<div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div>Regarding the use of malloc vs calloc, i had tried this a while back, though theoretically it should give performance improvement, i was not able to see this in practice for small memory allocations, may be due to the optimizations in the glibc and kernel memory management itself. When i ran a sample program for random size millions of malloc vs calloc and also use it(so that it results in page fault) resulted in no conclusive results on which is more performant, in some runs calloc was slower in some runs malloc was slower. In glibc the calloc is not exactly malloc + memset, after browsing and looking through __libc_calloc(), the kernel always gives zeroed pages for security purpose(with COW), when calloc implementation in libc reuses the memory allocated by its own process then there is a difference in perf of malloc and calloc, but even these are optimised to some extent in glibc. I would personally prefer keeping calloc and use malloc only if its large buffers, and trade off this one nano perf improvement, for the ease of coding and uniformity across the project. As a thumb rule we have been using calloc for smaller memory, we should discuss and conclude on this so it can go in developer-guide(./doc/developer-guide/coding-standard.md)</div><div><br></div><div>Also, while at the memory alloc optimization discussions, would like to mention about one another observation, long back had tried doing some changes to gf_malloc and gf_calloc to use from
the per thread mem pool if available, rather than send it to libc. We saw that there were
millions of small mallocs/callocs happening so we tried to modify the
code to lookup the mempool for availability for pre allocated memory,
with this change we saw no improvement! But there were improvements wrt using iobuf_pool and other pre allocated poo for dict, inode etcl. something we could explore.</div><div><br></div><div>Regards,</div><div>Poornima<br></div></div></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 27, 2019 at 2:25 PM Niels de Vos <<a href="mailto:ndevos@redhat.com">ndevos@redhat.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On Tue, Mar 26, 2019 at 02:52:33PM -0700, Vijay Bellur wrote:<br>
> On Thu, Mar 21, 2019 at 8:44 AM Yaniv Kaul <<a href="mailto:ykaul@redhat.com" target="_blank">ykaul@redhat.com</a>> wrote:<br>
> <br>
> ><br>
> ><br>
> > On Thu, Mar 21, 2019 at 5:23 PM Nithya Balachandran <<a href="mailto:nbalacha@redhat.com" target="_blank">nbalacha@redhat.com</a>><br>
> > wrote:<br>
> ><br>
> >><br>
> >><br>
> >> On Thu, 21 Mar 2019 at 16:16, Atin Mukherjee <<a href="mailto:amukherj@redhat.com" target="_blank">amukherj@redhat.com</a>> wrote:<br>
> >><br>
> >>> All,<br>
> >>><br>
> >>> In the last few releases of glusterfs, with stability as a primary theme<br>
> >>> of the releases, there has been lots of changes done on the code<br>
> >>> optimization with an expectation that such changes will have gluster to<br>
> >>> provide better performance. While many of these changes do help, but off<br>
> >>> late we have started seeing some diverse effects of them, one especially<br>
> >>> being the calloc to malloc conversions. While I do understand that malloc<br>
> >>> syscall will eliminate the extra memset bottleneck which calloc bears, but<br>
> >>> with recent kernels having in-built strong compiler optimizations I am not<br>
> >>> sure whether that makes any significant difference, but as I mentioned<br>
> >>> earlier certainly if this isn't done carefully it can potentially introduce<br>
> >>> lot of bugs and I'm writing this email to share one of such experiences.<br>
> >>><br>
> >>> Sanju & I were having troubles for last two days to figure out why<br>
> >>> <a href="https://review.gluster.org/#/c/glusterfs/+/22388/" rel="noreferrer" target="_blank">https://review.gluster.org/#/c/glusterfs/+/22388/</a> wasn't working in<br>
> >>> Sanju's system but it had no problems running the same fix in my gluster<br>
> >>> containers. After spending a significant amount of time, what we now<br>
> >>> figured out is that a malloc call [1] (which was a calloc earlier) is the<br>
> >>> culprit here. As you all can see, in this function we allocate txn_id and<br>
> >>> copy the event->txn_id into it through gf_uuid_copy () . But when we were<br>
> >>> debugging this step wise through gdb, txn_id wasn't exactly copied with the<br>
> >>> exact event->txn_id and it had some junk values which made the<br>
> >>> glusterd_clear_txn_opinfo to be invoked with a wrong txn_id later on<br>
> >>> resulting the leaks to remain the same which was the original intention of<br>
> >>> the fix.<br>
> >>><br>
> >>> This was quite painful to debug and we had to spend some time to figure<br>
> >>> this out. Considering we have converted many such calls in past, I'd urge<br>
> >>> that we review all such conversions and see if there're any side effects to<br>
> >>> it. Otherwise we might end up running into many potential memory related<br>
> >>> bugs later on. OTOH, going forward I'd request every patch<br>
> >>> owners/maintainers to pay some special attention to these conversions and<br>
> >>> see they are really beneficial and error free. IMO, general guideline<br>
> >>> should be - for bigger buffers, malloc would make better sense but has to<br>
> >>> be done carefully, for smaller size, we stick to calloc.<br>
> >>><br>
> >><br>
> >>> What do others think about it?<br>
> >>><br>
> >><br>
> >> I believe that replacing calloc with malloc everywhere without adequate<br>
> >> testing and review is not safe and am against doing so for the following<br>
> >> reasons:<br>
> >><br>
> ><br>
> > No patch should get in without adequate testing and thorough review.<br>
> ><br>
> <br>
> <br>
> There are lots of interesting points to glean in this thread. However, this<br>
> particular one caught my attention. How about we introduce a policy that no<br>
> patch gets merged unless it is thoroughly tested? The onus would be on the<br>
> developer to provide a .t test case to show completeness in the testing of<br>
> that patch. If the developer does not or cannot for any reason, we could<br>
> have the maintainer run tests and add a note in gerrit explaining the tests<br>
> run. This would provide more assurance about the patches being tested<br>
> before getting merged. Obviously, patches that fix typos or that cannot<br>
> affect any functionality need not be subject to this policy.<br>
> <br>
> As far as review thoroughness is concerned, it might be better to mandate<br>
> acks from respective maintainers before merging a patch that affects<br>
> several components. More eyeballs that specialize in particular<br>
> component(s) will hopefully catch some of these issues during the review<br>
> phase.<br>
<br>
Both of these points have always been strongly encouraged. They are also<br>
documented in the<br>
<a href="https://docs.gluster.org/en/latest/Contributors-Guide/Guidelines-For-Maintainers/" rel="noreferrer" target="_blank">https://docs.gluster.org/en/latest/Contributors-Guide/Guidelines-For-Maintainers/</a><br>
<a href="https://github.com/gluster/glusterdocs/blob/master/docs/Contributors-Guide/Guidelines-For-Maintainers.md" rel="noreferrer" target="_blank">https://github.com/gluster/glusterdocs/blob/master/docs/Contributors-Guide/Guidelines-For-Maintainers.md</a><br>
(formatting is broken in the 1st link, but I dont know how to fix it)<br>
<br>
We probably need to apply our own guidelines a little better, and<br>
remember developers that > 90% of the patch(series) should come with a<br>
.t file or added test in an existing one.<br>
<br>
And a big +1 for getting reviews or at least some involvement of the<br>
component maintainers.<br>
<br>
Niels<br>
_______________________________________________<br>
Gluster-devel mailing list<br>
<a href="mailto:Gluster-devel@gluster.org" target="_blank">Gluster-devel@gluster.org</a><br>
<a href="https://lists.gluster.org/mailman/listinfo/gluster-devel" rel="noreferrer" target="_blank">https://lists.gluster.org/mailman/listinfo/gluster-devel</a><br>
</blockquote></div>