Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
11.4, 11.8, 12.0
-
None
Description
Test binlog_encryption.encrypted_master_switch_to_unencrypted_gtid fails on buildbot sporadically starting in 11.4 with error output:
binlog_encryption.encrypted_master_switch_to_unencrypted_gtid 'mix' w3 [ fail ]
|
Test ended at 2025-03-08 22:17:20
|
|
|
CURRENT_TEST: binlog_encryption.encrypted_master_switch_to_unencrypted_gtid
|
mysqltest: At line 129: Slave without encryption configured should fail to read encrypted binlog (expected 0-1-5 but got )
|
|
|
The result from queries just before the failure was:
|
< snip >
|
#####################################################
|
# Part 3: restart master again without encryption
|
#####################################################
|
connection default;
|
connection server_1;
|
CREATE TABLE table3_no_encryption (
|
pk INT AUTO_INCREMENT PRIMARY KEY,
|
ts TIMESTAMP NULL,
|
b BLOB
|
) ENGINE=MyISAM;
|
INSERT INTO table3_no_encryption VALUES (NULL,NOW(),'data_no_encryption');
|
INSERT INTO table3_no_encryption SELECT NULL,NOW(),b FROM table3_no_encryption;
|
INSERT INTO table3_no_encryption SELECT NULL,NOW(),b FROM table3_no_encryption;
|
#####################################################
|
# Check: resume replication and check how it goes
|
#####################################################
|
connection server_2;
|
start slave;
|
include/wait_for_slave_io_error.inc [errno=1236]
|
# Ensuring slave was unable to replicate any encrypted transactions..
|
Attachments
Issue Links
- is caused by
-
MDEV-4991 GTID binlog indexing
-
- Closed
-
Activity
Right, I see. Yes, the test case waits for the IO thread to stop with an error, then checks the value of @@GLOBAL.gtid_slave_pos. If the SQL thread gets delayed for some reason, it is possible for the test case to read the value before the SQL thread has had time to update it. There's simply missing a wait for the SQL thread to be ready.
Failure can be repeated reliably with the below patch that puts a small delay at the critical point in the SQL thread.
diff --git a/sql/slave.cc b/sql/slave.cc
|
index 8b5797de935..ac1c4386302 100644
|
--- a/sql/slave.cc
|
+++ b/sql/slave.cc
|
@@ -5790,6 +5790,7 @@ pthread_handler_t handle_slave_sql(void *arg)
|
/* Read queries from the IO/THREAD until this thread is killed */
|
|
thd->set_command(COM_SLAVE_SQL);
|
+my_sleep(500000);
|
while (!sql_slave_killed(serial_rgi))
|
{
|
THD_STAGE_INFO(thd, stage_reading_event_from_the_relay_log);
|
Hi knielsen!
Assigning to you, as the snippet that fails was added in MDEV-4991:
--echo # Ensuring slave was unable to replicate any encrypted transactions..
|
--let $gsp= `SELECT @@global.gtid_slave_pos`
|
if (`SELECT strcmp("$gsp","$last_unencrypted_gtid")`)
|
{
|
die Slave without encryption configured should fail to read encrypted binlog (expected $last_unencrypted_gtid but got $gsp);
|
}
|
Pushed to 11.4