Re: [PATCH] isar: Clean mount point on bitbake exit

[Alexander Smirnov <asmirnov@ilbers.de>]

On 02/06/2018 11:56 PM, Jan Kiszka wrote:
> On 2018-02-06 21:45, Alexander Smirnov wrote:
>> On 02/06/2018 11:31 PM, Jan Kiszka wrote:
>>> On 2018-02-06 20:55, Alexander Smirnov wrote:
>>>> 8<--
>>>>
>>>> That's it! Branch 'asmirnov/devel', please test and enjoy :-)
>>>>
>>>> 8<--
>>>>
>>>> Now each multiconfig has registered handler for BuildCompleted event
>>>> (see
>>>> class 'isar-event.bbclass'). Moreover, the '/proc/mounts' file contains
>>>> all the active mounts. In addition, from event handler we could derive
>>>> all the variables like ${TMPDIR}, ${DISTRO} etc. So it's possible to
>>>> find all the active mounts for current multiconfig and clean them.
>>>>
>>>> NOTE: if build is interrupted by double ^C, some mount points could stay
>>>> uncleaned. This is caused by remaining processes started by bitbake, for
>>>> example:
>>>>    - 'chroot build.sh ...'
>>>>    - 'multistrap ...'
>>>
>>> Can you explain what the race condition is exactly? It seems to work for
>>> me so far, no forgotten mounts.
>>
>> Sometimes after double ^C I see several mounted artifacts in
>> buildchroot. But I've also noticed, that there are running processes
>> like 'multistrap' and 'build.sh' running *exactly* in this buildchroot.
>>
>> So, simple test, if you press double ^C and run 'ps ax' after, you will
>> likely see running processes. IIRC the same bitbake behavior I saw with
>> Yocto, interrupting bitbake doesn't mean to kill the process with
>> current task.
> 
> Yes, there is a settling phase, but everything is dead and gone after a
> couple of seconds, at least here.

My multistrap process stays for a few minutes after, but it happens once 
per about 10 times. Probably you are lucky and interrupt the build in 
suitable place. For sure I press double ^C quickly without delay in between.

In general I'm not sure if we can handle this, it sounds like bitbake 
should do this. The correct way to stop the build is to press single ^C, 
this should work without any issues, but you have to wait until current 
task is finished.

Anyway please let me know if the patch is ok and solves your issue with 
'/dev/null', so I can apply it.

Alex