[CRIU] PID mismatch problem

[Cyrill Gorcunov]

On Tue, Dec 22, 2015 at 08:39:30AM +0100, Federico Reghenzani wrote:
>    Ok loaded modules are:
>    Module                  Size  Used by
>    evdev                  24576  0
>    pcspkr                 16384  0
>    coretemp               16384  0
>    xen_netfront           28672  0
>    sch_fq_codel           20480  3
>    ip_tables              28672  0
>    x_tables               28672  1 ip_tables
>    ext4 Â  Â  Â  Â  Â  Â  Â  Â  Â 507904 Â 2
>    crc16 Â  Â  Â  Â  Â  Â  Â  Â  Â 16384 Â 1 ext4
>    mbcache                20480  1 ext4
>    jbd2 Â  Â  Â  Â  Â  Â  Â  Â  Â  90112 Â 1 ext4
>    xen_blkfront           36864  4
>    It's a standard archlinux install with a recompiled kernel 4.2.5 (to
>    enable the C/R flag) on a Xen server. If you think that this platform
>    setup may be problematic, just tell me and I will try on a standard
>    non-virtual machine.

No no, all seems to be fine (there some module, iirc it was pktgen which
creates own kernel thread upon namespace creation which cause criu to
fail because we can't create namespace with pid needed, but this is
not your case).

>    I attached the tree generated by criu_restore_child() (orted-restore is
>    the software that calls CRIU APIs, and orted is the successfully restored
>    application with its child "sleep 50"). I got a 50%Â rate of successful
>    restore and 50% of pid mismatch.

Btw, maybe you would be able to start restore under the strace? Something
like "strace -f -o /log.txt -s 512 criu restore ..." and show us the /log.txt?
If run under criu service, then the service itself might be started with
strace if i'm not missing something obvious.

>    Also I have no stdout/stderr from orted, but I have the stdout/stderr of
>    its children, I'm not sure if it is orted or criu problem (before the dump
>    stdout/stderr works), I'm investigating on it.Â