Description Philipp Hahn 2014-02-17 14:27:52 CET

Ticket#: 2014021721004459

I a large Xen environment with 18 VMs running on a single host, xenstored generated so much load, that "virsh list" took ages and "virsh dumpxml $DOM" didn't return for several minutes.

This is caused by
xen/tools/python/xen/xend/XendDomainInfo.py # _update_consoles()
1638                     # The value is unchanged. Just for xenstore watcher
1639                     console_ctrl.writeBackend(0, 'uuid', console_uuid)
which writes back the DB for each VM looked up. Each lookup took roughly 1 seconds, for a total of 15 seconds for all 18 VMs. As UVMMd is querying all VMs every 30 seconds, this seems to get in some kind of loop.
("virsh dumpxml $DOM" also iterates over all VMs, even wen $DOM matches the name/UUID/ID.)

Putting the tdb on a tempfs solved the problem. We should do something like the following for Xen systems:
  echo "xenstored /var/lib/xenstored tmpfs rw,mode=755 0 0" >>/etc/fstab

+++ This bug was initially created as a clone of Bug #28309 +++
The original bug only added zapping the /var/lib/xenstored/tdb file for reboots.
In newer Debian the file is in /var/run/=/run/=tmpfs by default.