Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Cannot Reproduce
-
2.3.16
-
MXS-SPRINT-99, MXS-SPRINT-101
Description
It looks like: https://jira.mariadb.org/browse/MXS-2408
Is still on 2.3.16:
2020-01-23 15:42:15 error : (872) Lost connection to the master server 'server', closing session. Lost connection to master server while waiting for a result. Connection has been idle for 0.0 seconds. Error caused by: #HY000: Lost connection to backend server: network error. Last close reason: <none>
Attachments
Issue Links
- relates to
-
-
- Closed
-
Activity
Managed to find out a few cases where the network socket error is cleared before the error handler function gets to read it. With this fix, we should be able to see the actual error message that caused the connection to be closed.
Closing as Cannot Reproduce as this hasn't happened with latest releases.
Since this was on MaxScale 2.3.16 which has the improved error logging, we know this event wasn't generated by MaxScale (no (Generated event) after the network error part). This means that it is not a regression of
MXS-2408orMXS-2410and it is in fact a real network error event.The connection idle time calculation could be wrong if a network input event (EPOLLIN) occurs at the same time that a network error does (EPOLLERR). Currently the idle time is unconditionally updated whenever network input events arrive. By updating the idle time only when data is successfully read, we should be able to preserve the real idle time of the connection.