[Devel] [Q] migrating inotify assigned on ploop device

Igor igor at parallels.com
Tue Jan 12 01:26:09 PST 2016


On 01/12/2016 11:57 AM, Cyrill Gorcunov wrote:
> On Tue, Jan 12, 2016 at 12:20:34AM +0300, Cyrill Gorcunov wrote:
>> Hi guys, today I've been investigating a problem of container's migration
>> with inotify assigned on /dev/ploop10308p1 device. The main problem is that
>> the mark is set to devtmpfs where device ID and inode number is not persistent
>> by default so to workaraound such problem we keep a real "path" to the watchee
>> inside image which eventually get used on restore procedure, but because of
>> ploop nature at the moment of restore the former /dev/ploop10308p1 is transformed
>> into the new device /dev/ploop46339p1 so as expected the restore procedure fails.
>>
>> Been talking to Pavel and the proper general solutions is "device namespace" which
>> require significant kernel update/rework so I don't consider it as a solution
>> which could be implemented in a day or two. Instead another option might be
>> to use CRIT tool and decode|encode device path on the fly via libvzctl or
>> p.haul project.
>>
>> Any other ideas comments?
> After another talk with Pavel it seems the only reasonable solution is
> to recode ploop names in all images when migrating with help of
> crit tool and p.haul? Nikita, could you please take a look and
> figure out if this may be implemented?
This is not only migration problem but suspend/resume in general, on CT 
resume ploop decide is autogenerated.




More information about the Devel mailing list