December 2016 - developers - lists.mariadb.org

[Maria-developers] MDEV-11485 Split Item_func_between::val_int() into virtual methods in Type_handler
by Alexander Barkov 05 Dec '16

05 Dec '16

Hello Sanja, Please review a patch for MDEV-11485. Thanks!

1 0

[Maria-developers] Please review MDEV-10717 Assertion `!null_value' failed in virtual bool Item::send(Protocol*, String*)
by Alexander Barkov 05 Dec '16

05 Dec '16

Hello Sergei, Please review a patch for MDEV-10717. Thanks!

2 1

Re: [Maria-developers] Working on spider patches, MDEV-7698
by Michael Widenius 05 Dec '16

05 Dec '16

Hi! Now working on MDEV-7700 Spiral patch 002_mariadb-10.0.15.spider.diff +++ ./002_mariadb-10.1.8-spider/sql/ha_partition.cc 2015-10-14 01:48:53.392665313 +0900 @@ -327,7 +327,9 @@ void ha_partition::init_handler_variable m_file_buffer= NULL; m_name_buffer_ptr= NULL; m_engine_array= NULL; +/* m_connect_string= NULL; +*/ m_file= NULL; m_file_tot_parts= 0; m_reorged_file= NULL; @@ -1516,4 +1518,6 @@ int ha_partition::prepare_new_partition( if ((error= set_up_table_before_create(tbl, part_name, create_info, p_elem))) goto error_create; +/* tbl->s->connect_string = p_elem->connect_string; +*/ I don't think this it will work removing the usage of p_elem->connect_string This is because each partition may have a different connect string. Here is an example from fedarated_partion.test: eval create table t1 (s1 int primary key) engine=federated partition by list (s1) (partition p1 values in (1,3) connection='mysql://root@127.0.0.1:$SLAVE_MYPORT/federated/t1_1', partition p2 values in (2,4) connection='mysql://root@127.0.0.1:$SLAVE_MYPORT/federated/t1_2'); The above works in mariadb 10.2 but not in your spider tree. >From this patch I will, for now, only take the code related to HA_EXTRA_WRITE_CAN_REPLACE HA_EXTRA_WRITE_CANNOT_REPLACE --- ./001_mariadb-10.1.8-partition_cond_push/sql/sql_plugin.cc 2015-10-13 23:49:10.188129839 +0900 +++ ./002_mariadb-10.1.8-spider/sql/sql_plugin.cc 2015-10-14 01:48:54.296665317 +0900 @@ -2757,6 +2757,7 @@ static void update_func_str(THD *thd, st *(char**) tgt= my_strdup(value, MYF(0)); else *(char**) tgt= 0; - my_free(old); + if (old) + my_free(old); As my_free is safe to call with NULL, the above is not needed diff -Narup ./001_mariadb-10.1.8-partition_cond_push/sql/sql_priv.h ./002_mariadb-10.1.8-spider/sql/sql_priv.h --- ./001_mariadb-10.1.8-partition_cond_push/sql/sql_priv.h 2015-10-13 23:49:10.189129839 +0900 +++ ./002_mariadb-10.1.8-spider/sql/sql_priv.h 2015-10-14 01:48:54.642665315 +0900 @@ -27,6 +27,8 @@ #ifndef SQL_PRIV_INCLUDED #define SQL_PRIV_INCLUDED +#define PLUGIN_VAR_CAN_MEMALLOC + #ifndef MYSQL_CLIENT The above is not needed, as all code that is testing this is doing: #if defined(MARIADB_BASE_VERSION) && MYSQL_VERSION_ID >= 100000 Which is always true in MariaDB 10.x Regards, Monty

2 7

[Maria-developers] Working on spider patches, MDEV-7701
by Michael Widenius 05 Dec '16

05 Dec '16

Hi! Next patch, MDEV 7701 +++ ./003_mariadb-10.2.0-vp/sql/ha_partition.cc 2016-05-05 21:25:04.289731712 +0900 @@ -7238,13 +7239,30 @@ int ha_partition::extra(enum ha_extra_fu } /* Category 9) Operations only used by MERGE */ case HA_EXTRA_ADD_CHILDREN_LIST: + DBUG_RETURN(loop_extra(operation)); case HA_EXTRA_ATTACH_CHILDREN: + int result; + handler **file; + if ((result = loop_extra(operation))) + DBUG_RETURN(result); + m_num_locks = 0; + additional_table_flags = + (HA_CAN_BG_SEARCH | HA_CAN_BG_INSERT | HA_CAN_BG_UPDATE); + file = m_file; + do + { + m_num_locks += (*file)->lock_count(); + additional_table_flags &= ~((ulonglong) ((*file)->ha_table_flags() ^ + (HA_CAN_BG_SEARCH | HA_CAN_BG_INSERT | HA_CAN_BG_UPDATE))); The above test could be simple: additional_table_flags &= (ulonglong) ((*file)->ha_table_flags(); Should be the same thing. I did however remove this code, as HA_CAN_BG... are not used anywhere in the current spider code. See comment below. I also changed so that m_num_locks always contains the upper limit number of locks needed. Without this change, the call to lock_count() would have returneed m_total_parts * 'new-counted-locks' which would be been way too much. @@ -9123,6 +9141,19 @@ const COND *ha_partition::cond_push(cons handler **file= m_file; COND *res_cond = NULL; DBUG_ENTER("ha_partition::cond_push"); +#ifdef HANDLER_HAS_TOP_TABLE_FIELDS + if (set_top_table_fields) + { + do + { + if ((*file)->set_top_table_and_fields(top_table, + top_table_field, + top_table_fields)) + DBUG_RETURN(cond); + } while (*(++file)); + file= m_file; + } +#endif I added this without the define, but I merged it to the loop for cond_push. Instead of having defines in the sql code, I will fix that in spider and vp that I will set the proper defines based on the MariaDB version. However, I would like to have a clear comment the purpose of the function set_top_table_and_fields(). struct st_mysql_storage_engine partition_storage_engine= { MYSQL_HANDLERTON_INTERFACE_VERSION }; +++ ./003_mariadb-10.2.0-vp/sql/handler.h 2016-05-05 21:25:04.314731713 +0900 @@ -255,6 +257,11 @@ enum enum_alter_inplace_result { */ #define HA_BINLOG_FLAGS (HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE) +#define HA_CAN_BG_SEARCH (1LL << 45) +#define HA_CAN_BG_INSERT (1LL << 46) +#define HA_CAN_BG_UPDATE (1LL << 47) +#define HA_CAN_MULTISTEP_MERGE (1LL << 48) Do you really need HA_CAN_BG_XXX flags? I checked your spider tree and this is not used anywhere, so I think we should leave them out for now. If we don't need these, we can remove the loop for setting them in HA_EXTRA_ATTACH_CHILDREN too. If we need them, I would like to have a comment for why they are needed and when we will need then add these in a separate patch. @@ -3525,6 +3540,29 @@ public: Pops the top if condition stack, if stack is not empty. */ virtual void cond_pop() { return; }; + virtual int set_top_table_and_fields(TABLE *top_table, + Field **top_table_field, + uint top_table_fields) + { + if (!set_top_table_fields) + { + set_top_table_fields = TRUE; + this->top_table = top_table; + this->top_table_field = top_table_field; + this->top_table_fields = top_table_fields; + } + return 0; + } Why is this int and not void ? Do you have in mind that this may be possible to fail in the future? If it's never going to change, then better to do this void. I will make it void for now and change the relevant code. We can make it int again if needed in the future. diff -Narup ./002_mariadb-10.2.0-spider/sql/sql_base.cc ./003_mariadb-10.2.0-vp/sql/sql_base.cc --- ./002_mariadb-10.2.0-spider/sql/sql_base.cc 2016-04-17 02:47:27.000000000 +0900 +++ ./003_mariadb-10.2.0-vp/sql/sql_base.cc 2016-05-05 21:25:04.319731713 +0900 @@ -1475,6 +1475,10 @@ unique_table(THD *thd, TABLE_LIST *table table= table->find_table_for_update(); - if (table->table && table->table->file->ht->db_type == DB_TYPE_MRG_MYISAM) - { - TABLE_LIST *child; + if (table->table && + ( + table->table->file->ht->db_type == DB_TYPE_MRG_MYISAM || + (table->table->file->ha_table_flags() & HA_CAN_MULTISTEP_MERGE) + ) + ) + continue I added HA_CAN_MULTISTEP_MERGE as a flag to DB_TYPE_MRG_MYISAM, which allwoed me to make the above tests and the following identical ones much simpler. This is now pushed into 10.2-spider Regards, Monty

2 2

Re: [Maria-developers] MDEV-11330 Split Item_func_hybrid_field_type::val_xxx() into methods in Type_handler
by Vicențiu Ciorbaru 05 Dec '16

05 Dec '16

Hi Alexander! I find this version much cleaner! Looks good to me. What I'd really like to see later is if we can write some unit tests for these conversions. Not mtr linked tests but actual unit tests specifically using these Items. I am up for chipping in and writing some myself. One thing I'd like to discuss later in the context of reentrant functions are these side effects such as "return (null_value= true);" when fetching data from an Item. I think they should go from base items and only perhaps happen in "cache" items. We can potentially then turn all these methods to be const methods. Vicențiu On Thu, 1 Dec 2016 at 15:18 Alexander Barkov <abarkov73(a)gmail.com> wrote: > Hello Vicențiu, > > Please review a modified version. > > As agreed, I'm not using the cast approach for the built-in types > and define a set of data type conversion methods > directly Type_handler_hybrid_field_type methods instead. > > Thanks! >

1 0

[Maria-developers] MariaRocks port: why handler::init_with_fields() call is needed + call for comments
by Sergey Petrunia 02 Dec '16

02 Dec '16

Hello, I think me and Serg have tried having this discussion before on several occasions but were not successful due to me not having the details back then. Now I'm trying with the details. == Description of the issue == Look at MariaDB, the code in TABLE_SHARE::init_from_binary_frm_image (or open_binary_frm() in mysql-5.6 codebase) The code goes like this: ... /* Allocate handler */ if (!(handler_file= get_new_handler(share, thd->mem_root, plugin_hton(se_plugin)))) // ^(LOCATION-1) ... /* Fix key->name and key_part->field */ if (key_parts) { longlong ha_option= handler_file->ha_table_flags(); // ^(LOCATION-2) ... for (uint key=0 ; key < keys ; key++,keyinfo++) { ... for (i=0; i < key_parts; key_part++, i++) { ... field= key_part->field= share->field[key_part->fieldnr-1]; // (LOCATION-3) ... } } Note that - a handler object is created at (LOCATION-1). A TABLE_SHARE object exists at that point, but it has table_share->key_info==NULL. This means the storage object has no information about the indexes in the table it is representing. - At (LOCATION-2), the code retrieves ha_table_flags(). A bit below that point it checks for "ha_option & HA_PRIMARY_KEY_IN_READ_INDEX" (there are also other checks, both harmful and harmless, but let's focus on this one for now). In MyRocks, whether HA_PRIMARY_KEY_IN_READ_INDEX flag is set depends on the datatypes of the primary key columns. However, ha_rocksdb object is requested to provide ha_table_flags() before it has a chance of looking at the table DDL and actually knowing the value. == Solution in MyRocks == It's here: https://github.com/facebook/mysql-5.6/commit/c1f1f0b64eaecd132fc6ec21663faf… Basically, there is a new handler::init_with_fields() call which SQL layer calls after information about table fields is filled. == Questions == Any objections to this approach? Alternative solutions that do not require a lot of re-writing? BR Sergei -- Sergei Petrunia, Software Developer MariaDB Corporation | Skype: sergefp | Blog: http://s.petrunia.net/blog

1 0

Re: [Maria-developers] [Commits] cf04c06: MDEV-10435 crash with bad stat tables.
by Sergei Golubchik 01 Dec '16

01 Dec '16

Hi, Alexey! On Nov 02, Alexey Botchkov wrote: > revision-id: cf04c06a58e3d0e700491f8f9167e9323cf1de1d (mariadb-10.1.18-28-gcf04c06) > parent(s): c18054deb2b5cfcf1f13aa71574406f2bafb87c6 > committer: Alexey Botchkov > timestamp: 2016-11-02 13:02:32 +0400 > message: > > MDEV-10435 crash with bad stat tables. > > Functions from sql/statistics.cc don't seem to expect > stat tables to fail or to have inadequate structure. > Table open errors suppressed and some validity checks added. > > diff --git a/sql/sql_statistics.cc b/sql/sql_statistics.cc > index 4020cbc..3f341ac 100644 > --- a/sql/sql_statistics.cc > +++ b/sql/sql_statistics.cc > @@ -129,6 +129,30 @@ inline void init_table_list_for_single_stat_table(TABLE_LIST *tbl, > } > > > +static > +inline int stat_tables_are_inadequate(TABLE_LIST *tables) > +{ > + TABLE_SHARE *cur_s; > + > + /* If the number of tables changes, we should revise this function. */ > + DBUG_ASSERT(STATISTICS_TABLES == 3); > + > + cur_s= tables[TABLE_STAT].table->s; > + if (cur_s->fields < TABLE_STAT_N_FIELDS || cur_s->keys == 0) > + return TRUE; > + > + cur_s= tables[COLUMN_STAT].table->s; > + if (cur_s->fields < COLUMN_STAT_N_FIELDS || cur_s->keys == 0) > + return TRUE; > + > + cur_s= tables[INDEX_STAT].table->s; > + if (cur_s->fields < INDEX_STAT_N_FIELDS || cur_s->keys == 0) > + return TRUE; > + > + return FALSE; > +} I'd suggest to use Table_check_intact interface instead. Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

1 0

[Maria-developers] 10.2 tree, innodb_gis.alter_spatial_index test broken after your push
by Sergey Petrunia 01 Dec '16

01 Dec '16

Hi Jan, Your push https://github.com/MariaDB/server/commit/dc9f919f27fccfeb0de3ab392f33bc5efd… broke the innodb_gis.alter_spatial_index test in 10.2 tree. BR Sergei -- Sergei Petrunia, Software Developer MariaDB Corporation | Skype: sergefp | Blog: http://s.petrunia.net/blog

1 0

Re: [Maria-developers] [Commits] 2c9bb42: MDEV-11432 Change the informational redo log format tag to "MariaDB 10.2.3"
by Jan Lindström 01 Dec '16

01 Dec '16

Hi, ok to push. R: Jan On Thu, Dec 1, 2016 at 8:36 AM, <marko.makela(a)mariadb.com> wrote: > revision-id: 2c9bb42d901fc4f48f4884e4a85af74eae6d0929 > (mariadb-10.2.2-91-g2c9bb42) > parent(s): dc9f919f27fccfeb0de3ab392f33bc5efdfd59a0 > author: Marko Mäkelä > committer: Marko Mäkelä > timestamp: 2016-12-01 08:28:59 +0200 > message: > > MDEV-11432 Change the informational redo log format tag to "MariaDB 10.2.3" > > MariaDB 10.2 incorporates MySQL 5.7. MySQL 5.7.9 (the first GA release > of the series) introduced an informational field to the InnoDB redo log > header, which identifies the server version where the redo log files > were created (initialized, resized or updated), in > WL#8845: InnoDB: Redo log format version identifier. > > The informational message would be displayed to the user, for example > if someone tries to start up MySQL 8.0 after killing a MariaDB 10.2 server. > In the current MariaDB 10.2 source code, the identifier string would > misleadingly say "MySQL 5.7.14" (using the hard-coded version number in > univ.i) instead of "MariaDB 10.2.3" (using the contents of the VERSION > file, the build system copies to config.h and my_config.h). > > This is only a cosmetic change. The compatibility check is based on a > numeric identifier. > > We should probably also change the numeric identifier and some logic > around it. MariaDB 10.2 should refuse to recover from a crashed MySQL 5.7 > instance, because the redo log might contain references to shared > tablespaces, > which are not supported by MariaDB 10.2. Also, when MariaDB 10.2 creates > an encrypted redo log, there should be a redo log format version tag that > will prevent MySQL 5.7 or 8.0 from starting up. > > --- > storage/innobase/include/log0log.h | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/storage/innobase/include/log0log.h > b/storage/innobase/include/log0log.h > index c1f2570..caa067c 100644 > --- a/storage/innobase/include/log0log.h > +++ b/storage/innobase/include/log0log.h > @@ -555,7 +555,11 @@ or the MySQL version that created the redo log file. > */ > /** End of the log file creator field. */ > #define LOG_HEADER_CREATOR_END (LOG_HEADER_CREATOR + 32) > /** Contents of the LOG_HEADER_CREATOR field */ > -#define LOG_HEADER_CREATOR_CURRENT "MySQL " INNODB_VERSION_STR > +#define LOG_HEADER_CREATOR_CURRENT \ > + "MariaDB " \ > + IB_TO_STR(MYSQL_VERSION_MAJOR) "." \ > + IB_TO_STR(MYSQL_VERSION_MINOR) "." \ > + IB_TO_STR(MYSQL_VERSION_PATCH) > > /** The redo log format identifier corresponding to the current format > version. > Stored in LOG_HEADER_FORMAT. */ > _______________________________________________ > commits mailing list > commits(a)mariadb.org > https://lists.askmonty.org/cgi-bin/mailman/listinfo/commits

1 0

Re: [Maria-developers] Please review MDEV-11298 Split Item_func_hex::val_str_ascii() into virtual methods in Type_handler
by Alexander Barkov 01 Dec '16

01 Dec '16

Hello Alexey, Thanks for review! I addressed your suggestions. Also, Vicentiu expressed a concern (when reviewing another patch with the same approach) that this kind of cast that I used in sql_type.cc might be unsafe, because it's non-standard. We're discussion other possible solutions now. So I thought that in the meanwhile for the built-in data types we can create methods: String *val_str_ascii_from_val_int(String *str); String *val_str_ascii_from_val_real(String *str); String *val_str_ascii_from_val_str(String *str); and use them from Type_handler_xxx. A new patch is attached. Please see comments below. On 11/30/2016 05:46 PM, Alexey Botchkov wrote: > Hi, Alexander! > > Basically ok with the patch. > Few comments though: > > > +String *Item_func_hex::val_str_ascii_from_int(String *str, ulonglong > num) > > { > ... > > + char ans[65], *ptr; > > + if (!(ptr= longlong2str(num, ans, 16)) || > > + str->copy(ans, (uint32) (ptr - ans), &my_charset_numeric)) > > + return make_empty_result(); // End of memory > > + return str; > > } > > > We can avoid extra copying here like this: > > { > char *n_end; > str->set_charset(&my_charset_numeric); > if (str->alloc(65) || > !(n_end= longlong2str(num, (char *) str->ptr(), 16))) > return make_empty_result(); > str->length((uint32) (n_end - str->ptr()); > return str; > } Thanks for the idea. It's nice to fix it at once. I moved this to String rather than Item_func_hex though: - this helps to avoid (char*) str->ptr(). - it can be useful for other purposes I tried to move octex2hex() from passport.c to int2str.c and reuse it in sql_string.cc, but failed because of strange compilation failures in mysqlbinlog.cc which includes both password.c and sql_string.cc using #include. From my understanding there is a bug in password.c: instead of #include <mysql.h> it should be: #include "mysql.h" I would like to avoid making changes in here for now, so I gave up trying to reuse octet2str(). Instead of that I added APPEND_HEX() and reused it in 3 places in sql_string.cc. > > > class Item_func_hex :public Item_str_ascii_func > > { > .... > > Shouldn't we implement this too: ? > const Type_handler *Item_func_hex::type_handler() > { return m_handler; } Not really. Type handler for Item_func_hex is type_handler_varchar, and it does not depend on the data type of the argument. To avoid confusion, I renamed m_handler to m_arg0_type_handler. > Personally i wouldn't add the m_handler member at all and just do > const Type_handler *Item_func_hex::type_handler() > { return args[0]->m_handler; } > Are you sure the m_handler member works faster? Well you can just forget > that > comment by now, but i'd like to test this someday. I think we should cache it, because it can be expensive to get the handler of args[0] on every row, as it involves virtual methods, which can call more virtual methods recursively. I added a comment about this. > > > +class Item_func_hex_str: public Item_func_hex > > +{ > > +public: > > + String *val_str_ascii(String *str) > > + { > > + /* Convert given string to a hex string, character by character */ > > + String *res= args[0]->val_str(str); > > + if ((null_value= (!res || tmp_value.alloc(res->length() * 2 + 1)))) > > + return 0; > > + tmp_value.length(res->length() * 2); > > + tmp_value.set_charset(&my_charset_latin1); > > + octet2hex((char*) tmp_value.ptr(), res->ptr(), res->length()); > > + return &tmp_value; > > + } > > +}; > > I think it makes more sence to switch the *str and tmp_value > usage here. So the caller gets the 'str' value he sends to the > function as a result. Not that it changes a lot, still i think it's > nicer. Good idea! Done. > > Best regards. > HF >

1 0