Inconsistent Handling of the Process Threads

[PeterZaitsev]

It looks like different data is handled inconsistently for threads belonging to the process

for namedprocess_namegroup_cpu_user_seconds_total the value seems to include all the threads, same for namedprocess_namegroup_read_bytes_total however namedprocess_namegroup_context_switches_total seems to only report value from the main thread which is not meaningful value for threaded processes.

[ncabatoff]

Thanks for the bug report, sadly you appear to be correct. The documentation for all this isn't wonderful, but I guess for anything that can be different from one thread to another (like context switches) there's no way to obtain the aggregate value without visiting every thread's proc files.

Unfortunately this means I'm going to have to either eliminate the -threads option and make it mandatory, or make the context switch metrics contingent on providing that option. I'll have to think about that, but for now I'll start by getting something working.

[ncabatoff]

I fixed this in the pre-release v0.3.11, by reading the thread data whether or not -threads is given. The thread-based metrics are only emitted when -threads is given, as before. I haven't yet decided whether to eliminate the option altogether and always emit them.