[RADIATOR] AuthHEIMDALDIGEST and defunct kdigest processes

[Johan Wassberg]

> On 12 Feb 2019, at 15:41, Heikki Vatiainen <hvn at open.com.au> wrote:
> 
> Added Cc: so that others here are kept up-to-date.

Sorry, my intention was not to forget answer directly to you and forgot to add
the list as cc.

> 
> On 08/02/2019 17.41, Johan Wassberg wrote:
> 
>> Not sure what that says about our environment, but I can reproduce this
>> in our test environment with no load at all, one client trying to auth.
>> Might need to get some better hardware…
> 
> Here all components were running on the same machine which may be the main cause of timing difference.
> 
>> So I enabled `LogMicroseconds` and added a logline after my earlier
>> modifications (Right after the if sentence for the result of `waitpid`).
>> Then I run two tests with different flags for `waitpid`, `WNOHANG` and `0`.
> 
>> Fri Feb  8 11:58:07 2019 341772: DEBUG: AuthHEIMDALDIGEST digest command output: tickets=no
>> Fri Feb  8 11:58:07 2019 341960: ERR: AuthHEIMDALDIGEST kdigest_challenge waitpid missed child (10162), waitpid returned 0
>> Fri Feb  8 11:58:07 2019 342017: ERR: AuthHEIMDALDIGEST kdigest_challenge waitpid completed
>> Fri Feb  8 11:58:07 2019 342091: DEBUG: Radius::AuthHEIMDALDIGEST REJECT: AuthBy HEIMDALDIGEST Password check failed: jocar [jocar]
> 
> 
>> With `0`:
> 
>> Fri Feb  8 11:55:52 2019 420102: DEBUG: AuthHEIMDALDIGEST digest command output: tickets=no
>> Fri Feb  8 11:55:52 2019 420300: ERR: AuthHEIMDALDIGEST kdigest_challenge waitpid completed
>> Fri Feb  8 11:55:52 2019 420382: DEBUG: Radius::AuthHEIMDALDIGEST REJECT: AuthBy HEIMDALDIGEST Password check failed: jocar [jocar]
> 
>> Is the time difference small enough you think?
> 
> I'd say it is. When the total wait time is considered, waitpid for 1-2 ms does not affect much. And also, it's called when we know the called program is exiting which should mean it's not going to be wait for very long.
> 
> I think the way to go is waitpid with 0.
> 
> Thanks for taking a look at this!

Cool!
When do you think this can be included in any package? In the next version or
patches? (Not sure what qualifies as a patch or as a fix for later version…)

-- 
jocar