Marko Mäkelä added a comment - 2017-12-11 08:06 I am unable to repeat the problem (the test fails differently for me), but I do have a proposed fix: diff --git a/storage/innobase/btr/btr0defragment.cc b/storage/innobase/btr/btr0defragment.cc index cdbe47d48f8..86df5a077a8 100644 --- a/storage/innobase/btr/btr0defragment.cc +++ b/storage/innobase/btr/btr0defragment.cc @@ -831,6 +831,7 @@ DECLARE_THREAD(btr_defragment_thread)(void*) mtr_commit(&mtr); /* Reaching the end of the index. */ dict_stats_empty_defrag_stats(index); + trx->error_state = DB_SUCCESS; ut_d(trx->persistent_stats = true); ++trx->will_lock; dberr_t err = dict_stats_save_defrag_stats(index, trx); diff --git a/storage/innobase/dict/dict0defrag_bg.cc b/storage/innobase/dict/dict0defrag_bg.cc index e43a75b2cf7..5f5edafc9fb 100644 --- a/storage/innobase/dict/dict0defrag_bg.cc +++ b/storage/innobase/dict/dict0defrag_bg.cc @@ -213,6 +213,7 @@ dict_stats_process_entry_from_defrag_pool(trx_t* trx) index_id_t index_id; ut_ad(!srv_read_only_mode); + ut_ad(trx->persistent_stats); /* pop the first index from the auto defrag pool */ if (!dict_stats_defrag_pool_get(&table_id, &index_id)) { @@ -242,6 +243,7 @@ dict_stats_process_entry_from_defrag_pool(trx_t* trx) } mutex_exit(&dict_sys->mutex); + trx->error_state = DB_SUCCESS; ++trx->will_lock; dberr_t err = dict_stats_save_defrag_stats(index, trx); @@ -294,6 +296,9 @@ dict_stats_save_defrag_summary(dict_index_t* index, trx_t* trx) dberr_t dict_stats_save_defrag_stats(dict_index_t* index, trx_t* trx) { + ut_ad(trx->error_state == DB_SUCCESS); + ut_ad(trx->persistent_stats); + if (dict_index_is_ibuf(index)) { return DB_SUCCESS; } diff --git a/storage/innobase/dict/dict0stats_bg.cc b/storage/innobase/dict/dict0stats_bg.cc index 0f76c73bc96..3f5419a0751 100644 --- a/storage/innobase/dict/dict0stats_bg.cc +++ b/storage/innobase/dict/dict0stats_bg.cc @@ -378,6 +378,7 @@ dict_stats_process_entry_from_recalc_pool(trx_t* trx) dict_stats_recalc_pool_add(table); } else { + trx->error_state = DB_SUCCESS; ++trx->will_lock; dict_stats_update(table, DICT_STATS_RECALC_PERSISTENT, trx); if (trx->state != TRX_STATE_NOT_STARTED) { diff --git a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc index a87ab677d04..e7a4532a604 100644 --- a/storage/innobase/handler/ha_innodb.cc +++ b/storage/innobase/handler/ha_innodb.cc @@ -6480,6 +6480,7 @@ ha_innobase::open(const char* name, int, uint) } ut_ad(!trx->persistent_stats); ut_d(trx->persistent_stats = true); + trx->error_state = DB_SUCCESS; ++trx->will_lock; dict_stats_init(ib_table, trx); innobase_commit_low(trx); @@ -13281,6 +13282,7 @@ create_table_info_t::create_table_update_dict() innobase_copy_frm_flags_from_create_info(innobase_table, m_create_info); ++m_trx->will_lock; + m_trx->error_state = DB_SUCCESS; dict_stats_update(innobase_table, DICT_STATS_EMPTY_TABLE, m_trx); innobase_commit_low(m_trx); @@ -14045,6 +14047,7 @@ ha_innobase::rename_table( normalize_table_name(norm_from, from); normalize_table_name(norm_to, to); + trx->error_state = DB_SUCCESS; ++trx->will_lock; ret = dict_stats_rename_table(norm_from, norm_to, errstr, sizeof errstr, trx); diff --git a/storage/innobase/handler/handler0alter.cc b/storage/innobase/handler/handler0alter.cc index d2401bd4e2d..44c40d05b4c 100644 --- a/storage/innobase/handler/handler0alter.cc +++ b/storage/innobase/handler/handler0alter.cc @@ -9073,6 +9073,7 @@ ha_innobase::commit_inplace_alter_table( } row_mysql_unlock_data_dictionary(trx); + trx->error_state = DB_SUCCESS; ++trx->will_lock; /* TODO: The following code could be executed In these places, we are reusing a transaction object that was committed. Transaction commit will not reset the error_state, and sometimes we have error==DB_SUCCESS despite error_state being something else. Apparently this was the case in ha_innobase::rename_table(), most likely because saving the statistics failed for some partition of the table, due to concurrent activity from a background thread that updates persistent statistics.

Julien Muchembled added a comment - 2018-01-12 20:36 Since we upgraded from 10.2.11 to 10.2.12, we also have Lock wait timeouts (but no assertion failure) and regressions in the test suite of our software : > show full processlist; Id User Host db Command Time State Info 2 system user   NULL Daemon NULL InnoDB purge worker NULL 1 system user   NULL Daemon NULL InnoDB purge coordinator NULL 3 system user   NULL Daemon NULL InnoDB purge worker NULL 4 system user   NULL Daemon NULL InnoDB purge worker NULL 5 system user   NULL Daemon NULL InnoDB shutdown handler NULL 164 root localhost test_neo1 Query 47 Opening tables REPLACE INTO config VALUES ('version', '1') 165 root localhost test_neo2 Query 47 creating table CREATE TABLE IF NOT EXISTS pt ( rid INT UNSIGNED NOT NULL, nid INT NOT NULL, state TINYINT UNSIGNED NOT NULL, PRIMARY KEY (rid, nid) ) ENGINE=InnoDB 166 root localhost test_neo3 Query 47 creating table CREATE TABLE config ( name VARBINARY(255) NOT NULL PRIMARY KEY, value VARBINARY(255) NULL ) ENGINE=InnoDB 167 root localhost test_neo4 Query 47 creating table CREATE TABLE config ( name VARBINARY(255) NOT NULL PRIMARY KEY, value VARBINARY(255) NULL ) ENGINE=InnoDB 168 root localhost NULL Query 0 init show full processlist 10 rows in set (0.00 sec)