> On Feb. 28, 2014, 10:37 a.m., Jim Youll wrote:
> > would it be very hard to make that a command line / config file var rather 
> > than an env var? Then it would be set just like all other configurable 
> > params in a config file or with a command line override, and would be more 
> > sustainable.

I don't think client-side health checking is long for this world and 
executor-side health checking already supports this via 
HealthCheckConfig.timeout_secs. The way the code is currently factored makes 
adding a command-line flag somewhat messy.

That said you can add this to your .aurora config with

import os
os.putenv('AURORA_HTTP_SIGNALER_TIMEOUT_SECS', '2.0')

And it will be respected. I'd recommend this over forcing the invoker to 
remember to set an environment variable before invoking 'aurora'.


- Kevin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/18602/#review35810
-----------------------------------------------------------


On Feb. 27, 2014, 6:37 p.m., Kevin Sweeney wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/18602/
> -----------------------------------------------------------
> 
> (Updated Feb. 27, 2014, 6:37 p.m.)
> 
> 
> Review request for Aurora, Mark Chu-Carroll, Maxim Khutornenko, and Brian 
> Wickman.
> 
> 
> Bugs: AURORA-224
>     https://issues.apache.org/jira/browse/AURORA-224
> 
> 
> Repository: aurora
> 
> 
> Description
> -------
> 
> Make http signaler timeout secs configurable via an environment variable. 
> This is a band-aid for the issue described in AURORA-224. Further 
> improvements would be made possible by refactoring the updater to completely 
> remove client-side health checking (delegating to thermos, which already 
> supports a much richer system of delegated health checks).
> 
> I'm open to suggestions on how to make this feature more Aurora-y - I know 
> environment variables have typically been discouraged since they can lead to 
> flaky deploys, but I don't think this setting adds unnecessary risk and I see 
> it as a short-term fix for some counter-intuitive behavior on our part.
> 
> 
> Diffs
> -----
> 
>   src/main/python/apache/aurora/common/http_signaler.py 
> b19b72d6400cef1a2e469b1853a7d49c063d5a25 
> 
> Diff: https://reviews.apache.org/r/18602/diff/
> 
> 
> Testing
> -------
> 
> ./build-support/jenkins/build.sh
> 
> 
> Thanks,
> 
> Kevin Sweeney
> 
>

Reply via email to