Hi, We've got a number of sles9 servers with the same issue, all of them seem to have process start times in the future.

An example (I've replaced the machine name with <machine> and the username with <user> before anyone asks).

<machine> /proc> date ; ps -f
Mon Dec 4 12:23:16 GMT 2006
UID PID PPID C STIME TTY TIME CMD
<user> 18123 18122 0 12:07 pts/3 00:00:00 -csh
<user> 16880 18123 0 12:43 pts/3 00:00:00 ps -f

As you can see the STIME is 20min later than the current system time, system time is set via ntp,the hardware clock is out but by 1 hour (probably due to BST)

<machine>:/ # date;hwclock --show
Mon Dec 4 13:24:57 GMT 2006
Mon 04 Dec 2006 02:24:58 PM GMT -0.438085 seconds

Looking in /proc the processes seem to be created at the correct time:

<machine> /proc> sleep 100 &; ps -f ;
[1] 30263
UID PID PPID C STIME TTY TIME CMD
<user> 18123 18122 0 12:07 pts/3 00:00:00 -csh
<user> 30263 18123 0 12:58 pts/3 00:00:00 sleep 100
<user> 30264 18123 0 12:58 pts/3 00:00:00 ps -f
<machine> /proc> ls -ltd 30263
dr-xr-xr-x 3 <user> users 0 2006-12-04 12:38 30263

It's fairly obviously its something that drifts the longer that the machine is up:

sles9test:~ # uptime;date;ps -f
1:37pm up 6 days 21:36, 2 users, load average: 0.00, 0.00, 0.00
Mon Dec 4 13:37:43 GMT 2006
UID PID PPID C STIME TTY TIME CMD
root 22732 22729 0 13:37 pts/0 00:00:00 -bash
root 22753 22732 0 13:37 pts/0 00:00:00 ps -f

<machine1> /home/<user>> uptime;date;ps -f
1:01pm up 99 days 2:20, 3 users, load average: 2.45, 1.77, 1.63
Mon Dec 4 13:01:17 GMT 2006
UID PID PPID C STIME TTY TIME CMD
<user> 16061 16059 0 13:19 pts/5 00:00:00 -csh
<user> 18540 16061 0 13:21 pts/5 00:00:00 ps -f

<machine2> /home/<user>> uptime;date;ps -f
1:01pm up 123 days 20:24, 3 users, load average: 1.21, 1.29, 1.34
Mon Dec 4 13:01:18 GMT 2006
UID PID PPID C STIME TTY TIME CMD
<user> 16709 16708 0 13:25 pts/3 00:00:00 -csh
<user> 17449 16709 0 13:25 pts/3 00:00:00 ps -f

<machine3> /home/<user>> uptime;date;ps -f
1:01pm up 123 days 2:38, 5 users, load average: 0.24, 0.22, 0.27
Mon Dec 4 13:01:14 GMT 2006
UID PID PPID C STIME TTY TIME CMD
<user> 13332 13331 0 13:23 pts/6 00:00:00 -csh
<user> 16519 13332 0 13:25 pts/6 00:00:00 ps -f

<machine4> /home/<user>> uptime;date;ps -f
1:01pm up 149 days 21:40, 3 users, load average: 1.82, 1.34, 1.01
Mon Dec 4 13:01:15 GMT 2006
UID PID PPID C STIME TTY TIME CMD
<user> 8433 8432 0 13:29 pts/2 00:00:00 -csh
<user> 10209 8433 0 13:31 pts/2 00:00:00 ps -f

Some more details just in case:
SLES9, kernel version 2.6.5-7.139-bigsmp, no service packs applied.

Does anyone know why it does this?
Is this actually an issue?
Is there a fix for it?

Thanks

David

The gradual drift in error can be accounted for as follows:

a) The time server(s) are blocking your attempts to synchronise with their servers because you have not requested permission to do so.

b) The time server(s) are blocking your attempts to synchronise with their servers because the frequency of your time checks exceeds the frequency they require. That is, your time hacks occur too frequently, consuming bandwidth, so they are blocking you.

So, the thing to do would be to verify that you are not synchronising too often, and that you are not being blocked because you failed to notify them of your use of their service.

There may be other factors at work, but these are the factors that come to mind most readily.

Nono, you've missunderstood, nothing to do with ntp, the *time* is correct and ntp is syncing correctly but whenever you do a 'ps' the start times for the processes are in the future. Wouldn't matter what the time is, you shouldn't start a new process and see the start time in the future!

The times seem to advance ~10 minutes for every 50 days the server is up, a reboot sorts it.

I've since found the same thing happens on SuSE 9.1 but doesn't do it on redhat 9, I've not got a 10/10.1 box that's been up long enough to see if it also does it on those. :-/

Thanks

David

<bump>
Anyone?