[CRIU] Dumping a restored apache process

[Pavel Emelyanov]

On 04/17/2014 12:12 PM, Cyrill Gorcunov wrote:
> On Thu, Apr 17, 2014 at 12:07:45PM +0400, Pavel Emelyanov wrote:
>> On 04/16/2014 09:24 PM, Cyrill Gorcunov wrote:
>>> On Wed, Apr 16, 2014 at 07:48:07PM +0300, Deyan Doychev wrote:
>>>> On 04/16/2014 07:01 PM, Cyrill Gorcunov wrote:
>>>>> On Wed, Apr 16, 2014 at 06:18:01PM +0300, Deyan Doychev wrote:
>>>>>> 3. criu restore --tcp-established --file-locks --restore-detached
>>>>>>     It successfully restores.
>>>>>>
>>>>>> 4. criu dump --tcp-established --file-locks -t 9499 -v4
>>>>>>     Here I get the error described above.
>>>>> Could you please provide complete dump log for step 4)
>>>>
>>>> Here is the complete dump log.
>>>>
>>>> This time I hit the error on the third dump.
>>>>
>>>
>>> Thanks! Indeed for some reason the file descriptor is treated as
>>> different one keeping the same socket
>>
>> Since it's the after-restore problem, maybe we _restore_ these files
>> as two different ones?
> 
> I can't imaging how it's possible. Deyan, could you please save
> complete dump/restore images somewhere (ie every c/r cycle to
> write into own directory with help of -D option). Then, once
> it fails -- send me them privately (or put somewhere where
> I could fetche them from if this would be more convenient). I
> would like to see what happened in images...

Another idea is that we we have this ID in fdid cache before we
enter the fd_id_generate for the 2nd time. But this should be only
possible when we put this ID into cache via fd_id_generate_special.

Can we get dump logs with Cyrill's additional logging on the 1st
dump to see how the IDs are generated in normal situation?

Thanks,
Pavel