On 09.07.2012 12:54, Michal Privoznik wrote:
[I am pasting your patch here again so I can point out some nits]
On 06.07.2012 03:29, MATSUDA, Daiki wrote:
>
However, I will hold the push for a while. There is one general problem with this patch.
Unlike qemu monitor, guest agent has several commands which don't return any
successful message (guest-suspend-* family). with current implementation libvirt takes an
event on command as confirmation of success. That is - we block until an event is
received. But with this patch, if user will issue such command the whole API would block
endlessly. I am not sure how we should fix this. Either we will take the current
implementation (what if qemu will ever come with new command which doesn't report
successful run?) or we introduce a flag DONT_WAIT_FOR_REPLY (tricky, we want to wait for
error response - but what timeout should we choose? moreover, timeouts are bad).
Therefore, if anybody has any suggestions I am more than happy to hear them.
What about this: by default there will be some reasonable timeout (like
30 seconds) for obtaining a reply from agent. And we will have two flags:
1) DONT_WAIT - which will just write command and don't wait for any reply
2) WAIT_INF - which will block endlessly for a reply.
Hence if user knows the command should return he can set WAIT_INF flag
(okay, maybe these are badly named) and vice versa.
Thoughts?
Michal