Bug#657078: patches to reduce the footprint of the nfs4 idmapper
Hi,
On 04/24/2012 10:45 PM, Jonathan Nieder wrote:
Could you describe the problem in more detail? Under what
circumstances does the idmapper provoke allocation failures, and how
effectively do the patches avoid trouble?
We have two big NFSv4 file servers that also automount from each other
to provide homedirs over samba.
These machines are accessed by a few hundred NFSv4 clients. On these
(RHEL6) servers we see this issue come up sometimes twice a day,
sometimes twice a week (since 6.2).
The are reports that Debian is affected by this bug as well, for example
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=636306
I've tried to reproduce this issue in the past but failed, so it's very
hard to determine if the patches help a lot. But from what I've been
able to find the patches dramatically reduce the memory needed (now
order 4).
I don't know if you can see the following Red Hat bug report?
https://bugzilla.redhat.com/show_bug.cgi?id=730045
An upstream thread that starts talking about the memory issue can be
found here:
http://www.spinics.net/lists/linux-nfs/msg27350.html
An initial patch brought this down from order 4 to order 2.
A bit further down the thread they mention:
----
Much, much better! Nice work.
Built on a different machine this time with a different compiler, so
sizes are a little different. Not certain why, but they're fairly close
-- maybe differences in padding or something:
Without either patch:
(gdb) p sizeof(struct idmap)
$1 = 39168
With just the first patch:
(gdb) p sizeof(struct idmap)
$1 = 8448
With both patches:
(gdb) p sizeof(struct idmap)
$1 = 272
----
Since the thread is also about the new nfs ID mapper, it could be that
this large reduction is only (partialy) achieved if the new id mapper is
in use. But since they applied it in Fedora 16, which also still has the
old id mapper, I assume it's good.
NFSv4: Further reduce the footprint of the idmapper
commit 685f50f9188ac1e8244d0340a9d6ea36b6136cec
NFSv4: Reduce the footprint of the idmapper
commit d073e9b541e1ac3f52d72c3a153855d9a9ee3278
Whether or not these get applied in the stable tree (which would be
nice if it can be justified), I think we definitely want them in
Debian, so cloning the bug to track that.
I agree! (but I'm biased).
Regards,
Rik
--
Rik Theys
Senior System Engineer
KU Leuven - Dept. Elektrotechniek (ESAT)
Kasteelpark Arenberg 10
B-3001 LEUVEN - HEVERLEE
Tel.: +32(0)16/32.11.07
----------------------------------------------------------------
<<Any errors in spelling, tact or fact are transmission errors>>
Reply to: