On 10/29/2015 12:15 PM, Laine Stump wrote:
On my system, these are set to 1, 1, and 1000, and I've found that DAD
takes something between 5.7 and 6.8 seconds to complete (it was at the
lower end with a single IP address, and at the high end with 70 IP
addresses (or also with 20 IPs, so I don't think it's going to get
substantially higher). (Too bad I didn't do this *before* I pushed,
rather than relying on a successful build and reports of proper
operation on your system :-/)
Based on that, I think it makes sense to push a patch that sets the
timeout to 20 seconds (and push Luyao's patch ragardless). Since the
timeout is meant to catch an "infinite" wait, I think 20 seconds is
okay; I don't want to add yet another tunable parameter unless we really
need to.
I'm pushing Luyao's patch now, but will wait for an ACK to push the
patch I've attached here. Anyone?
Looks reasonable to me, and safe for inclusion in the current release. ACK
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org