On 04/27/2012 07:22 AM, Michal Privoznik wrote:
This patch set tries to fix corner cases where libvirt runs on huge
system, e.g. 4K CPU monster. In these cases, capabilities XML is
enormously big, as we are transferring info about each singe CPU core
(to which NUMA node it belongs, etc.). This XML is bigger than our
RPC limit, therefore users cannot get it as it is dropped on server,
leaving them with inability to connect. Therefore we need to increase
those limits (whole RPC message and RPC string). However, simple
lifting up will work, but increase mem usage.
Therefore I've reworked RPC buffer handling: changed it from
'statically' to dynamically allocated.
So in most cases - when small messages are sent - this will even
decrease our memory consumption. Leaving us flexible for corner cases
described above.
On the other hand, I realize we've had our history with RPC breakage.
So I think I'll require more than 1 ACK before pushing.
What's your opinion?
I think it's worth making a change like this, but holding off until
after the 0.9.12 release to make sure we have time to hammer on it and
address any unintended fallout. I still haven't closely reviewed the
patches.
--
Eric Blake eblake(a)redhat.com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org