[gentoo-amd64] Re: Time Drift

[Duncan]

Björn Bredohl <bjoern@×××××××.de> posted 466A5C29.4020701@×××××××.de,
excerpted below, on  Sat, 09 Jun 2007 09:52:09 +0200:

> It works like this: hwclock keeps a file, /etc/adjtime, that keeps some
> historical information. This is called the adjtime file.

In addition, the Gentoo clock initscript is setup such that you can both 
control where adjtime is kept, and choose not to use it at all, 
recommended if you are using ntpd or the like.  The comment from the 
default /etc/conf.d/clock config, as of baselayout-1.13.0_alpha12 (~amd64 
AFAIK, tho I had unmasked and was testing it before it went ~arch), and 
some example settings:

# Newer FHS specs say this file should live in /var/lib rather than
# /etc.  If you care about such things, feel free to change this value.
# Note that a blank value means that you do not wish to even use the
# adjtime facility.  This is the default behavior as adjtime can be
# very fragile.  If the clock is updated without updating the adjtime
# file (which is common when using services such as ntp), then the
# clock can be screwed up when it gets updated at next boot.

#CLOCK_ADJTIME="/var/lib/adjtime"
#CLOCK_ADJTIME="/etc/adjtime"
#CLOCK_ADJTIME=""

That's straight out of my conf.d/clock config, so as you can see, I'm 
using the default, which is as noted, NOT to use /etc/adjtime.  FWIW, 
that's a fairly recent change here.  I was using it for quite awhile 
without serious issue I could see from either ntpd or the clock 
initscript, but having read about the possible issues, I decided to be 
safe rather than sorry, and disable it.

I then erased /etc/adjtime, and it hasn't been recreated, so the clock 
initscript indeed seems to be honoring my config.

Typically after an overnight suspend to disk (aka hibernate, here 
accomplished with a script that pauses the ntpd, ntp-client and clock 
scripts, syncing the hardware clock to the system clock at clock service 
pause as I've got that configured as well, then restarts them after the 
restart), when I restart and the services restart, ntp-client makes a 
correction of between 0.2 and 0.6 seconds, which the clock service was 
apparently originally taking into account with adjtime.  Now that I'm not 
using adjtime, the clock service isn't adjusting for that, but as I said, 
ntp-client catches it in the time-sync before ntpd starts (and the 
updated system clock is in turn copied to hwclock at hibernate or system 
shutdown, completing the update cycle), and that should be much more 
robust.

Note that this was AFAIK one of the features that changed a bit with 
baselayout-1.13, so those on stable, still using baselayout-1.12.x, will 
probably have a slightly different clock config, with a different way to 
disable /etc/adjtime.  If worse comes to worse, it should be possible to 
add a rm /etc/adjtime call to /etc/conf.d/local.start or whatever, so the 
file is removed automatically shortly after creation and thus can't be 
used by the clock initscript.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

-- 
gentoo-amd64@g.o mailing list