[CRIU] [PATCH] zdtm: add a dummy action-script

Andrey Wagin avagin at gmail.com
Thu Aug 7 13:58:21 PDT 2014


2014-08-07 21:12 GMT+04:00 Christopher Covington <cov at codeaurora.org>:
> On 08/07/2014 11:02 AM, Andrew Vagin wrote:
>> On Thu, Aug 07, 2014 at 05:14:40PM +0400, Pavel Emelyanov wrote:
>>> On 08/07/2014 04:16 PM, Christopher Covington wrote:
>>>> On 08/07/2014 03:06 AM, Pavel Emelyanov wrote:
>>>>> On 08/06/2014 05:05 PM, Andrey Vagin wrote:
>>>>>> Just to be sure that it affects nothing
>>>>>
>>>>> This would affect the full tests run time significantly.
>>>>
>>>> What if it went in at the same time as an optimization so that overall runtime
>>>> stayed constant or dropped?
>>>
>>> Hardly. The sigaction restore saves us 64 syscalls per-task for tests that have
>>> more than one of them (quite a few), while each --action-script results in 3-4
>>> system() calls performed :)
>>>
>>>> Speaking of test run-time, it appears to me that zdtm.sh is currently pretty
>>>> single threaded. Adding parallelization to the test harness may provide a
>>>> useful speedup on quad code and greater systems (maybe the kind of speed up
>>>> that could hide the slowdown of always running a script).
>>>
>>> Andrew knows some way of launching ns/ tests in parallel. Andrey?
>>
>> make -C test -j 4 zdtm_ns
>
> Why does this only work for namespace tests?

A processe is restored with the same pid. So if we execute tests
concurrently, one test can reuse a pid of another one, while it has been
dumped, but it isn't restored yet.

We can execute zdtm.sh for each test in a separate pidns. Look at the
attached patch.

Thanks,
Andrew

> Thanks,
> Christopher
>
> --
> Employee of Qualcomm Innovation Center, Inc.
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> hosted by the Linux Foundation.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch
Type: application/octet-stream
Size: 880 bytes
Desc: not available
URL: <http://lists.openvz.org/pipermail/criu/attachments/20140808/df79ddd8/attachment.obj>


More information about the CRIU mailing list