mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexander Rukletsov <ruklet...@gmail.com>
Subject Re: Review Request 59746: Stopped accounting aborted container launches as failures.
Date Tue, 22 Aug 2017 13:51:10 GMT


> On June 2, 2017, 1:39 p.m., Jan Schlicht wrote:
> > src/slave/slave.cpp
> > Line 5147 (original), 5147 (patched)
> > <https://reviews.apache.org/r/59746/diff/1/?file=1740554#file1740554line5147>
> >
> >     Should we also cover the `PENDING` state of a future? `!future.isReady()` could
also mean that it's in `PENDING` state. The old code (wrongly) logged a "future discarded"
but covered that case.
> 
> Alexander Rukletsov wrote:
>     I don't think so. IIUC, this continuation is only called when a future, on which
this continuation is chained, has entered a terminal state. However, it might make sense to
`CHECK` that the future is not pending, since we consider this an internal invariant.
> 
> Jan Schlicht wrote:
>     Agreed, a `CHECK` makes sense here as we wouldn't expect a `PENDING` state at this
point.

We don't have such checks in the rest of the file (not in master.cpp). Given I'm going a different
way now, I'll punt on this.


- Alexander


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/59746/#review176766
-----------------------------------------------------------


On Aug. 22, 2017, 1:47 p.m., Alexander Rukletsov wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/59746/
> -----------------------------------------------------------
> 
> (Updated Aug. 22, 2017, 1:47 p.m.)
> 
> 
> Review request for mesos, Ian Downes, Jie Yu, Joseph Wu, and Jan Schlicht.
> 
> 
> Bugs: MESOS-7601
>     https://issues.apache.org/jira/browse/MESOS-7601
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> The container launch future might be failed or discarded (depending
> on the containerizer implementation) if the launch has been aborted,
> for example, a framework might have stopped while its task are being
> started. Such failures should not be accounted as launch errors.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.cpp 50d2a10cd68f6611efd4e691e5325e6e0c06f33a 
> 
> 
> Diff: https://reviews.apache.org/r/59746/diff/2/
> 
> 
> Testing
> -------
> 
> `make check` on several Linux distros.
> 
> Additional manual tests for (1) mesos and (1) docker containerizers. The framework is
asked to exit right after it submits the task to mesos.
> 
> (1) With mesos c-zer
> m: `./bin/mesos-master.sh --work_dir=./m`
> a: `GLOG_v=1 sudo ./bin/mesos-agent.sh --master=<ip>:5050 --work_dir=./a --containerizers=mesos
--image_providers="DOCKER" --isolation=filesystem/linux,docker/runtime`
> f: `./src/mesos-execute --master=<ip>:5050 --containerizer=mesos --docker_image=fedora:25
--name=pull-test --command="sleep 1000"`
> 
> (2) With docker c-zer
> m: `./bin/mesos-master.sh --work_dir=./m`
> a: `GLOG_v=1 sudo ./bin/mesos-agent.sh --master=<ip>:5050 --work_dir=./a --containerizers=docker`
> f: `./src/mesos-execute --master=<ip>:5050 --containerizer=docker --docker_image=fedora:25
--name=pull-test --command="sleep 1000"`
> 
> 
> Thanks,
> 
> Alexander Rukletsov
> 
>


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