Univention Bugzilla – Bug 36605
libvirt-check.sh: libvirt/check/timeout=5s too low
Last modified: 2015-01-29 11:41:03 CET
+++ This bug was initially created as a clone of Bug #33966 +++ The UCRV libvirt/check/timeout=5 is much too low as an "virsh list" on a Xen host with 12 VMs took 5s. This leads to libvirtd being killed needlessly, which breaks the connection to UVMM. As that happened each time, UVMM takes minutes to recover a working connection to the host. As "virsh list" depends on the number of VMs running on the cost, please also consider switching to another command which is not O(n) but O(1), like "virsh nodeinfo" (see "virsh help host" for a list of possible commands). But on the affected host this also takes ~5s: # time virsh nodeinfo >/dev/null real 0m4.980s user 0m0.212s sys 0m0.060s # time virsh capabilities >/dev/null real 0m6.885s user 0m0.196s sys 0m0.080s
Please also check a backport to UCS 3.2.
r57187 | Bug #36605 libvirt: Copyright 2015 r57188 | Bug #36605 libvirt: Increase timeout in check script Package: univention-virtual-machine-manager-node Version: 3.0.4-1.87.201501081418 Branch: ucs_4.0-0 Scope: errata4.0-0 2014-12-08-univention-virtual-machine-manager-node.yaml Increase timeout for libvirtd check to 30 seconds. Still using "list" as this is the most portable one.
OK: New timeout 30s OK: On update with unchanged old default: set to timeout to 30s OK: YAML
<http://errata.univention.de/ucs/4.0/54.html>