Implement timeout for mysql-lvm's FLUSH TABLES WITH READ LOCK feature
flush tables with read lock can take a very long time. A timeout should be added to kill the query after a period of time. The easiest way would be via two connections - one which just sits and waits for a timeout before issuing a kill query and the other which actually issues the flush commands. I don't think there is a sane way to interrupt MySQLdb otherwise
Blueprint information
- Status:
- Started
- Approver:
- Andrew Garner
- Priority:
- Undefined
- Drafter:
- Andrew Garner
- Direction:
- Needs approval
- Assignee:
- Andrew Garner
- Definition:
- Drafting
- Series goal:
- None
- Implementation:
-
Started
- Milestone target:
- None
- Started by
- Andrew Garner
- Completed by
Related branches
Related bugs
Sprints
Whiteboard
We currently use MySQLdb but any libmysqlclient based connector would have the same properties.
The basic problem (from a python perspective) is that all signals are trapped while we run a FLUSH TABLES WITH READ LOCK. So implementing a timeout is a little non-trivial.
I think the following approach will work well for now, and we can adjust this to a simpler mechanism if we ever move to another connector with nicer signal properties:
import threading
main_connection = MySQLdb.
sentinel_connection = MySQLdb.
def cancel_
cursor = connection.cursor()
cursor.
LOG.
cursor.close()
timer = threading.
timer.start()
try:
main_
except MySQLError:
# probably cancelled, but other errors might occur
LOG.error(...)
raise
else:
timer.cancel()
LOG.info("Lock acquired")
