Please guide me

[birdie-github]

I feel really stupid.

The old python iotop could show processes IO data for their entire lifetimes and you could show only processes that have actually read or written something, while not showing processes that have had zero IO activity.

I don't understand how it can be done with with iotop-c.

I run iotop 1.30.

I press o/O - OK, everything has disappeared.

I press a/A multiple times - I still see nothing while I absolutely have to see a couple of processes at the very least.

I don't understand how it all works.

[bbonev]

A has two modes, use h to see which mode is the current one. It may help to switch off o and check all processes or use search / to find a particular one...

iotop does not see the processes lifetime IO, it will only show from starting iotop to the current moment.

Also note that even if you enable taskstats, that does not affect the already running processes. Only the ones started afterwards. Enable taskstats from the kernel command line (e.g. taskstats=1) to have that data on all processes.

[birdie-github]

I've toggled A several times and used o. I either see no processes at all (my laptop barely reads/writes anything) or all the processes despite them having 0 for their activity.

I'm even more confused.

I presume your application reads /proc/PID/io counters that exist for the entire lifetime of a process.

I would love to be able to see those and IO per second in addition to that.

Is that possible?

          PID TotalReadBytes TotalWriteBytes ReadBytesPerSec WriteBytesPerSec IOusePercentage
processA 
processB
processC

This is what I want.

And only for the processes that have at least written or read a single byte (in Linux sadly the granularity AFAIK is in 4K pages only).

I don't want to see 5 pages of processes that haven't read or written anything at all.

[bbonev]

No, the data does not come from /proc/PID/io but from taskstats over netlink, which should not be much different - most probably the accumulated value is already there...

iotop uses the counter difference to display the current rate, but for accumulated it sums those differences for iotop's lifetime, which may differ from process' lifetime. Changing that logic will only require to get the process' start time in CLOCK_MONOTONIC in order to calculate the average rate for the lifetime... Does it make more sense in this way?