-----BEGIN PGP SIGNED MESSAGE-----
Hash: RIPEMD160


I switched my stratum one box to rc1 and it's died twice with "out of memory" - has anybody else
seen this?

FreeBSD 5.4 on a Soekris 4801, unmodified rc1 code (configure, make, make install)

syslog:
Mar 14 02:08:34 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync disabled 2307
Mar 14 02:09:28 gatekeeper.no-such-agency.net ntpd[51354]: pps sync disabled
Mar 14 02:09:37 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync enabled 2107
Mar 14 02:24:42 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync disabled 2307
Mar 14 02:25:38 gatekeeper.no-such-agency.net ntpd[51354]: pps sync disabled
Mar 14 02:25:48 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync enabled 2107
Mar 14 02:46:11 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync disabled 2307
Mar 14 02:47:08 gatekeeper.no-such-agency.net ntpd[51354]: pps sync disabled
Mar 14 02:47:17 gatekeeper.no-such-agency.net ntpd[51354]: kernel time sync enabled 2107
Mar 14 03:11:49 gatekeeper.no-such-agency.net ntpd[51354]: Exiting: No more memory!

ntp.conf has

driftfile /var/db/ntpd.drift
keys /etc/ntp.keys
trustedkey 1
enable auth
filegen peerstats file peerstats type day enable
filegen loopstats file loopstats type day enable
filegen clockstats file clockstats type day enable
discard average 30
discard minimum 3
restrict default kod nomodify limited
restrict 127.0.0.1
restrict 192.168.2.0 mask 255.255.255.0
restrict 192.168.3.0 mask 255.255.255.0
restrict 64.142.114.146
tos mindist 0.01
server 127.127.20.0  mode 2 version 4 prefer minpoll 6 maxpoll 6
fudge 127.127.20.0 flag3 1
server 192.168.3.252 key 1 minpoll 6  maxpoll 6
server skipjack.no-such-agency.net  key 1 minpoll 6
server someserver minpoll 7
server someotherserver minpoll 7
server yetmoreserver minpoll 7
server a.n.other.server minpoll 7
broadcast 192.168.2.255 key 1  minpoll 4
server john-desk-2.localnet noselect
server solar.localnet noselect


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (MingW32)

iD8DBQFEFyNNaVyA7PElsKkRA7+yAJ997nJisn3U4qllDGU0B4Q/jSGOugCgvC0W
SU0ZX6GAZiAbu85y6nhpOpY=
=IUS5
-----END PGP SIGNATURE-----

"John Pettitt"  wrote:

> ...
> Mar 14 03:11:49 gatekeeper.no-such-agency.net ntpd[51354]: Exiting: No 
> more memory!
> ...

If there was some correlation with the ntp load, then, according to a notice 
on www.pool.ntp.org, try the ulimit command.

--
Karel Sandler 

>>> In article <dv7oql$ttb$1@ns.felk.cvut.cz>, "Karel Sandler" <sandler@ujf.cas.cz> writes:

Karel> If there was some correlation with the ntp load, then, according to a
Karel> notice on www.pool.ntp.org, try the ulimit command.

Where is this notice on that website?  I don't see it...

H

>>> Where is this notice on that website?  I don't see it...

I'm running rc1 on FreeBSD 6.0 and I'm not seeing any memory problems;
but I did find the reference on the pool.ntp.org configuration page.
It is being reported as a LINUX issue and, as such, may not have
anything to do with the out of memory being reported here.

http://www.pool.ntp.org/join/configuration.html
>>Typical problems
>>"Out of memory"
>>Some linux distributions defaults to only allowing an application to lock 32KB memory. An ntpd in the >>pool requires more than that. In the ntpd startup script (often /etc/init.d/ntpd) add this before ntpd is >>started:
>>ulimit -l 8192

I also found this:

http://forums.fedoraforum.org/archive/index.php/t-36725.html
>>Thomas Zehetbauer:
>>I just hit the very same problem. Whenever my server was included in
>>pool.ntp.org it died within hours.
>>After some hours of experimentation and research I think I have found
>>this problem to be a combination of
>>1) misconfigured clients using 'peer' instead of 'server'; use 'nopeer'
>>for all client access restrict lines
>>2) ntpd trying to lock itself into memory; compile ntp without
>>HAVE_MLOCKALL
>>3) RLIMIT_MEMLOCK is only 32kB by default; raise with ulimit -l before
>>starting ntpd

-----BEGIN PGP SIGNED MESSAGE-----
Hash: RIPEMD160

Karel Sandler wrote:
> "John Pettitt"  wrote:
> 
>> ...
>> Mar 14 03:11:49 gatekeeper.no-such-agency.net ntpd[51354]: Exiting: No 
>> more memory!
>> ...
> 
> If there was some correlation with the ntp load, then, according to a notice 
> on www.pool.ntp.org, try the ulimit command.
> 
> --
> Karel Sandler 
> 
> 


This is on freebsd 5.3 (and 5.4) boxes - going back to stable-20060224 on the external S1 fixes it
the memory footprint is steady at 3.6mb over 7 hours - on my internal test S1 server (only 6
clients) it's grown from 2.2mb to 7.8mb since noon today.

At noon'ish when I restarted it on the internal server
> jeeves# ps -alx | egrep ntpd
>     0 27045     1   1  96  0  2140  2160 select Ss    ??    0:00.02 /usr/sbin/ntpd -p /var/run/ntpd.pid -f /var/db/ntpd.drift


at 7.05 pm
> jeeves# ps -alx | egrep ntpd
>     0 27045     1   0  96  0  7820  7844 select Ss    ??    0:05.24 /usr/sbin/ntpd -p /var/run/ntpd.pid -f /var/db/ntpd.drift


looks like a memory leak to me.

John
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (MingW32)

iD8DBQFEF4TmaVyA7PElsKkRAwSpAJ4uk+wQhdJHHBWBmfHUPAhhC5ybegCgnPmK
7Uar2O7DnL36IqhdLnVApro=
=ipfd
-----END PGP SIGNATURE-----

John Pettitt wrote:
> 
> I switched my stratum one box to rc1 and it's died twice with "out of memory" - has anybody else
> seen this?
> 
> FreeBSD 5.4 on a Soekris 4801, unmodified rc1 code (configure, make, make install)
> 
>> Mar 14 03:11:49 gatekeeper.no-such-agency.net ntpd[51354]: Exiting: No more memory!
> 
> ntp.conf has
> 
> driftfile /var/db/ntpd.drift
> keys /etc/ntp.keys
> trustedkey 1
> enable auth
> filegen peerstats file peerstats type day enable
> filegen loopstats file loopstats type day enable
> filegen clockstats file clockstats type day enable
> discard average 30
> discard minimum 3
> restrict default kod nomodify limited
> restrict 127.0.0.1
> restrict 192.168.2.0 mask 255.255.255.0
> restrict 192.168.3.0 mask 255.255.255.0
> restrict 64.142.114.146
> tos mindist 0.01
> server 127.127.20.0  mode 2 version 4 prefer minpoll 6 maxpoll 6
> fudge 127.127.20.0 flag3 1

Can you comment out your NMEA refclock and see if it stabilizes? You
shouldn't see it increasing memory by much after the first few minutes.

Thanks,

Danny
_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

Danny Mayer wrote:

> 
> Can you comment out your NMEA refclock and see if it stabilizes? You
> shouldn't see it increasing memory by much after the first few minutes.
> 




I cut ntp conf back to this

driftfile /var/db/ntpd.drift
server 192.168.3.254

and it's still growing.

John

I'm also seeing something that looks like a memory leak. I've run rc1
on a couple of
systems, with --disable-all-clocks and also with --disable-all-clock
--enable-HOPFSERIAL
--enable-ATOM --enable-RAWDCF; in all cases, I observed monotonically
increasing
memory occupation (with the Solaris 'ps' command). On the smallest
system
(Sun SPARC, Solaris 7) ntpd crashed, once with an explicit
out-of-memory and
once with a bus error.

When it's running, rc1 shows significantly better convergence than
v4.2.0.

Paul

"Harlan Stenn" <stenn@ntp.isc.org> wrote in message 
news:ywn9acbscuj6.fsf@ntp1.isc.org...
>>>> In article <dv7oql$ttb$1@ns.felk.cvut.cz>, "Karel Sandler" 
>>>> <sandler@ujf.cas.cz> writes:
>
> Karel> If there was some correlation with the ntp load, then, according to 
> a
> Karel> notice on www.pool.ntp.org, try the ulimit command.
>
> Where is this notice on that website?  I don't see it...
>
I apologize for this short reference. At 2am I used only my memory and also 
forget the relation to LINUX, as was noticed by jkline in another message in 
this thread.

Karel Sandler 

"John Pettitt" <jpp@cloudview.com> wrote in message 
news:4417234D.3020200@cloudview.com...
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: RIPEMD160
>
>
> I switched my stratum one box to rc1 and it's died twice with "out 
> of memory" - has anybody else
> seen this?

I'm running RC1 on 4.11, no clients, approx 60 servers conf'd, and 
memory usage has been constant.  If a bunch of people want to try 
slamming it with clients, I'll open the restrictions for a bit, just 
don't expect stellar time from it. : )

Joshua Coombs