set a hard limit for internal temporary tables
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Percona Cluster Charm |
Triaged
|
Medium
|
Unassigned |
Bug Description
Today I handled an alert on a customer cloud (bionic-rocky) in which one of the mysql units was *very rapidly* running out of space.
It turned out the culprit was a >300GB ibtmp1 file, which according to[0] and [1]:
* grows indefinitely unless capped
* is never shrunk
* is only truncated at service restart
The recommendation is therefore to explicitly limit its size with:
innodb_
This does not avoid the need to restart the mysql service if ibtmp1 reaches the threshold, as some queries would in that case fail, but at least ensures that a growing temporary file won't bring down an entire node by filling up all available space.
For the reason above, an nrpe check alerting when the file does reach the limit should also be provided (to signal an operator that a service restart should be performed when possible).
This seems to only apply to mysql 5.7.
[0] https:/
[1] https:/
Changed in charm-percona-cluster: | |
status: | New → Triaged |
importance: | Undecided → Medium |
Subscribing field medium, the only known work around to this is monitoring and service restarts which isn't a great solution and could still have a cluster go down before it's resolved.