summaryrefslogtreecommitdiff
path: root/mysql-test/main/flush_corruption.test
diff options
context:
space:
mode:
Diffstat (limited to 'mysql-test/main/flush_corruption.test')
-rw-r--r--mysql-test/main/flush_corruption.test78
1 files changed, 78 insertions, 0 deletions
diff --git a/mysql-test/main/flush_corruption.test b/mysql-test/main/flush_corruption.test
new file mode 100644
index 00000000000..97ef416307f
--- /dev/null
+++ b/mysql-test/main/flush_corruption.test
@@ -0,0 +1,78 @@
+#
+# Test of MyISAM tables with multisession and many waits.
+#
+# This test takes rather long time so let us run it only in --big-test mode
+--source include/big_test.inc
+# We use some debug-only features in this test
+--source include/have_debug.inc
+# We use INFORMATION_SCHEMA.PROCESSLIST in this test
+--source include/not_embedded.inc
+
+--disable_warnings
+drop table if exists t1,t2,t3,t4,t5,t6;
+--enable_warnings
+
+--echo #
+--echo # Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE
+--echo # corrupts a MERGE table
+--echo # Problem #3
+--echo #
+# Two FLUSH TABLES within a LOCK TABLES segment could invalidate the lock.
+# This did *not* require a MERGE table.
+#
+# To increase reproducibility it was necessary to enter a sleep of 2
+# seconds at the end of wait_for_tables() after unlock of LOCK_open. In
+# 5.0 and 5.1 the sleep must be inserted in open_and_lock_tables() after
+# open_tables() instead. wait_for_tables() is not used in this case. The
+# problem was that FLUSH TABLES releases LOCK_open while having unlocked
+# and closed all tables. When this happened while a thread was in the
+# loop in mysql_lock_tables() right after wait_for_tables()
+# (open_tables()) and before retrying to lock, the thread got the lock.
+# And it did not notice that the table needed a refresh after the
+# [re-]open. So it executed its statement on the table.
+#
+# The first FLUSH TABLES kicked the INSERT out of thr_multi_lock() and
+# let it wait in wait_for_tables() (open_table()). The second FLUSH
+# TABLES must happen while the INSERT was on its way from
+# wait_for_tables() (open_table()) to the next call of thr_multi_lock().
+# This needed to be supported by a sleep to make it repeatable.
+#
+CREATE TABLE t1 (c1 INT) ENGINE= MyISAM;
+LOCK TABLE t1 WRITE;
+#SELECT NOW();
+ connect (con1,localhost,root,,);
+ let $con1_id= `SELECT CONNECTION_ID()`;
+ SET @orig_debug=@@global.debug_dbug;
+ SET GLOBAL debug_dbug="+d,sleep_open_and_lock_after_open";
+ send INSERT INTO t1 VALUES (1);
+connection default;
+--echo # Let INSERT go into thr_multi_lock().
+#--sleep 8
+#SELECT ID,STATE,INFO FROM INFORMATION_SCHEMA.PROCESSLIST;
+let $wait_condition= SELECT 1 FROM INFORMATION_SCHEMA.PROCESSLIST
+ WHERE ID = $con1_id AND STATE = 'Waiting for table metadata lock';
+--source include/wait_condition.inc
+#SELECT NOW();
+--echo # Kick INSERT out of thr_multi_lock().
+FLUSH TABLES;
+#SELECT NOW();
+--echo # Let INSERT go through open_tables() where it sleeps.
+#--sleep 8
+#SELECT ID,STATE,INFO FROM INFORMATION_SCHEMA.PROCESSLIST;
+let $wait_condition= SELECT 1 FROM INFORMATION_SCHEMA.PROCESSLIST
+ WHERE ID = $con1_id AND STATE = 'Waiting for table metadata lock';
+--source include/wait_condition.inc
+#SELECT NOW();
+--echo # Unlock and close table and wait for con1 to close too.
+FLUSH TABLES;
+#SELECT NOW();
+--echo # This should give no result.
+SELECT * FROM t1;
+#SELECT NOW();
+UNLOCK TABLES;
+ connection con1;
+ reap;
+ SET GLOBAL debug_dbug=@orig_debug;
+ disconnect con1;
+connection default;
+DROP TABLE t1;