List:Commits« Previous MessageNext Message »
From:Sergey Vojtovich Date:September 6 2007 12:05pm
Subject:bk commit into 5.1 tree (svoj:1.2569) BUG#30583
View as plain text  
Below is the list of changes that have just been committed into a local
5.1 repository of svoj. When svoj does a push these changes will
be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet@stripped, 2007-09-06 17:05:07+05:00, svoj@stripped +3 -0
  BUG#30583 - Partition on DOUBLE key + INNODB + count(*) == crash
  
  Issuing SELECT COUNT(*) against partitioned InnoDB table may cause
  server crash.
  
  Fixed that not all required fields were included into read_set.

  mysql-test/r/partition_innodb.result@stripped, 2007-09-06 17:05:05+05:00, svoj@stripped +7 -0
    A test case for BUG#30583.

  mysql-test/t/partition_innodb.test@stripped, 2007-09-06 17:05:05+05:00, svoj@stripped +8 -0
    A test case for BUG#30583.

  sql/ha_partition.cc@stripped, 2007-09-06 17:05:05+05:00, svoj@stripped +18 -0
    We may get empty read_set for engines that support partial column
    read and ::info doesn't return exact row count. This may happen
    e.g. with SELECT COUNT(*) FROM t1. We must ensure that all columns
    of current key are included into read_set, as partitioning requires
    them for sorting (see ha_partition::handle_ordered_index_scan).

diff -Nrup a/mysql-test/r/partition_innodb.result b/mysql-test/r/partition_innodb.result
--- a/mysql-test/r/partition_innodb.result	2007-04-16 21:16:12 +05:00
+++ b/mysql-test/r/partition_innodb.result	2007-09-06 17:05:05 +05:00
@@ -129,3 +129,10 @@ insert into t1 (time, first_name, last_n
 SELECT * FROM t1 WHERE first_name='Andy' OR last_name='Jake';
 id	time	first_name	last_name
 drop table t1;
+CREATE TABLE t1 (a DOUBLE NOT NULL, KEY(a)) ENGINE=InnoDB
+PARTITION BY KEY(a) PARTITIONS 10;
+INSERT INTO t1 VALUES(1),(2);
+SELECT COUNT(*) FROM t1;
+COUNT(*)
+2
+DROP TABLE t1;
diff -Nrup a/mysql-test/t/partition_innodb.test b/mysql-test/t/partition_innodb.test
--- a/mysql-test/t/partition_innodb.test	2007-04-15 21:45:06 +05:00
+++ b/mysql-test/t/partition_innodb.test	2007-09-06 17:05:05 +05:00
@@ -134,3 +134,11 @@ SELECT * FROM t1 WHERE first_name='Andy'
 
 drop table t1;
 
+#
+# BUG#30583 - Partition on DOUBLE key + INNODB + count(*) == crash
+#
+CREATE TABLE t1 (a DOUBLE NOT NULL, KEY(a)) ENGINE=InnoDB
+PARTITION BY KEY(a) PARTITIONS 10;
+INSERT INTO t1 VALUES(1),(2);
+SELECT COUNT(*) FROM t1;
+DROP TABLE t1;
diff -Nrup a/sql/ha_partition.cc b/sql/ha_partition.cc
--- a/sql/ha_partition.cc	2007-08-02 09:50:52 +05:00
+++ b/sql/ha_partition.cc	2007-09-06 17:05:05 +05:00
@@ -3324,6 +3324,24 @@ int ha_partition::index_init(uint inx, b
   */
   if (m_lock_type == F_WRLCK)
     bitmap_union(table->read_set, &m_part_info->full_part_field_set);
+  else if (!(m_table_flags & HA_STATS_RECORDS_IS_EXACT) &&
+           m_table_flags & HA_PARTIAL_COLUMN_READ)
+  {
+    /*
+      We may get empty read_set for engines that support partial column
+      read and ::info doesn't return exact row count. This may happen
+      e.g. with SELECT COUNT(*) FROM t1. We must ensure that all columns
+      of current key are included into read_set, as partitioning requires
+      them for sorting (see ha_partition::handle_ordered_index_scan).
+
+      TODO: optimize this workaround away from here, as we actually do not
+      need to perform ordered index scan in this case.
+    */
+    uint i;
+    for (i= 0; i < m_curr_key_info->key_parts; i++)
+      bitmap_set_bit(table->read_set,
+                     m_curr_key_info->key_part[i].field->field_index);
+  }
   file= m_file;
   do
   {
Thread
bk commit into 5.1 tree (svoj:1.2569) BUG#30583Sergey Vojtovich6 Sep
  • Re: bk commit into 5.1 tree (svoj:1.2569) BUG#30583Sergei Golubchik7 Sep