[CRIU] zdtm with proxy-cache

Andrew Vagin avagin at virtuozzo.com
Wed Feb 10 16:40:12 PST 2016


On Mon, Feb 08, 2016 at 01:56:27AM +0000, Rodrigo Bruno wrote:
> Hi,
> 
> I am testing CRIU with proxy-cache migration (using sockets) using zdtm.py script
> but I ran into two problems.
> 
> 1- I start both the proxy and cache with --pidfile and their pids gets written 
> into specific files. However, the pids of both processes (proxy and cache) change
> after getting written and therefore, from zdtm.py, I cannot kill them reading the
> pidfile. 
> 
> I looked into the page-server and it seems to quit before the zdtm.py script 
> attempts to kill it. If it does not quit, zdtm.py will fail to kill it as well
> (because the pid inside the pid file is no longer the real one).
> 
> Do you know why this happens, why the real pid does not match the pid inside
> the pidfile?

Maybe you try to kill a process from another pidns? zdtm_ct creates a
new pidns.
> 
> 2- I cannot run tests using namespaces. I think the problem is related to a 
> unix socket that is used to communicate between the cache and CRIU Restore.
> As soon as the restore operation attempts to mount the namespaces, new 
> connections to the unix socket no longer succeed. 

It may be due to network namespaces. You need to connect to a cache
before forking the root task and then always use this socket, because
the root task will be executed in another set of namespaces.

Thanks,
Andrew

> 
> Do you have any idea of a potential reason for this? Where should I look at?
> 
> Thanks,
> Rodrigo Bruno
> 
> -- 
> Rodrigo Bruno <rbruno at gsd.inesc-id.pt>
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu


More information about the CRIU mailing list