aurora-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kai Huang <texasred2...@hotmail.com>
Subject Re: Review Request 51536: Scheduler updater will not use watch_sec if health check is enabled
Date Tue, 06 Sep 2016 22:11:24 GMT


> On Sept. 6, 2016, 8:56 p.m., Stephan Erb wrote:
> > The current default of `watch_secs` is 45 seconds. Should we drop that to 0 and
also adapt the docs accordingly? That would optimize the default settings for the health check
driven updates, which I think would be a good thing.
> 
> Maxim Khutornenko wrote:
>     If we drop it to 0 now it will break many clients currently relying on it. I think
we'll have to wait until the next release if we want to set a new default.

Thanks Stephan, we will release in the order of: scheduler -> executor -> client.


- Kai


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/51536/#review147917
-----------------------------------------------------------


On Sept. 6, 2016, 6:46 p.m., Kai Huang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/51536/
> -----------------------------------------------------------
> 
> (Updated Sept. 6, 2016, 6:46 p.m.)
> 
> 
> Review request for Aurora, Joshua Cohen, Maxim Khutornenko, and Zameer Manji.
> 
> 
> Bugs: AURORA-894
>     https://issues.apache.org/jira/browse/AURORA-894
> 
> 
> Repository: aurora
> 
> 
> Description
> -------
> 
> - Scheduler updater will not use watch_sec if health check is enabled.
> 
> This feature intends to improve reliability and performance of the Aurora scheduler job
updater by relying on health check status rather than watch_secs timeout when deciding an
individual instance update state. 
> 
> See this epic: https://issues.apache.org/jira/browse/AURORA-894 
> and the design doc: https://docs.google.com/document/d/1ZdgW8S4xMhvKW7iQUX99xZm10NXSxEWR0a-21FP5d94/edit#
for more details and background.
> 
> After discussion on Aurora dev list, we decided to keep the watch_secs infrastructure
intact on scheduler side. Our final conclusion is that we adopt the following implementation:
> 
> 1. If the users want purely health checking driven updates they can set watch_secs to
0 and enable health checks.
> 
> 2. If they want to have both health checking and time driven updates they can set watch_secs
to the time that they care about, and doing health checks at STARTING state as well.
> 
> 3. If they just want time driven updates they can disable health checking and set watch_secs
to the time that they care about.
> 
> In this review, there will be only one scheduler change: 
> Currently scheduler does not accept zero value for watch_secs, we need to relax this
constraint.
> 
> Executor change to do (in a separate review):
> The executor starts health check at STARTING, if a successful health check is performed
before initial_interval_sec expires, the executor will sends a status message for RUNNING.
> 
> 
> Diffs
> -----
> 
>   src/main/java/org/apache/aurora/scheduler/updater/UpdateFactory.java ac8df3e5a2da8cf22e1ba8a90944546e19ccdcaa

>   src/test/java/org/apache/aurora/scheduler/updater/JobUpdaterIT.java 04551f17999d742c53dfb1b36286b119b448550e

> 
> Diff: https://reviews.apache.org/r/51536/diff/
> 
> 
> Testing
> -------
> 
> ./gradlew build
> 
> ./gradlew :test --tests "org.apache.aurora.scheduler.updater.JobUpdaterIT"
> 
> ./build-support/jenkins/build.sh
> 
> 
> Thanks,
> 
> Kai Huang
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message