Re: [PATCH] tests/avocado: Cancel BootLinux tests in case there is no free port

[Thomas Huth]

On 07/03/2022 19.48, Daniel P. Berrangé wrote:
> On Mon, Mar 07, 2022 at 07:31:50PM +0100, Thomas Huth wrote:
> > On 07/03/2022 13.50, Daniel P. Berrangé wrote:
> > > On Mon, Feb 28, 2022 at 12:43:25PM +0100, Thomas Huth wrote:
> > > > The BootLinux tests are currently failing with an ugly python
> > > > stack trace on my RHEL8 system since they cannot get a free port
> > > > (likely due to the firewall settings on my system). Let's properly
> > > > check the return value of find_free_port() instead and cancel the
> > > > test gracefully if it cannot get a free port.
> > > >
> > > > Signed-off-by: Thomas Huth <thuth@redhat.com>
> > > > ---
> > > >    Unfortunately, it still takes > 70 seconds for each and every
> > > >    tests from tests/avocado/boot_linux.py to get canceled, so
> > > >    tests/avocado/boot_linux.py still renders "make check-avocado"
> > > >    for me pretty unusable... looking at the implementation of
> > > >    find_free_port() in Avocado, I wonder whether there isn't a
> > > >    better way to get a free port number in Python? Brute-forcing
> > > >    all ports between 1024 and 65536 seems just quite cumbersome
> > > >    to me...
> > >
> > > Even in the worst case of testing every single port,
> > > for INET and INET6 and for STREAM and DGRAM sockets,
> > > that find_free_port port completes in a couple of
> > > seconds.
> >
> > Weird, on my system, the test runs for 70 seconds, just to finally
> > discovered that there was no free port available.
>
> Incidentally I'm really suprised you even hit the 'no free port'
> scenario. I've never seen that happen unless the machine was
> basically doomed due to something leaking open sockets by the
> 10's of 1000's.
>
> You mention firewall settings above, but I don't think that's
> relevant. The find_free_port() call is with no args, so it
> gets set to 'address=localhost' which means is_port_free
> takes the bind() codepath. The firewall has no interaction
> with the bind() codepath in the kernel AFAIK.

Yes, I've now had another closer look, and indeed, the firewall is not the 
problem here - but is_port_free() seems to be very buggy. The problem is 
that I do not have any IPv6 address configured on my system, and in that 
case the current code fails.
See https://github.com/avocado-framework/avocado/issues/5273 ... I need this 
patch to fix it:

diff --git a/avocado/utils/network/ports.py b/avocado/utils/network/ports.py
--- a/avocado/utils/network/ports.py
+++ b/avocado/utils/network/ports.py
@@ -60,7 +60,7 @@ def is_port_free(port, address):
                     if localhost:
                         return False
                 sock.close()
-        return True
+                return True
     finally:
         if sock is not None:
             sock.close()

 Thomas