mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chun-Hung Hsiao <chhs...@apache.org>
Subject Re: Review Request 70084: Avoid dereferencing removed executors and launching containers for them.
Date Fri, 15 Mar 2019 19:37:57 GMT


> On March 15, 2019, 2:40 a.m., Meng Zhu wrote:
> > src/slave/slave.cpp
> > Lines 2991 (patched)
> > <https://reviews.apache.org/r/70084/diff/1/?file=2128192#file2128192line2991>
> >
> >     We have an implicit constructor from repeated resource. Can we get rid of the
explicit construction here?

Unfortunately no since both `executorInfo.resources()` and `task->resources()` are both
`RepeatedPtrField<Resource>`, no `operator+` is defined for it. Dropping.


> On March 15, 2019, 2:40 a.m., Meng Zhu wrote:
> > src/slave/slave.cpp
> > Lines 2998 (patched)
> > <https://reviews.apache.org/r/70084/diff/1/?file=2128192#file2128192line2998>
> >
> >     lets use the new `*` operator:
> >     `... = *(flags.default_container_info);`

It seems that we don't need the parentheses ;) Will do.

BTW is `operator*` backported to 1.7?


> On March 15, 2019, 2:40 a.m., Meng Zhu wrote:
> > src/slave/slave.cpp
> > Lines 3001 (patched)
> > <https://reviews.apache.org/r/70084/diff/1/?file=2128192#file2128192line3001>
> >
> >     `publishResources` are put ahead of the other calls.
> >     I guess the simplicity outweights the performance concern of wasted work.
> >     
> >     As one with limited knowledge of this part of the code, I was wondering if we
need any cleanup of the `publishResources` in case something goes ary in between. Not in this
patch, but adding some comments regarding the cleanup and idempotency of the call would be
great.

There are comments in `publishResources` but I'll copy them over here and other callsites
as well.


> On March 15, 2019, 2:40 a.m., Meng Zhu wrote:
> > src/slave/slave.cpp
> > Line 2978 (original), 3014 (patched)
> > <https://reviews.apache.org/r/70084/diff/1/?file=2128192#file2128192line3014>
> >
> >     I see that this is equivalent to the original logic, but brevity makes it harder
to grasp. A comment regarding taskGroup would be appreaciated.

Can you elaborated more on what should be clarified?


- Chun-Hung


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70084/#review213740
-----------------------------------------------------------


On March 2, 2019, 1:04 a.m., Chun-Hung Hsiao wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70084/
> -----------------------------------------------------------
> 
> (Updated March 2, 2019, 1:04 a.m.)
> 
> 
> Review request for mesos, Gilbert Song, Greg Mann, Jie Yu, Meng Zhu, and Vinod Kone.
> 
> 
> Bugs: MESOS-8467
>     https://issues.apache.org/jira/browse/MESOS-8467
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> When launching executors and tasks, there is no guarantee that the
> executors still remain after `Slave::publishResources` is returned. If
> not, the executor struct should not be dereferenced and the executor
> containers should not be launched at all.
> 
> NOTE: The patch makes `Slave::launchExecutor` called asynchronously even
> if there is no secret generator. However this should not affect the
> correctness of executor launching.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.hpp 7ad495504e4ff144ac31812fbd4a3a1f4da86f02 
>   src/slave/slave.cpp e3c2c005d865b5c333e92e50e49ef398fe06ad79 
> 
> 
> Diff: https://reviews.apache.org/r/70084/diff/1/
> 
> 
> Testing
> -------
> 
> sudo make check
> 
> 
> Thanks,
> 
> Chun-Hung Hsiao
> 
>


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