mesos-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benno Evers <bev...@mesosphere.com>
Subject Re: [Proposal] Use jemalloc as default memory allocator for Mesos
Date Tue, 22 Aug 2017 10:16:10 GMT
Hi Alexander,

in general, jemalloc and tcmalloc are very similar, and seem to be taking
ideas from each other (in fact the jeprof executable started as a copy of
pprof and there are still references the pprof documentation in some
comments)

>From what I've seen, the main difference is that the profiling seems
better-suited to multi-threaded programs, in particular the profile file
format includes per-thread memory statistics and the profiling features can
be turned on and off individually per thread. From an API perspective, all
settings can be accessed by the mallctl() function, while it seems that
tcmalloc requires some options to be set by environment variable (
https://gperftools.github.io/gperftools/heapprofile.html). Finally, I also
found the documentation to be more thorough.

But again, the two are very similar, so I think the main decision here
isn't whether to choose jemalloc or tcmalloc but whether to switch to a
custom memory allocator that has support for profiling heap memory usage.


On Mon, Aug 21, 2017 at 4:26 PM, Alexander Rojas <alexander@mesosphere.io>
wrote:

> Hi Benno,
>
> This does sound like a great addition to Mesos. Can you however explain
> how jemalloc is better than tcmalloc? I think that for such important
> change, we probably need some more information.
>
> Your comment in MESOS-7876 mentions that we already have tcmalloc since it
> is part of gperftools, so I would like to have a whole picture of the
> advantages and disadvantages of both options.
>
> Alexander Rojas
> alexander@mesosphere.io
>
>
>
>
> > On 18. Aug 2017, at 12:49, Benno Evers <bevers@mesosphere.com> wrote:
> >
> > Hi all,
> >
> > I would like to propose bundling jemalloc as a new dependency
> > under `3rdparty/`, and to link Mesos against this new memory
> > allocator by default.
> >
> >
> > # Motivation
> >
> > The Mesos master and agent binaries are, ideally, very long-running
> > processes. This makes them susceptible to memory issues, because
> > even small leaks have a chance to build up over time to the point
> > where they become problematic.
> >
> > We have seen several such issues on our internal Mesos installations,
> > for example https://issues.apache.org/jira/browse/MESOS-7748
> > or https://issues.apache.org/jira/browse/MESOS-7800.
> >
> > I imagine any organization running Mesos for an extended period
> > of time has had its share of similar issues, so I expect this
> > proposal to be useful for the whole community.
> >
> >
> > # Why jemalloc?
> >
> > Given that memory issues tend to be most visible after a given
> > process has been running for a long time, it would be great to
> > have the option to enable heap tracking and profiling at runtime,
> > without having to restart the process. (This ability could then
> > be connected to a Mesos endpoint, similar to how we can adjust
> > the log level at runtime)
> >
> > The two production-quality memory allocators that have this
> > ability currently seem to be tcmalloc and jemalloc. Of these,
> > jemalloc does produce in our experience better and more
> > detailed statistics.
> >
> >
> > # What is the impact on users who do not need this feature?
> >
> > Naturally, not every single user of Mesos will have a need
> > for this feature. To ensure these users would not experience serious
> > performance regressions as a result of this change, we conducted
> > a preliminary set of benchmarks whose results are collected
> > under https://issues.apache.org/jira/browse/MESOS-7876
> >
> > It turns out that we could probably even expect a small speedup (1% - 5%)
> > as a nice side-effect of this change.
> >
> > Users who compile Mesos themselves would of course have the option
> > to disable jemalloc at configuration time or replace it with their
> > memory allocator of choice.
> >
> >
> >
> > I'm looking forward to hear any thoughts and comments.
> >
> >
> > Thanks,
> > --
> > Benno Evers
> > Software Engineer, Mesosphere
>
>


-- 
Benno Evers
Software Engineer, Mesosphere

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