Re: [Maria-developers] MDEV-16329 ALTER ONLINE TABLE

Hi, Nikita, this is a review of the combined diff of your branch. Of `git diff github/10.8...github/bb-10.8-online-alter`, that is of cf7cc376bac fix savepoints in myisam 4a128ca5458 fix main.delayed 7300151cc84 fix skipping rocksdb 19504dfdbcc add binlog/standalone combinations 55e993595aa add rocksdb combination 06645b17450 savepoints 9ffcd13b9fe rename tests 730926ae962 MDEV-16329 [5/5] ALTER ONLINE TABLE cf52ea33a70 MDEV-16329 [4/5] Refactor MYSQL_BIN_LOG: extract Event_log ancestor 0445c8f1ee9 MDEV-16329 [3/5] use binlog_cache_data directly in most places 3f3da0f65a3 MDEV-16329 [2/5] refactor binlog and cache_mngr dc79667ca73 MDEV-16329 [1/5] add THD::binlog_get_cache_mngr 6f9dd00ce2d rpl: repack table_def faf8680df8d Copy_field: add const to arguments It was mostly fine the last time already, so only few comments below:

diff --git a/mysql-test/include/have_log_bin_off.require b/mysql-test/include/have_log_bin_off.require new file mode 100644 index 00000000000..979dbe75f80 --- /dev/null +++ b/mysql-test/include/have_log_bin_off.require @@ -0,0 +1,2 @@ +Variable_name Value +log_bin OFF

better use `if (...) { skip }` pattern *.require is an obsolete 20-year old technique from before if- and skip-times

diff --git a/mysql-test/main/alter_table.test b/mysql-test/main/alter_table.test index 31c69783248..91423ee8c1f 100644 --- a/mysql-test/main/alter_table.test +++ b/mysql-test/main/alter_table.test @@ -1532,8 +1532,12 @@ ALTER TABLE t1 DROP INDEX i1, DROP INDEX i2, DROP INDEX i3, DROP INDEX i4; ALTER TABLE t1 ADD INDEX i1(b), ALGORITHM= INPLACE, LOCK= NONE; ALTER TABLE t1 ADD INDEX i2(b), ALGORITHM= INPLACE, LOCK= SHARED; ALTER TABLE t1 ADD INDEX i3(b), ALGORITHM= INPLACE, LOCK= EXCLUSIVE; ---error ER_ALTER_OPERATION_NOT_SUPPORTED_REASON +--disable_info +--disable_warnings +--error 0,ER_ALTER_OPERATION_NOT_SUPPORTED_REASON

why, because of embedded?

ALTER TABLE t1 ADD INDEX i4(b), ALGORITHM= COPY, LOCK= NONE; +--enable_warnings +--enable_info ALTER TABLE t1 ADD INDEX i5(b), ALGORITHM= COPY, LOCK= SHARED; ALTER TABLE t1 ADD INDEX i6(b), ALGORITHM= COPY, LOCK= EXCLUSIVE;

diff --git a/mysql-test/main/alter_table_locknone.result b/mysql-test/main/alter_table_locknone.result new file mode 100644 index 00000000000..ea040925efe --- /dev/null +++ b/mysql-test/main/alter_table_locknone.result @@ -0,0 +1,270 @@ +create table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')) engine=myisam; +insert into t1 (a) values (1),(2),(3); +alter online table t1 modify b int default 5, alter c set default 'X'; +alter online table t1 change b new_name int; +alter online table t1 modify e enum('a','b','c'); +alter online table t1 comment "new comment"; +alter table t1 add constraint q check (a > 0); +alter online table t1 drop constraint q; +alter online table t1 algorithm=INPLACE, lock=NONE; +alter online table t1; +alter table t1 algorithm=INPLACE; +alter table t1 lock=NONE; +show create table t1; +Table Create Table +t1 CREATE TABLE `t1` ( + `a` int(11) NOT NULL, + `new_name` int(11) DEFAULT NULL, + `c` varchar(80) DEFAULT 'X', + `e` enum('a','b','c') DEFAULT NULL, + PRIMARY KEY (`a`) +) ENGINE=MyISAM DEFAULT CHARSET=latin1 COMMENT='new comment' +drop table t1; +create temporary table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')); +insert into t1 (a) values (1),(2),(3); +alter online table t1 modify b int default 5, alter c set default 'X'; +alter online table t1 change b new_name int; +alter online table t1 modify e enum('a','b','c'); +alter online table t1 comment "new comment"; +alter online table t1 rename to t2; +show create table t2; +Table Create Table +t2 CREATE TEMPORARY TABLE `t2` ( + `a` int(11) NOT NULL, + `new_name` int(11) DEFAULT NULL, + `c` varchar(80) DEFAULT 'X', + `e` enum('a','b','c') DEFAULT NULL, + PRIMARY KEY (`a`) +) ENGINE=MyISAM DEFAULT CHARSET=latin1 COMMENT='new comment' +drop table t2; +create table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')) engine=aria; +insert into t1 (a) values (1),(2),(3); +alter online table t1 modify b int default 5; +alter online table t1 change b new_name int; +alter online table t1 modify e enum('a','b','c'); +alter online table t1 comment "new comment"; +show create table t1; +Table Create Table +t1 CREATE TABLE `t1` ( + `a` int(11) NOT NULL, + `new_name` int(11) DEFAULT NULL, + `c` varchar(80) DEFAULT NULL, + `e` enum('a','b','c') DEFAULT NULL, + PRIMARY KEY (`a`) +) ENGINE=Aria DEFAULT CHARSET=latin1 PAGE_CHECKSUM=1 COMMENT='new comment' +alter online table t1 page_checksum=1; +alter online table t1 page_checksum=0; +ERROR 0A000: LOCK=NONE is not supported for this operation. Try LOCK=SHARED

Why LOCK=NONE would be not supported now? Shouldn't your new feature allow just everything with LOCK=NONE?

+drop table t1; +create table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')); +insert into t1 (a) values (1),(2),(3); +alter online table t1 drop column b, add b int; +ERROR 0A000: LOCK=NONE is not supported for this operation. Try LOCK=SHARED ... diff --git a/mysql-test/main/alter_table_online.combinations b/mysql-test/main/alter_table_online.combinations new file mode 100644 index 00000000000..ae144432c68 --- /dev/null +++ b/mysql-test/main/alter_table_online.combinations @@ -0,0 +1,2 @@ +[innodb] +[rocksdb]

what about non-trans tables?

diff --git a/mysql-test/main/alter_table_online.test b/mysql-test/main/alter_table_online.test index 6ef9661c43b..815483d252d 100644 --- a/mysql-test/main/alter_table_online.test +++ b/mysql-test/main/alter_table_online.test @@ -1,335 +1,596 @@ -# -# Test of ALTER ONLINE TABLE syntax -# - ---source include/have_innodb.inc ---source include/have_partition.inc -# -# Test of things that can be done online -# - -create table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')) engine=myisam; -insert into t1 (a) values (1),(2),(3); - -alter online table t1 modify b int default 5, alter c set default 'X'; -alter online table t1 change b new_name int; -alter online table t1 modify e enum('a','b','c'); -alter online table t1 comment "new comment"; -alter table t1 add constraint q check (a > 0); -alter online table t1 drop constraint q; - -# No OPs - -alter online table t1 algorithm=INPLACE, lock=NONE; -alter online table t1; -alter table t1 algorithm=INPLACE; -alter table t1 lock=NONE; -show create table t1; -drop table t1; +-- source include/have_debug_sync.inc +-- source include/not_embedded.inc +-- source alter_table_online_binlog.inc

-# -# everything with temporary tables is "online", i.e. without locks -# -create temporary table t1 (a int not null primary key, b int, c varchar(80), e enum('a','b')); -insert into t1 (a) values (1),(2),(3); - -alter online table t1 modify b int default 5, alter c set default 'X'; -alter online table t1 change b new_name int; -alter online table t1 modify e enum('a','b','c'); -alter online table t1 comment "new comment"; -alter online table t1 rename to t2; -show create table t2; -drop table t2; +-- disable_query_log +-- if ($MTR_COMBINATION_INNODB) { +-- source include/have_innodb.inc +set default_storage_engine= innodb; +-- } +-- if ($MTR_COMBINATION_ROCKSDB) { +-- source include/have_rocksdb.inc +set default_storage_engine= rocksdb; +-- } +-- enable_query_log + +-- let $default_engine= `select @@default_storage_engine` + +--connect (con2, localhost, root,,) +--connection default + + +--echo # +--echo # Test insert +--echo # + +--echo # Insert and add column +create or replace table t1 (a int) engine=innodb;

why do you specify the engine explicitly if you run it in two engine combinations?

+insert t1 values (5); + +--connection con2 +--send +set debug_sync= 'now WAIT_FOR ended'; + +--connection default +set debug_sync= 'alter_table_copy_end SIGNAL ended WAIT_FOR end'; + ... + +--connection default +--reap +--sorted_result +select * from t1; + +--echo # +--echo # MYISAM. Only Inserts can be tested.

why?

+--echo # + +create or replace table t1 (a int) engine=myisam; +insert t1 values (5); + +--connection con2 +--send +set debug_sync= 'now WAIT_FOR ended'; + +--connection default +set debug_sync= 'alter_table_copy_end SIGNAL ended WAIT_FOR end'; + +--send +alter table t1 add b int NULL, algorithm= copy, lock= none; + ... diff --git a/mysql-test/main/mdl_sync.test b/mysql-test/main/mdl_sync.test index 3df19aca806..8c6bc76f5ca 100644 --- a/mysql-test/main/mdl_sync.test +++ b/mysql-test/main/mdl_sync.test @@ -42,7 +42,7 @@ connection con1; set debug_sync='mdl_upgrade_lock SIGNAL parked WAIT_FOR go'; --send alter table t1 rename t3

-connection default; + connection default;

eh? (here and many other changes in this file)

--echo connection: default set debug_sync= 'now WAIT_FOR parked';

diff --git a/mysql-test/suite/gcol/inc/gcol_select.inc b/mysql-test/suite/gcol/inc/gcol_select.inc index 2386c55fdbc..8f6314233bd 100644 --- a/mysql-test/suite/gcol/inc/gcol_select.inc +++ b/mysql-test/suite/gcol/inc/gcol_select.inc @@ -872,7 +872,9 @@ CREATE TABLE t1(a INT); INSERT INTO t1 VALUES(2147483647); ALTER TABLE t1 ADD COLUMN h INT AS (a) VIRTUAL; ALTER TABLE t1 CHANGE h i INT AS (a) VIRTUAL, ALGORITHM=COPY; +--error ER_WARN_DATA_OUT_OF_RANGE,ER_ALTER_OPERATION_NOT_SUPPORTED_REASON

why two? embedded, again?

ALTER TABLE t1 ADD COLUMN b SMALLINT AS (a) VIRTUAL, ALGORITHM=COPY, LOCK=NONE; +--error ER_WARN_DATA_OUT_OF_RANGE,ER_ALTER_OPERATION_NOT_SUPPORTED_REASON ALTER TABLE t1 ADD COLUMN e SMALLINT AS (a) VIRTUAL, ALGORITHM=COPY, LOCK=NONE; ALTER TABLE t1 ADD COLUMN f SMALLINT AS (a) VIRTUAL, ALGORITHM=COPY, LOCK=SHARED; ALTER TABLE t1 ADD COLUMN g SMALLINT AS (a) VIRTUAL, ALGORITHM=COPY, LOCK=EXCLUSIVE; diff --git a/sql/handler.h b/sql/handler.h index fe61666bf20..5e46fa59d7e 100644 --- a/sql/handler.h +++ b/sql/handler.h @@ -3538,6 +3543,7 @@ class handler :public Sql_alloc /** to be actually called to get 'check()' functionality*/ int ha_check(THD *thd, HA_CHECK_OPT *check_opt); int ha_repair(THD* thd, HA_CHECK_OPT* check_opt); + virtual void open_read_view(){}

why did you change to that from start_consistent_snapshot()?

void ha_start_bulk_insert(ha_rows rows, uint flags= 0) { DBUG_ENTER("handler::ha_start_bulk_insert"); diff --git a/sql/log.cc b/sql/log.cc index 70ceecc66f8..9beab80773c 100644 --- a/sql/log.cc +++ b/sql/log.cc @@ -101,6 +102,9 @@ static int binlog_flush_cache(THD *thd, binlog_cache_mngr *cache_mngr, Log_event *end_ev, bool all, bool using_stmt, bool using_trx, bool is_ro_1pc);

+int binlog_online_alter_commit(THD *thd, bool all); +void binlog_online_alter_rollback(THD *thd, bool all);

static?

+ static const LEX_CSTRING write_error_msg= { STRING_WITH_LEN("error writing to the binary log") };

@@ -3565,6 +3613,91 @@ bool MYSQL_BIN_LOG::open_index_file(const char *index_file_name_arg, }

+bool Event_log::open(const char *log_name, + const char *new_name, ulong next_file_number, + enum cache_type io_cache_type_arg) +{ + bool error= false; + if (log_name || new_name) + { + error= MYSQL_LOG::open( +#ifdef HAVE_PSI_INTERFACE + 0, +#endif + log_name, LOG_NORMAL, new_name, next_file_number, io_cache_type_arg); + } + else + { +#ifdef HAVE_PSI_INTERFACE + /* Keep the key for reopen */ + m_log_file_key= 0; +#endif + error= init_io_cache(&log_file, -1, LOG_BIN_IO_SIZE, + io_cache_type_arg, 0, 0, + MYF(MY_WME | MY_NABP | MY_WAIT_IF_FULL)); + + log_state= LOG_OPENED; + inited= true; + } + if (error) + return error; + + longlong bytes_written= write_description_event( + (enum_binlog_checksum_alg)binlog_checksum_options, + encrypt_binlog, false); + error= bytes_written < 0; + return error; +} + +longlong +Event_log::write_description_event(enum_binlog_checksum_alg checksum_alg, + bool encrypt, bool dont_set_created) +{ + Format_description_log_event s(BINLOG_VERSION); + /* + don't set LOG_EVENT_BINLOG_IN_USE_F for SEQ_READ_APPEND io_cache + as we won't be able to reset it later + */ + if (io_cache_type == WRITE_CACHE)

this was incorrectly rebased. in the previous patch you *moved* this block of code from MYSQL_BIN_LOG::open into a separate method. here you copied it. you forgot to delete it from MYSQL_BIN_LOG::open().

+ s.flags |= LOG_EVENT_BINLOG_IN_USE_F; + s.checksum_alg= checksum_alg; + + crypto.scheme = 0; + DBUG_ASSERT(s.checksum_alg != BINLOG_CHECKSUM_ALG_UNDEF); + if (!s.is_valid()) + return -1; + s.dont_set_created= dont_set_created; + if (write_event(&s, 0, &log_file)) + return -1; + + if (encrypt) + { + uint key_version= encryption_key_get_latest_version(ENCRYPTION_KEY_SYSTEM_DATA); + if (key_version == ENCRYPTION_KEY_VERSION_INVALID) + { + sql_print_error("Failed to enable encryption of binary logs"); + return -1; + } + + if (key_version != ENCRYPTION_KEY_NOT_ENCRYPTED) + { + if (my_random_bytes(crypto.nonce, sizeof(crypto.nonce))) + return -1; + + Start_encryption_log_event sele(1, key_version, crypto.nonce); + sele.checksum_alg= s.checksum_alg; + if (write_event(&sele, 0, &log_file)) + return -1; + + // Start_encryption_log_event is written, enable the encryption + if (crypto.init(sele.crypto_scheme, key_version)) + return -1; + } + } + return (longlong)s.data_written; +} + + /** Open a (new) binlog file.

@@ -7252,6 +7477,160 @@ class CacheWriter: public Log_event_writer bool first; };

+int cache_copy(IO_CACHE *to, IO_CACHE *from) +{ + DBUG_ENTER("cache_copy"); + if (reinit_io_cache(from, READ_CACHE, 0, 0, 0)) + DBUG_RETURN(ER_ERROR_ON_WRITE); + size_t bytes_in_cache= my_b_bytes_in_cache(from); + + do + { + my_b_write(to, from->read_pos, bytes_in_cache); + + from->read_pos += bytes_in_cache; + bytes_in_cache= my_b_fill(from); + if (from->error || to->error) + DBUG_RETURN(ER_ERROR_ON_WRITE); + } while (bytes_in_cache); + + DBUG_RETURN(0); +} + +int binlog_online_alter_commit(THD *thd, bool all) +{ + DBUG_ENTER("online_alter_commit"); + int error= 0; +#ifdef HAVE_REPLICATION + + if (thd->online_alter_cache_list.empty()) + DBUG_RETURN(0); + + bool is_ending_transaction= ending_trans(thd, all); + + for (auto &cache_mngr: thd->online_alter_cache_list) + { + auto *binlog= cache_mngr.share->online_alter_binlog; + DBUG_ASSERT(binlog); + + error= binlog_flush_pending_rows_event(thd, + /* + do not set STMT_END for last event + to leave table open in altering thd + */ + false, + true, + binlog, + is_ending_transaction + ? &cache_mngr.trx_cache + : &cache_mngr.stmt_cache); + if (is_ending_transaction) + { + mysql_mutex_lock(binlog->get_log_lock()); + error= binlog->write_cache(thd, &cache_mngr.trx_cache.cache_log); + + mysql_mutex_unlock(binlog->get_log_lock()); + } + else + { + error= cache_copy(&cache_mngr.trx_cache.cache_log, + &cache_mngr.stmt_cache.cache_log);

Why do you need two caches here? It seems you could have just one trx_cache and if the statement fails you simply truncate it to the beginning of the statement.

+ } + + if (error) + { + my_error(ER_ERROR_ON_WRITE, MYF(ME_ERROR_LOG), + binlog->get_name(), errno); + binlog_online_alter_cleanup(thd->online_alter_cache_list, + is_ending_transaction); + DBUG_RETURN(error); + } + } + + binlog_online_alter_cleanup(thd->online_alter_cache_list, + is_ending_transaction); + + for (TABLE *table= thd->open_tables; table; table= table->next) + { + table->online_alter_cache= NULL;

Why?

+ } +#endif // HAVE_REPLICATION + DBUG_RETURN(error); +} + +void binlog_online_alter_rollback(THD *thd, bool all) +{ +#ifdef HAVE_REPLICATION + bool is_ending_trans= ending_trans(thd, all); + + /* + This is a crucial moment that we are running through + thd->online_alter_cache_list, and not through thd->open_tables to cleanup + stmt cache, though both have it. The reason is that the tables can be closed + to that moment in case of an error. + The same reason applies to the fact we don't store cache_mngr in the table + itself -- because it can happen to be not existing. + Still in case if tables are left opened + */ + binlog_online_alter_cleanup(thd->online_alter_cache_list, is_ending_trans); +#endif // HAVE_REPLICATION +} + +SAVEPOINT** find_savepoint_in_list(THD *thd, LEX_CSTRING name, + SAVEPOINT ** const list); + +SAVEPOINT* savepoint_add(THD *thd, LEX_CSTRING name, SAVEPOINT **list, + int (*release_old)(THD*, SAVEPOINT*));

may be in transaction.h?

+ +int online_alter_savepoint_set(THD *thd, LEX_CSTRING name) +{ + + DBUG_ENTER("binlog_online_alter_savepoint"); +#ifdef HAVE_REPLICATION + if (thd->online_alter_cache_list.empty()) + DBUG_RETURN(0); + + if (savepoint_alloc_size < sizeof (SAVEPOINT) + sizeof(my_off_t)) + savepoint_alloc_size= sizeof (SAVEPOINT) + sizeof(my_off_t); + + for (auto &cache: thd->online_alter_cache_list) + { + if (cache.share->db_type()->savepoint_set == NULL) + continue; + + SAVEPOINT *sv= savepoint_add(thd, name, &cache.sv_list, NULL);

wouldn't it be simpler to log SAVEPOINT and ROLLBACK/RELEASE as Query_log_event to the cache? Then you won't need to do any maintenance.

+ if(unlikely(sv == NULL)) + DBUG_RETURN(1); + my_off_t *pos= (my_off_t*)(sv+1); + *pos= cache.trx_cache.get_byte_position(); + + sv->prev= cache.sv_list; + cache.sv_list= sv; + } +#endif + DBUG_RETURN(0); +} + +int online_alter_savepoint_rollback(THD *thd, LEX_CSTRING name) +{ + DBUG_ENTER("online_alter_savepoint_rollback"); +#ifdef HAVE_REPLICATION + for (auto &cache: thd->online_alter_cache_list) + { + if (cache.share->db_type()->savepoint_set == NULL) + continue; + + SAVEPOINT **sv= find_savepoint_in_list(thd, name, &cache.sv_list); + // sv is null if savepoint was set up before online table was modified + my_off_t pos= *sv ? *(my_off_t*)(*sv+1) : 0; + + cache.trx_cache.restore_savepoint(pos); + } + +#endif + DBUG_RETURN(0); +} + /* Write the contents of a cache to the binary log.

diff --git a/sql/rpl_rli.h b/sql/rpl_rli.h index cc807852bf2..57757ecf3dd 100644 --- a/sql/rpl_rli.h +++ b/sql/rpl_rli.h @@ -900,14 +900,20 @@ struct rpl_group_info } }

- bool get_table_data(TABLE *table_arg, table_def **tabledef_var, TABLE **conv_table_var) const + bool get_table_data(TABLE *table_arg, table_def **tabledef_var, + TABLE **conv_table_var, + const Copy_field *copy[], const Copy_field **copy_end) const { DBUG_ASSERT(tabledef_var && conv_table_var); for (TABLE_LIST *ptr= tables_to_lock ; ptr != NULL ; ptr= ptr->next_global) if (ptr->table == table_arg) { - *tabledef_var= &static_cast(ptr)->m_tabledef; - *conv_table_var= static_cast(ptr)->m_conv_table; + auto *rpl_table_list= static_cast(ptr); + if (rpl_table_list->m_tabledef_valid)

when can it be false?

+ *tabledef_var= &rpl_table_list->m_tabledef; + *conv_table_var= rpl_table_list->m_conv_table; + *copy= rpl_table_list->m_online_alter_copy_fields; + *copy_end= rpl_table_list->m_online_alter_copy_fields_end; DBUG_PRINT("debug", ("Fetching table data for table %s.%s:" " tabledef: %p, conv_table: %p", table_arg->s->db.str, table_arg->s->table_name.str, diff --git a/sql/sql_table.cc b/sql/sql_table.cc index b07efb29bba..9dfe2178667 100644 --- a/sql/sql_table.cc +++ b/sql/sql_table.cc @@ -9533,6 +9539,19 @@ bool mysql_alter_table(THD *thd, const LEX_CSTRING *new_db, has been already processed. */ table_list->required_type= TABLE_TYPE_NORMAL; + + + if (alter_info->requested_lock == Alter_info::ALTER_TABLE_LOCK_SHARED + || alter_info->requested_lock == Alter_info::ALTER_TABLE_LOCK_EXCLUSIVE + || thd->locked_tables_mode == LTM_LOCK_TABLES + || thd->lex->sql_command == SQLCOM_OPTIMIZE + || alter_info->algorithm(thd) == Alter_info::ALTER_TABLE_ALGORITHM_NOCOPY)

why only NOCOPY? What about INPLACE and INSTANT? the rest of the condition looks fine.

+ online= false; + + if (online) + { + table_list->lock_type= TL_READ; + }

DEBUG_SYNC(thd, "alter_table_before_open_tables");

@@ -10961,6 +10983,58 @@ bool mysql_trans_commit_alter_copy_data(THD *thd) DBUG_RETURN(error); }

+#ifdef HAVE_REPLICATION +static int online_alter_read_from_binlog(THD *thd, rpl_group_info *rgi, + Cache_flip_event_log *log) +{ + MEM_ROOT event_mem_root; + Query_arena backup_arena; + Query_arena event_arena(&event_mem_root, Query_arena::STMT_INITIALIZED); + init_sql_alloc(key_memory_gdl, &event_mem_root, + MEM_ROOT_BLOCK_SIZE, 0, MYF(0)); + + int error= 0; + + IO_CACHE *log_file= log->flip(); + + thd_progress_report(thd, 0, my_b_write_tell(log_file)); + + Abort_on_warning_instant_set old_abort_on_warning(thd, 0); + do + { + const auto *descr_event= rgi->rli->relay_log.description_event_for_exec; + auto *ev= Log_event::read_log_event(log_file, descr_event, false); + if (!ev) + break; + + ev->thd= thd; + thd->set_n_backup_active_arena(&event_arena, &backup_arena); + error= ev->apply_event(rgi); + thd->restore_active_arena(&event_arena, &backup_arena); + + event_arena.free_items(); + free_root(&event_mem_root, MYF(MY_KEEP_PREALLOC)); + if (ev != rgi->rli->relay_log.description_event_for_exec) + delete ev; + thd_progress_report(thd, my_b_tell(log_file), thd->progress.max_counter); + DEBUG_SYNC(thd, "alter_table_online_progress"); + } while(!error); + + return error; +} +#endif // HAVE_REPLICATION + +static void online_alter_cleanup_binlog(THD *thd, TABLE_SHARE *s) +{ +#ifdef HAVE_REPLICATION + if (!s->online_alter_binlog) + return; + // s->online_alter_binlog->reset_logs(thd, false, NULL, 0, 0);

forgot to delete?

+ s->online_alter_binlog->cleanup(); + s->online_alter_binlog->~Cache_flip_event_log(); + s->online_alter_binlog= NULL; +#endif +}

static int copy_data_between_tables(THD *thd, TABLE *from, TABLE *to, @@ -11306,6 +11431,76 @@ copy_data_between_tables(THD *thd, TABLE *from, TABLE *to, cleanup_done= 1; to->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY);

+#ifdef HAVE_REPLICATION + if (likely(online && error < 0)) + { + Ha_trx_info *trx_info_save= thd->transaction->all.ha_list; + thd->transaction->all.ha_list = NULL;

why?

+ thd_progress_next_stage(thd); + Table_map_log_event table_event(thd, from, from->s->table_map_id, + from->file->has_transactions()); + Relay_log_info rli(false); + rpl_group_info rgi(&rli); + RPL_TABLE_LIST rpl_table(to, TL_WRITE, from, table_event.get_table_def(), + copy, copy_end); + Cache_flip_event_log *binlog= from->s->online_alter_binlog; + rgi.thd= thd; + rgi.tables_to_lock= &rpl_table; + + rgi.m_table_map.set_table(from->s->table_map_id, to); + + DBUG_ASSERT(binlog->is_open()); + + rli.relay_log.description_event_for_exec= + new Format_description_log_event(4); + + // We restore bitmaps, because update event is going to mess up with them. + to->default_column_bitmaps(); + + error= online_alter_read_from_binlog(thd, &rgi, binlog); + + DEBUG_SYNC(thd, "alter_table_online_before_lock"); + + int lock_error= + thd->mdl_context.upgrade_shared_lock(from->mdl_ticket, MDL_EXCLUSIVE, + (double)thd->variables.lock_wait_timeout); + if (!error) + error= lock_error; + + if (!error) + { + thd_progress_next_stage(thd); + error= online_alter_read_from_binlog(thd, &rgi, binlog); + } + + thd->transaction->all.ha_list = trx_info_save; + } + else if (unlikely(online)) // error was on copy stage + { + /* + We need to issue a barrier to clean up gracefully. + Without this, following possible: + T1: ALTER TABLE starts + T2: INSERT starts + T1: ALTER TABLE fails with error (i.e. ER_DUP_KEY) + T1: from->s->online_alter_binlog sets to NULL + T2: INSERT committs + T2: thd->online_alter_cache_list is not empty + T2: binlog_commit: DBUG_ASSERT(binlog); is issued.

1. do you have a test for that? 2. can it be fixed, like, on iterating thd->online_alter_cache_list thd notices that from->s->online_alter_cache_list is NULL, so it simply discards the cache? Then you won't need MDL_SHARED_NO_WRITE

+ */ + // Ignore the return result. We already have an error. + thd->mdl_context.upgrade_shared_lock(from->mdl_ticket, + MDL_SHARED_NO_WRITE, + thd->variables.lock_wait_timeout); + } +#endif + + if (error > 0 && !from->s->tmp_table) + { + /* We are going to drop the temporary table */ + to->file->extra(HA_EXTRA_PREPARE_FOR_DROP); + } + DEBUG_SYNC(thd, "copy_data_between_tables_before_reset_backup_lock"); if (backup_reset_alter_copy_lock(thd)) error= 1;

Regards, Sergei VP of MariaDB Server Engineering and security@mariadb.org