_timefunc.h' >> --- sql/item_timefunc.h 2012-12-28 12:41:46 +0000 >> +++ sql/item_timefunc.h 2013-06-11 11:19:03 +0000 >> @@ -492,6 +492,8 @@ class Item_timefunc :public Item_tempora >> Item_timefunc(Item *a,Item *b) :Item_temporal_func(a,b) {} >> Item_timefunc(Item *a, Item *b, Item *c) :Item_temporal_func(a, b ,c) {} >> enum_field_types field_type() const { return MYSQL_TYPE_TIME; } >> + int save_in_field(Field *field, bool no_conversions) >> + { return save_time_in_field(field); } > > why? > >> }; >> >> === modified file 'sql/time.cc' >> --- sql/time.cc 2013-01-08 20:23:03 +0000 >> +++ sql/time.cc 2013-06-11 10:51:31 +0000 >> @@ -213,6 +213,20 @@ ulong convert_month_to_period(ulong mont >> } >> >> >> +bool >> +check_date_with_warn(const MYSQL_TIME *ltime, uint fuzzy_date) >> +{ >> + int dummy_warnings; >> + if (check_date(ltime, fuzzy_date, &dummy_warnings)) >> + { >> + Lazy_string_time str(ltime); >> + make_truncated_value_warning(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN, >> + &str, MYSQL_TIMESTAMP_ERROR, 0); > > please, use this function also in Item_date_typecast::get_date() and > may be in other places. you might need to pass MYSQL_TIMESTAMP_xxx > value as an argument, though. > >> + return true; >> + } >> + return false; >> +} >> + >> /* >> Convert a timestamp string to a MYSQL_TIME value and produce a warning >> if string was truncated during conversion. >> > > Regards, > Sergei >

1 0

[Maria-developers] I_S plugin doubts
by Roberto Spadim 13 Jun '13

13 Jun '13

Hi i was reading locale_info and qc_info source that i sent in jira changing the last parameter from 0 to SKIP_OPEN_TABLE, change the table from aria engine to memory engine, why? this change something internally in I_S table ? example: from static ST_FIELD_INFO qc_info_fields_queries[]= { {"ENTRY_POSITION_IN_CACHE", MY_INT32_NUM_DECIMAL_DIGITS, MYSQL_TYPE_LONG, 0, 0, 0, 0}, ... to static ST_FIELD_INFO qc_info_fields_queries[]= { {"ENTRY_POSITION_IN_CACHE", MY_INT32_NUM_DECIMAL_DIGITS, MYSQL_TYPE_LONG, 0, 0, 0, SKIP_OPEN_TABLE}, ... -- Roberto Spadim

2 5

Re: [Maria-developers] MDEV-4581 , more query cache information fields
by Roberto Spadim 12 Jun '13

12 Jun '13

hi sergei, i sent the patch, could you check it? i think that's easy to read and understand maybe some code standard should be done since i don't know how code in mariadb cs https://mariadb.atlassian.net/browse/MDEV-4581

1 0

[Maria-developers] MDEV-4581 , more query cache information fields
by Roberto Spadim 11 Jun '13

11 Jun '13

Hi guys, i sent a patch for qc_info plugin in MDEV-4581 Could mariadb team review and tell me what i should do, to put it in the next release? i will rewrite it, at 10.0.3 release, and change some things: the add to query to cache, i will change to only a function call at sql_cache.cc, and put the function at sql_cache.h something like: qc->qc_info_set_informations(insert time, rows_sent, rows_read, etc etc etc) rewrite the part of query hit. to execute only a command too qc->qc_info_hit(current_time, expend_qc_time, etc etc etc) this will make sql_cache.cc code cleaner and sql_cache.h "dirty" but will be more easly to read and patch future versions. only add 2 lines to sql_cache.cc , and many lines to sql_cache.h since sql_cache.h have the structure it will be easly to write function call at struct than sql_cache.cc file bye guys -- Roberto Spadim

1 0

Re: [Maria-developers] audit plugin help with MDEV-575 (fail2ban like)
by Roberto Spadim 11 Jun '13

11 Jun '13

well i'm thinking about again, maybe just a syslog could do the job with fail2ban but in windows... well it will not work maybe the second option of a internal table could do the job too Host | bad_login_datetime if host have more than 'bad_login_tries' bad logins in 'bad_login_seconds' seconds before now it will be disconnect after xxx seconds or after each minute, or after each disconnect clean the internal table that have bad_login_datetime < now() - 'bad_login_seconds' the clean should be optimized in a not paintfull way (maybe like a garbage colector that runs after X disconnects, or a event that happens each x seconds, or both of them, i prefer the disconnect method) any other idea?

1 2

Re: [Maria-developers] [Commits] Rev 3551: MDEV-4612 SQ pushdown: Server crashes in make_join_statistics with materialization+semijoin, IN subqueries, constant table, impossible condition in file:///home/tsk/mprog/src/10.0-md83/
by Oleksandr Byelkin 11 Jun '13

11 Jun '13

Hi! 11.06.2013 11:11, timour(a)askmonty.org пишет: > At file:///home/tsk/mprog/src/10.0-md83/ > > ------------------------------------------------------------ > revno: 3551 > revision-id: timour(a)askmonty.org-20130611081135-rge7p00h8e9nhypj > parent: timour(a)askmonty.org-20130528122232-kisslvdw5vhrna0q > fixes bug: https://mariadb.atlassian.net/browse/MDEV-4612 > committer: timour(a)askmonty.org > branch nick: 10.0-md83 > timestamp: Tue 2013-06-11 11:11:35 +0300 > message: > MDEV-4612 SQ pushdown: Server crashes in make_join_statistics with materialization+semijoin, IN subqueries, constant table, impossible condition > > Analysis: > The subquery materialization execution method subselect_hash_sj_engine::exec() > didn't restore thd->lex->current_select through all return paths. This was > not a problem when the method is called during execution. When a subquery > was executed during the optimization phase of the outer query, current_select > remained to point to the wrong subquery. This resulted in update_ref_and_keys() > getting the wrong number of conditions via thd->lex->current_select->cond_count, > thus allocating a smaller buffer for sargable conditions. This later resulted > in memory overwrite. > > Solution: > Make sure that subselect_hash_sj_engine::exec() always restores current_select. The patch looks OK to push. [skip]

1 0

[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