MySQL patches by Codership

wsrep_consistency_check only handled on SQLCOM_INSERT_SELECT

Bug #1343209 reported by Ben Mildren on 2014-07-17

This bug affects 1 person

	Status	Importance	Assigned to	Milestone
MySQL patches by Codership	Status tracked in 5.6
5.5	Fix Committed	Undecided	Seppo Jaakola	MySQL patches by Codership 5.5.38-25.11
5.6	Fix Released	Undecided	Seppo Jaakola	MySQL patches by Codership 5.6.21-25.9
Percona XtraDB Cluster moved to https://jira.percona.com/projects/PXC	Status tracked in 5.6
5.5	Fix Released	Undecided	Unassigned	Percona XtraDB Cluster moved to https://jira.percona.com/projects/PXC 5.5.39-25.11
5.6	Fix Released	Undecided	Unassigned	Percona XtraDB Cluster moved to https://jira.percona.com/projects/PXC 5.6.20-25.7

Bug Description

pt-table-checksum uses REPLACE ... SELECT to create a checksum of chunks of data, I believe the wsrep_consistency_check was introduced to ensure the isolation would support this process, however it looks like the wsrep_consistency_check is only handled on SQLCOM_INSERT_SELECT and not SQLCOM_REPLACE_SELECT.
In order for pt-table-checksum to be re-runnable it makes sense for it to use SQLCOM_REPLACE_SELECT as opposed to SQLCOM_INSERT_SELECT, so it looks like the wsrep_consistency_check should also/alternatively be applied to the SQLCOM_REPLACE_SELECT.

----------------------
modified sql_parse.cc:
----------------------
#ifdef WITH_WSREP
    if (lex->sql_command == SQLCOM_INSERT_SELECT &&
thd->wsrep_consistency_check == CONSISTENCY_CHECK_DECLARED)
    {
      thd->wsrep_consistency_check = CONSISTENCY_CHECK_RUNNING;
      WSREP_TO_ISOLATION_BEGIN(first_table->db, first_table->table_name, NULL);
    }

....

if (lex->sql_command == SQLCOM_REPLACE_SELECT)
lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_REPLACE_SELECT);

-----------------------
pt-table-checksum code:
-----------------------
   my $checksum_dml = "REPLACE INTO $repl_table "
                    . "(db, tbl, chunk, chunk_index,"
                    . " lower_boundary, upper_boundary, this_cnt, this_crc) "
                    . "SELECT"
                    . ($cluster->is_cluster_node($master_cxn) ? ' /*!99997*/' : '')
                    . " ?, ?, ?, ?, ?, ?,";

See original description

Ben Mildren (ben-mildren) on 2014-07-17

description:

updated

Ben Mildren (ben-mildren) on 2014-07-17

description:

updated

Revision history for this message

Alex Yurchenko (ayurchen) wrote on 2014-07-17:

yes, it looks like removing SQLCOM_INSERT_SELECT check from the if() statement could fix it.

Changed in codership-mysql:
assignee:	nobody → Seppo Jaakola (seppo-jaakola)

Revision history for this message

Nilnandan Joshi (nilnandan-joshi) wrote on 2014-07-23:

Verified with PXC 5.5 and 5.6

----------------------
sql_parse.cc:
----------------------

#ifdef WITH_WSREP
    if (lex->sql_command == SQLCOM_INSERT_SELECT &&
thd->wsrep_consistency_check == CONSISTENCY_CHECK_DECLARED)
    {
      thd->wsrep_consistency_check = CONSISTENCY_CHECK_RUNNING;
      WSREP_TO_ISOLATION_BEGIN(first_table->db, first_table->table_name, NULL);
    }

-----------------------
pt-table-checksum code:
-----------------------

# ########################################################################
   # Checksum args and the DMS part of the checksum query for each table.
   # ########################################################################
   my %crc_args = $rc->get_crc_args(dbh => $master_dbh);
   my $checksum_dml = "REPLACE INTO $repl_table "
                    . "(db, tbl, chunk, chunk_index,"
                    . " lower_boundary, upper_boundary, this_cnt, this_crc) "
                    . "SELECT"
                    . ($cluster->is_cluster_node($master_cxn) ? ' /*!99997*/' : '')
                    . " ?, ?, ?, ?, ?, ?,";
   my $past_cols = " COUNT(*), '0'";