mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vinod Kone <vinodk...@apache.org>
Subject Re: Review Request 68866: Waited for TEARDOWN response in v1 Java scheduler shim.
Date Wed, 03 Oct 2018 15:48:10 GMT


> On Sept. 27, 2018, 6:23 p.m., Vinod Kone wrote:
> > src/java/jni/org_apache_mesos_v1_scheduler_V1Mesos.cpp
> > Lines 294 (patched)
> > <https://reviews.apache.org/r/68866/diff/1/?file=2092453#file2092453line294>
> >
> >     Looks like you are doing `call` here instead of `send` so that you have a future
to wait on? Skipping `send` which does validation and authentication seems wrong. Let's not
do that.
> >     
> >     I would recommend putting a sleep in the client code instead of here for now.
> 
> Alexander Rukletsov wrote:
>     But `call` does [validation](https://github.com/apache/mesos/blob/6e21e94ddca5b776d44636fe3eba8500bf88dc25/src/scheduler/scheduler.cpp#L269-L270)
and [authentication](https://github.com/apache/mesos/blob/6e21e94ddca5b776d44636fe3eba8500bf88dc25/src/scheduler/scheduler.cpp#L301)
too. Not sure I follow, Vinod.
> 
> Greg Mann wrote:
>     Yea I think we have the necessary validation/authentication in the `call()` path
as well. This solution is nice because it will wait for as little time as possible, up to
5 seconds, rather than always waiting for 5 seconds.
>     
>     In fact, since this code will return early if a response is actually received, I
would feel comfortable with a slightly longer await - maybe 10 or 20 seconds?

I didn't realize `call` was a newish interface method added to the library. At a quick glance
I thought it was the continuation of `send` and an internal method. My bad.

So I'm ok with doing a `call` here.

Suggestion: The comment above doesn't have enough context for a future to understand why the
library could be destructured. I would recommend expanding the comment a bit more explaining.
For example "The JVM could potentially call JNI `finalize` if the Java scheduler nullified
its reference to `V1Mesos` library immediately after sending TEARDOWN." Also, maybe link to
MESOS-9274 for posterity.


- Vinod


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68866/#review209074
-----------------------------------------------------------


On Sept. 27, 2018, 5:41 p.m., Alexander Rukletsov wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68866/
> -----------------------------------------------------------
> 
> (Updated Sept. 27, 2018, 5:41 p.m.)
> 
> 
> Review request for mesos, Gastón Kleiman, Greg Mann, and Nicholas Parker.
> 
> 
> Bugs: MESOS-9274
>     https://issues.apache.org/jira/browse/MESOS-9274
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> See summary.
> 
> 
> Diffs
> -----
> 
>   src/java/jni/org_apache_mesos_v1_scheduler_V1Mesos.cpp 2a5fbd79ac7bad933067cd96e38186849af8edc4

> 
> 
> Diff: https://reviews.apache.org/r/68866/diff/1/
> 
> 
> Testing
> -------
> 
> `make check`on various Linux distro
> 
> 
> Thanks,
> 
> Alexander Rukletsov
> 
>


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