Re: [RFC][PATCH] Use ioemu block drivers through blktap

[Kevin Wolf <kwolf@suse.de>]

Ian Jackson schrieb:
> If this cleanup is really necessary, then that will cause problems for
> future runs.  If not then the cleanup isn't really necessary :-).

I think in your classification they are "not really necessary" cleanups.
But that's no reason not to do them, right?

> Leftover named pipes, for example, aren't really a big problem and can
> just be garbage collected at some point (provided their names are
> sufficiently unique).
> 
> I haven't eyeballed all of the new qemu blktap arrangements in detail.
> What processes are there and why can't the process at the other end of
> the pipe do the cleanup ?

As for uniqueness, they have the domain ID as part of their filename. If
a pipe already exists it is reused, so no real problem here, it's just
not nice to leave too many of them lying around IMHO.

Think of qemu-dm as a server providing access to disk images. If you
want the process at the other end to do the cleanup, you have the client
cleaning up server pipes. Even though this sounds a bit odd, it might
work for us in principle as the client is a single blktapctrl process
(or even xend?). One problem would be that qemu-dm could provide a pipe
which is not used by any client (if you don't use tap:ioemu block
devices) and thus won't be cleaned up.

Kevin