Sounds like: https://issues.apache.org/jira/browse/CLOUDSTACK-5986
It's introduced by new dhcp_release mechanism, and I've fixed it in the related branch, including 4.2 and 4.3. There is a workaround can be used by manipulating edithosts.sh in the VR, by setting no_dhcp_release to 1 always. --Sheng On Tue, Feb 18, 2014 at 8:49 AM, Simon Waterhouse < [email protected]> wrote: > Hi > I am having a few problems with DHCP leases from Virtual Router > (CloudStack 4.2 / Basic Zone / XenServer) and wondered if this is a known > problem. > > Under normal circumstances everything is working fine, but occasionally a > running instance does not get its assigned IP address via DHCP: tcpdump > tells me the new instance is issuing a DHCPDISCOVER and getting no > response. Looking at the Dnsmasq logs on the VR it seems that the DHCP > server is declining to respond to the DHCPDISCOVER because the configured > IP address is leased to another MAC (this MAC is associated with the NIC of > a long-since-dead instance). Flushing the cache of existing leases in > Dnsmasq by restarting the service seems to fix the problem (for some > definition of "fix"). > > As Dnsmasq appears to be configured in /etc/dhcphosts.txt to issue > infinite leases can someone tell me how leases get cleaned up if there is > an abnormal termination of an instance such that it cannot send a > DHCPRELEASE on shutdown? > > Thanks > Simon >
