Back in February, when I upgraded to 4.2.8p2, I *had* to add "rlimit memlock
0" to ntp.conf in order to get them to even run.  Today I upgraded several
4.2.8p2 servers running on Centos 7 to 4.2.8p4.  They immediately ran out of
memory and stopped seconds after starting.  So now in 4.2.8p4, I've had to
remove it to prevent out of memory crashes.  Servers have 512-1024MB RAM and
does nothing else but NTP.

Is there a simplified release notes that is current (matching this: " It
addresses 13 low- and medium-severity security issues and provides about 100
bugfixes and enhancements over 4.2.8p3.") ? The one on the website is from
when v4 came out in 2011.
https://www.eecis.udel.edu/~mills/ntp/html/release.html.

Here are some related bugs from Bugzilla:

http://bugs.ntp.org/show_bug.cgi?id=2915        Bug 2915 - ntpd doesn't
start with "rlimit memlock 0"
http://bugs.ntp.org/show_bug.cgi?id=2908        Bug 2908 - Only runs for a
few seconds on two Linux versions.
http://bugs.ntp.org/show_bug.cgi?id=2817        Bug 2817 - Stop locking ntpd
into memory by default

In bug 2817, this behaviour is confirmed and still present:

David Taylor 2015-09-30 07:23:01 UTC says:
" Under Linux RasPi-8 4.0.8-v7+ for the Raspberry Pi, ntp-dev-4.3.72, the
line
"rlimit memlock 0" /must/ be present for ntpd.to work.  However, if that
line
is left in the ntp.conf for ntp-dev-4.3.73, ntpd does /not/ work, and the
line
needs to be removed."

Some critical thinking needs to be done on this rlimit memory management to
put it to bed as this is a real nuisance and is only going to cause more
people to delay upgrading to current coming from stable builds that 'just
work'.

Is this now true for everyone on linux, remove ANY rlimit value to get it to
run?  I don't have a big enough OS and memory sample to know if this is just
me or a lot of others.

This probably is bad in linux world, but I come from the philosophy that if
an upgrade REQUIRES a configuration change to operate, it should be done
automatically.  Or else have a really good documentation/changelog that
clearly points this out.

Best regards,
Mike


-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Harlan
Stenn
Sent: March 15, 2015 2:40 PM
To: Michael Meier
Cc: [email protected]
Subject: Re: [Pool] "out of memory" crashes since 4.2.8 upgrade

Michael Meier writes:
> > See http://bugs.ntp.org/2471 - this feels like the rlimit issue to me.
> > Try adding:
> >  rlimit memlock 128
> > to your ntp.conf file.  This is an issue with glibc and friends, and it
> > sucks to make the amount of memory locked for ntpd to be too big, and if
> > it's not big enough then ntpd will crash.  The default value of 32
> > megabytes is enough for most everybody.
> 
> Better yet, set "rlimit memlock 0" to disable the behaviour of ntpd
> trying to lock itself into memory altogether.
> I have yet to find a limit that doesn't crash on a very busy server
> after $random_timespan. And it's also a waste of your memory, because it
> forbids the OS to simply swap out things that are never used.

I like this idea best of all.

I realize now I've been focusing on getting the "right" number because I
*assumed* that folks were complaining about this because they *wanted*
to lock ntpd in memory.

I'm going to start a discusison about this on questions@.

Thanks, Michael!

H
_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool

_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool

Reply via email to