mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benjamin Mahler <bmah...@apache.org>
Subject Re: Review Request 64626: Prototype: Added LIBPROCESS_LOCAL_MESSAGE_PASSING option.
Date Thu, 14 Dec 2017 22:48:44 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/64626/#review193855
-----------------------------------------------------------



Hm.. perhaps we need a pattern based approach? From the existing benchmarks, we usually want
specific messages only to go through http, for example messages between master <-> scheduler.

- Benjamin Mahler


On Dec. 14, 2017, 10:33 p.m., Ilya Pronin wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/64626/
> -----------------------------------------------------------
> 
> (Updated Dec. 14, 2017, 10:33 p.m.)
> 
> 
> Review request for mesos and Benjamin Mahler.
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> The new option allows to disable local message passing when messages are
> sent between UPIDs in the same OS process. Enabled by default.
> 
> 
> Diffs
> -----
> 
>   3rdparty/libprocess/src/process.cpp 75cf1d3b6d3d257ba9bc81c68017a74a6511cebf 
> 
> 
> Diff: https://reviews.apache.org/r/64626/diff/1/
> 
> 
> Testing
> -------
> 
> Ran `make check` and `make check LIBPROCESS_LOCAL_MESSAGE_PASSING=0` in libprocess.
> Ran `make check` in Mesos. Disabled local message passing breaks Mesos tests that use
the replicated log, because it seemingly relies on "immediate" local message delivery (see
[CoordinatorProcess::checkLearnPhase()](https://github.com/apache/mesos/blob/784b7f599f8ad8350fc7d6703a4b584301b23edc/src/log/coordinator.cpp#L396)).
> Manually verified that messages are going through the network with `tcpdump`.
> 
> 
> Thanks,
> 
> Ilya Pronin
> 
>


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