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 62212: Send TASK_STARTING from the built-in executors. [1/2]
Date Tue, 12 Sep 2017 11:11:41 GMT


> On Sept. 12, 2017, 11:03 a.m., Andrei Budnik wrote:
> > src/docker/executor.cpp
> > Lines 141 (patched)
> > <https://reviews.apache.org/r/62212/diff/1/?file=1819447#file1819447line141>
> >
> >     Do we really need to send `TASK_STARTING`, if we know that right after sending
`TASK_STARTING`, `launchTask` might send `TASK_FAILED`, e.g. in case of following checks:
> >     ```c++
> >     if (run.isSome()) {
> >      // ...
> >      status.set_state(TASK_FAILED);
> >     }
> >     ...
> >     if (runOptions.isError()) {
> >      // ...
> >      status.set_state(TASK_FAILED);
> >     }
> >     ```
> >     What is the semantics of `TASK_STARTING` - should an executor send it always
and unconditionally, or can we omit `TASK_STARTING` when we are pretty sure that we won't
be able to launch a task?
> >     Omitting `TASK_STARTING` before sending `TASK_FAILED` might be a good optimization,
if it's not important to send `TASK_STARTING` first.

Agree with Andrei: we should send task starting if we try to start the task. If we know that
starting will never succeed, we don't even try, hence no task starting update : )


> On Sept. 12, 2017, 11:03 a.m., Andrei Budnik wrote:
> > src/launcher/executor.cpp
> > Lines 512 (patched)
> > <https://reviews.apache.org/r/62212/diff/1/?file=1819449#file1819449line512>
> >
> >     What happens if a buggy scheduler will try to launch multiple tasks with same
`task_id` using a single `command` executor? I guess, `command` executor will send `TASK_STARTING`,
then `TASK_FAILED`. If so, then is it ok to have multiple `TASK_STARTING` status updates (having
different timestamps)?

Multiple tasks for a single command executor are not allowed. We should send task starting
only once, for the task that we actually try to start.


- Alexander


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/62212/#review185156
-----------------------------------------------------------


On Sept. 11, 2017, 9:16 a.m., Benno Evers wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/62212/
> -----------------------------------------------------------
> 
> (Updated Sept. 11, 2017, 9:16 a.m.)
> 
> 
> Review request for mesos, Andrei Budnik and Alexander Rukletsov.
> 
> 
> Bugs: MESOS-7941
>     https://issues.apache.org/jira/browse/MESOS-7941
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This gives schedulers more information about a tasks status,
> in particular it gives a better estimate of a tasks start time
> and helps differentiating between tasks stuck in TASK_STAGING
> and tasks stuck in TASK_STARTING.
> 
> 
> Diffs
> -----
> 
>   docs/high-availability-framework-guide.md 73743aba31f9d0ca827d318e2ecb4752a91b1be0

>   src/docker/executor.cpp e9949f652cd8527991ebfdfbf14e68b4c958fe79 
>   src/launcher/default_executor.cpp 106b7f2e0244d211c66b237b5d1c51f43fc6e529 
>   src/launcher/executor.cpp 951597b576b4912541dd87d52dcb981393e58082 
> 
> 
> Diff: https://reviews.apache.org/r/62212/diff/1/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Benno Evers
> 
>


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