[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: High-Precision NFS Timestamps

From: Eric Blake
Subject: Re: High-Precision NFS Timestamps
Date: Wed, 12 Jan 2011 15:30:07 -0700
User-agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv: Gecko/20101209 Fedora/3.1.7-0.35.b3pre.fc14 Lightning/1.0b3pre Mnenhy/0.8.3 Thunderbird/3.1.7

[adding bug-make, replies can drop automake]

On 01/12/2011 03:08 PM, Eric Reischer wrote:
> I recently upgraded our NFS fileserver to an ext4 filesystem, and since
> then we've been having clock skew warnings from make (3.81).  Because
> the ext3 filesystem that was previously running on the NFS server didn't
> support high-precision timestamps, we didn't have clock skew warnings
> because an hourly cron job kept the clocks in relatively close sync. 
> However, since ext4 does support sub-second timestamps, we're getting
> warnings about clock skews on the order of a few milliseconds.  I tried
> adding our target objects to the .LOW_RESOLUTION_TIME target, however
> now we're getting a warning about
> make: *** Warning: .LOW_RESOLUTION_TIME file .lib_deps has a high
> resolution time stamp
> make:  warning: Clock skew detected.  Your build may be incomplete.
> Since asking to have system clocks of NFS clients and servers synced to
> sub-millisecond precision is a bit excessive, is there a way (or feature
> request: could there be a way) to set the threshold for clock skews
> either via a runtime Makefile directive, or an environment variable?  I
> can't see any other way to eliminate these warnings without reverting
> back to ext3 (which isn't happening).  If it's a hard-coded value, there
> should be at least a little room for slop in there (~5 milliseconds?)
> due to network lag, etc.  Even on a point-to-point network, there's
> going to be one-way transit times of a millisecond or two.

This is more likely a make issue than an automake issue.

That said, NFS timestamps are indeed a source of pain, and gnulib has
recently put in quite a bit of work dealing with timestamp skew across
NFS (including code to ignore relatively close timestamps, rather than
insist that sub-millisecond differences represent skew); some of that
may be worth incorporating into the make code base if it isn't there

Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

Attachment: signature.asc
Description: OpenPGP digital signature

reply via email to

[Prev in Thread] Current Thread [Next in Thread]