Juju Charms Collection
percona-cluster package

openstack-on-lxd: mysql server hits OOM

Bug #1556901 reported by Dimitri John Ledkov
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
percona-cluster (Juju Charms Collection)
New
Undecided
Unassigned

Bug Description

I have deployed openstack-on-lxd bundle, on xenial, on s390x.

That specifies:

charm: cs:~openstack-charmers-next/xenial/percona-cluster
options:
  max-connections: 20000

Unfortunately that runs into OOM and fails.

My machine has 12GB of RAM and 4GB of swap, is that not enough for this bundle?

I had to reduce max-connections down to 200 to get the bundle to deploy.

I see https://bugs.mysql.com/bug.php?id=68514 which mentions that various things needs to be sized correctly with large max-connections, is the charm in question do that? and/or should the percona database be patched to do same/similar as the fix in MySQL 5.7.6 as per comments there? Does percona include fixes from 5.7.6, if they are at all applicable?

Tags: s390x xenial
James Page (james-page)
affects: nova-cloud-controller (Juju Charms Collection) → percona-cluster (Juju Charms Collection)
Revision history for this message
James Page (james-page) wrote :

Trying to repro - this maybe a side effect of the number of cores the machine is presenting; every service will be creating processes based on cpu count x2 all of which require mysql connections, which are then pooled.

Revision history for this message
James Page (james-page) wrote :

I suspect this is also a side effect of the fact that every LXD container sees the total memory on the system, and there is no application of constraints by Juju, so the charm will try to configure PXC with a % of the system memory; reducing the dataset-size down to 256M or 512M might also be useful here to reduce the memory footprint of the server.

Revision history for this message
James Page (james-page) wrote :

So this will be a problem in 5.6 only - 5.5 does not pre-allocate and 5.7 does things incrementally as well. We only have 5.6 atm - so a suitable workaround might be provided by enabling control over whether the performance schema is enabled or not.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.
Subscribing...

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.