pidstat command

joemon83 · 08-19-2020, 10:04 AM

Hi,
Can someone explain, why kB_rd/s and kB_wr/s different between these 2 pidstat command:

Code:

pidstat -dl | egrep 'mfs|Command'
08:01:00 AM   UID       PID   kB_rd/s   kB_wr/s kB_ccwr/s  Command
08:01:00 AM  4937     34314    506.30    170.19      0.00  /opt/mapr/server/mfs -b -p 5660 -r 40 -T 40 -a 100 -m 61656 -O /opt/mapr/conf/mapr-clusters.conf -6 /opt/cores 


pidstat -dl 1 1 | egrep 'mfs|Command'
08:01:18 AM   UID       PID   kB_rd/s   kB_wr/s kB_ccwr/s  Command
08:01:19 AM  4937     34314 117444.44   9888.89      0.00  /opt/mapr/server/mfs -b -p 5660 -r 40 -T 40 -a 100 -m 61656 -O /opt/mapr/conf/mapr-clusters.conf -6 /opt/cores

berndbausch · 08-19-2020, 10:24 AM

Your question is an excellent showcase for the need of code tags. It's very hard to read your command output.

However, what makes you think that values measured at two different times should be identical? I consider it normal that IO metrics change over time.

EDIT: I agree the difference is huge. Still, this seems to be a map-reduce application, i.e. big data, i.e. potential for reading/writing big datasets. Sometimes it churns away calculating something, sometimes it deposits data somewhere and/or gets new data.

syg00 · 08-19-2020, 06:04 PM

One of the problems of relying on sampling tools without knowing the underlying data access pattern. The wider the sampling period, say typically 10 or 30 minutes, the less likely you are to even be aware this is happening. But it might (will) be affecting other I/O on the system and causing pain you can't even see.

Nice pick-up.