Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.6
Description
With the merge of MDEV-32365 into bb-10.6-semisync, the reporting of a net error results in a segfault. This is because the slave object is deleted, and then it is referenced for its server_id when reporting the error. See the following code snippet:
else if (net.last_errno == ER_NET_READ_ERROR)
|
{
|
it.remove();
|
if (net.last_errno > 0 && global_system_variables.log_warnings > 2)
|
sql_print_warning("Semisync ack receiver got error %d \"%s\" "
|
"from slave server-id %d",
|
net.last_errno, ER_DEFAULT(net.last_errno),
|
slave->server_id());
|
}
|
The statement it.remove(); should happen after the warning is issued.
Also, with the merge, the error log can report 3 separate messages when a slave disconnects, which seems potentially annoying/confusing for users:
2023-12-06 14:30:21 10 [Warning] Semisync ack receiver got error 1158 "Got an error reading communication packets" from slave server-id 2
|
2023-12-06 14:30:21 45 [Warning] Aborted connection 44 to db: 'unconnected' user: 'root' host: 'localhost' (KILLED)
|
2023-12-06 14:30:21 44 [Note] Stop semi-sync binlog_dump to slave (server_id: 2)
|
Attachments
Issue Links
- is caused by
-
MDEV-32365 detailize the semisync replication magic number error
-
- Closed
-
-
-
- Closed
-