Bug#851876: Bug#855193: RFS: slt/0.0.git20140301-2.1 [RC][NMU] -- TLS reverse-proxy with SNI multiplexing (TLS virtual hosts)

[Daniel Kahn Gillmor]

On Wed 2017-02-15 12:09:54 -0500, Roger Shimizu wrote:
> On Thu, Feb 16, 2017 at 1:36 AM, James Cowgill  wrote:
>> On 15/02/17 11:24, Roger Shimizu wrote:
>>> On Wed, Feb 15, 2017 at 7:51 PM, James Cowgill  wrote:

 So do you know why the tests only pass when using 2 CPUs? That seems
 pretty fishy to me. Maybe there is an underlying bug here?
>>>
>>> I'm not sure the reason.
>>> I was just trying to help on the FTBFS RC bug during BSP Tokyo.
>>
>> I found the actual cause - there's a race condition in the testsuite
>> which will usually (100% in practice) cause it to deadlock on single
>> processor systems. You can look at the bug I filed upstream if you want.
>
> Thanks for catching the race condition bug!
>
> But can we do something to prevent the autoremove of the package soon?
> I guess the testcase issue can be lower to non-RC level then.

Thanks to both James for the investigation and to Roger for the proposed
workaround.

I've just pushed 0.0.git20140301-3 to the archive with Roger's
workaround, and noted James' diagnosis and upstream bug report.

Regards,

 --dkg


signature.asc
Description: PGP signature

Bug#851876: Bug#855193: RFS: slt/0.0.git20140301-2.1 [RC][NMU] -- TLS reverse-proxy with SNI multiplexing (TLS virtual hosts)

[Roger Shimizu]

On Thu, Feb 16, 2017 at 1:36 AM, James Cowgill  wrote:
>
> On 15/02/17 11:24, Roger Shimizu wrote:
>> Dear James,
>>
>> On Wed, Feb 15, 2017 at 7:51 PM, James Cowgill  wrote:
>>>
>>> So do you know why the tests only pass when using 2 CPUs? That seems
>>> pretty fishy to me. Maybe there is an underlying bug here?
>>
>> I'm not sure the reason.
>> I was just trying to help on the FTBFS RC bug during BSP Tokyo.
>
> I found the actual cause - there's a race condition in the testsuite
> which will usually (100% in practice) cause it to deadlock on single
> processor systems. You can look at the bug I filed upstream if you want.

Thanks for catching the race condition bug!

But can we do something to prevent the autoremove of the package soon?
I guess the testcase issue can be lower to non-RC level then.

Cheers,
-- 
Roger Shimizu, GMT +9 Tokyo
PGP/GPG: 4096R/6C6ACD6417B3ACB1

Bug#851876: Bug#855193: RFS: slt/0.0.git20140301-2.1 [RC][NMU] -- TLS reverse-proxy with SNI multiplexing (TLS virtual hosts)

[James Cowgill]

retitle 851876 slt: FTBFS on uniprocessor systems - testsuite deadlocks
forwarded 851876 https://github.com/inconshreveable/slt/issues/5
thanks

Hi,

On 15/02/17 11:24, Roger Shimizu wrote:
> Dear James,
> 
> On Wed, Feb 15, 2017 at 7:51 PM, James Cowgill  wrote:
>>
>> So do you know why the tests only pass when using 2 CPUs? That seems
>> pretty fishy to me. Maybe there is an underlying bug here?
> 
> I'm not sure the reason.
> I was just trying to help on the FTBFS RC bug during BSP Tokyo.

I found the actual cause - there's a race condition in the testsuite
which will usually (100% in practice) cause it to deadlock on single
processor systems. You can look at the bug I filed upstream if you want.

> What I know is this is not rare, like another one I fixed:
> https://bugs.debian.org/854500

This is at best a big hack, since it almost certainly hides a real bug,
but I guess in this situation it does kind of work since the bug is in
the testsuite.

> Could you kindly help to upload?
> Buildd [0] shows it actually build fine for all ARCHs I should build,
> so I don't what to trigger the autoremove ...

James



signature.asc
Description: OpenPGP digital signature