compiling on nfs directories

[Russell L. Carter]

On 12/16/14 20:39, Russell L. Carter wrote:
>
>
> On 12/16/14 11:37, John Baldwin wrote:
>> On Monday, December 15, 2014 3:59:29 pm Rick Macklem wrote:
>
> [...]
>
>>> What I suspect might cause this is one of two things:
>>> 1 - The modify time of the file is now changing at a time the Linux
>>>      client doesn't expect, due to changes in ZFS or maybe TOD clock
>>>      resolution. (At one time, the TOD clock was only at a resolution
>>>      of 1sec, so the client wouldn't see the modify time change often.
>>>      I think it is now at a much higher resolution, but would have to
>>>      look at the code/test to be sure.)
>>
>> No, it's still only a second resolution on FreeBSD by default.  You can
>> make this precise on the NFS server by setting the
>> vfs.timestamp_precision
>> sysctl to 3.  We should probably be using that by default for at least
>> server-class systems.
>>
>
> Hmm, what's this?  Let's see:
>
> rcarter at feyerabend> uname -a
> FreeBSD feyerabend.n1.pinyon.org 10.1-STABLE FreeBSD 10.1-STABLE #1
> r275516+3a52b5f(stable-jhb-em): Sat Dec  6 10:37:16 MST 2014
> toor at feyerabend.n1.pinyon.org:/usr/obj/usr/src/sys/RLCGSV  amd64
> rcarter at feyerabend> man -k vfs.timestamp_precision
> vfs.timestamp_precision: nothing appropriate
> rcarter at feyerabend> sysctl -d vfs.timestamp_precision
> vfs.timestamp_precision: File timestamp precision (0: seconds, 1: sec +
> ns accurate to 1/HZ, 2: sec + ns truncated to ms, 3+: sec + ns (max.
> precision))
> rcarter at feyerabend> sysctl vfs.timestamp_precision
> vfs.timestamp_precision: 0
>
> Ah, that's *VERY* interesting.  I am unfortunately leaving the
> physical vicinity of my server farm soon, so not the right time for
> experiments.  But I have been whining for some time now about what
> looks to be very similar to gerrit.kuehn's symptoms.  I see them on
> installworlds via NFS v4.1, on -current or stable/10-trunk.  About 9
> out of 10 installs fail trying to rebuild parts of the tree.  I
> finally resorted to copying /usr/obj* around and then just mounting
> /usr/src via NFS.  ick.  Oh, and also buildworld/buildkernel -j1.  A
> pity on a cluster where 8 cores/system are the norm.  But now I have
> something sensible to try.  Looking forward to it.

After figuring out a way to test this reversibly, I tried the following:

server & client vfs.timestamp_precision=3, make -j12 buildworld/kernel,
and make installworld -j1 on the client => fail, in /usr/src/sys/boot

server & client vfs.timestamp_precision=0, make -j1 build/install,
succeeds.

Worth a shot anyway...

Cheers,
Russell

> Happy holidays, and cheers!
> Russell
> _______________________________________________
> freebsd-net at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-net
> To unsubscribe, send any mail to "freebsd-net-unsubscribe at freebsd.org"