[CRIU] zdtm and --check-only with zdtm/static/socket-tcpbuf

Pavel Emelyanov xemul at virtuozzo.com
Mon Nov 14 05:20:38 PST 2016


On 11/10/2016 03:03 PM, Adrian Reber wrote:
> I am currently going through the test cases which are failing with
> '--check-only'. zdtm/static/socket-tcpbuf is one of the tests which
> fails. If I understand it correctly the following happens.
> 
> zdtm dumps the test cases with --check-only which lets the processes
> running. To end the process after check-only dumping I am using:
> 
> pstree_signal(self.__test.getpid(), signal.SIGKILL)
> 
> Which kills both instances of socket-tcpbuf.
> 
> Without check-only the following happens:
> 
> 1. zdtm starts 2 instances of socket-tcpbuf
> 2. zdtm dumps and ends one instance of socket-tcpbuf
> 3. zdtm restores the second instance of socket-tcpbuf
> 4. all good
> 
> With check-only following happens:
> 
> 1. zdtm starts 2 instances of socket-tcpbuf
> 2. zdtm dumps one instance of socket-tcpbuf and it keeps on running

With the --check-only, right?

> 3. using pstree_signal(self.__test.getpid(), signal.SIGKILL) both
>    instances of socket-tcpbuf get killed
> 3. using os.kill(int(self.__test.getpid()), signal.SIGKILL) both
>    instances of socket-tcpbuf also get killed
> 4. restore does not work as one process for communication is missing
> 
> So, my question is, what is the right approach to run socket-tcpbuf in
> check-only mode? Does that test have to be changed or should I just
> exclude it from check-only mode? Is there a way to exclude a test for
> certain use cases?

I would make the --check-only testing to work in pair with regular dump-restore
test. Like this -- before doing dump you do dump --check-only + restore --check-only
and if it works, then proceed with regular dump (and restore). Would it work?

> 		Adrian
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu
> .
> 



More information about the CRIU mailing list