Hi Dan, On Wed, 2 Nov 2016, 11:30 AM 積丹尼 Dan Jacobson <[email protected]> wrote:
ps should have a way to get the process start time from the local instance of the VPS, not the entire underlying computer. For real VPS they do because there is a separate instance of the /proc filesystem. $ sleep 4& ps -eo pid,cmd,start_time,bsdstart,start,lstart,stime|grep sleep|grep -v grep [1] 978 978 [sleep] 21:26 21:26 21:26:48 Wed Nov 2 21:26:48 2016 21:26 $ uptime 21:26:53 up 3 min, 1 user, load average: 0.66, 0.18, 0.06 [1]+ Done sleep 4 $ date Wednesday 2 November 21:26:57 AEDT 2016 That is on a kvm instance that has a different uptime to the underlying computer. Start times are basically get_btime plus the proc_start_time / Hertz get_btime is from grep btime /proc/stat proc_start_time is 22nd field in /proc/<PID>/stat Hertz is usually 100 So, like this: *$ *sleep 100 & [1] 10204 *$ *grep btime /proc/stat | cut -f 2 ; cut -f 22 -d' ' /proc/10204/stat btime 1477391669 69159649 *$ *date -d @`expr 1477391669 + 691596` Wednesday 2 November 21:41:05 AEDT 2016 *$ *ps -o stime 10204 STIME 21:41 Note I manually divided the processes start time by 100. So that field is nonsense... Actually no, whatever software the VPS uses is using a bad implementation of /proc. I suspect that /proc/stat is the culprit here. ps and the rest of the procps tools just interprets what it finds in /proc, so if /proc is wrong they will happily show the wrong thing. This doesn't sound like a real VPS but a container. Sometimes LXC or docker shells show this sort of odd behaviour. Do you know what virtualisation they are using there? - Craig
