On Sun, Nov 7, 2021 at 4:26 PM Richard W.M. Jones <rjones(a)redhat.com> wrote:
On Sun, Nov 07, 2021 at 04:22:26PM +0200, Nir Soffer wrote:
> On Sun, Nov 7, 2021 at 4:00 PM Richard W.M. Jones <rjones(a)redhat.com> wrote:
> >
> > On Sun, Nov 07, 2021 at 03:27:47PM +0200, Nir Soffer wrote:
> > > When using qemu-nbd with systemd socket activation, trying to connect
> > > additional clients concurrently fails with:
> > >
> > > nbd_connect_uri: connect: server backlog overflowed, see
> > >
https://bugzilla.redhat.com/1925045: Resource temporarily
> > > unavailable
> > >
> > > But this bug was fixed in qemu-nbd long time ago; turns out that libnbd
> > > has the same bug, creating the server socket with:
> > >
> > > listen(s, 1);
> > >
> > > Use SOMAXCONN so we can have multiple connections for better
> > > performance. I think this can be very useful for nbdcopy.
> >
> > Apart from the problems in API mentioned in the other review, this
> > patch ordering isn't bisectable.
> >
> > What's the motivation for this patch series?
>
> Why is it not bissectable?
Doesn't the test fail after the first patch is applied and
before the second one?
No, it requires using concurrent connections to fail. I have an application
using this that fails reliably without the fix.