disk_io_counters() values on NetBSD are too high

[giampaolo]

$ make install && python -c "import psutil; from pprint import pprint as pp; pp(psutil.disk_io_counters(1))"
{'wd0': sdiskio(read_count=13512837104935596480L, write_count=13511736081481412928L, read_bytes=13513043022849399608L, write_bytes=110518291808L)}

...also, they all look similar.
We're definitively doing something wrong.

[giampaolo]

Also they never increment. I tried writing a big file and checking the numbers but they remain the same.

[0-wiz-0]

Seems I was late to the party. Sorry.
Thanks for fixing it so quickly!

[giampaolo]

In case you want to take it a shot at busy time....
I was trying this:

(long long) PSUTIL_KPT2DOUBLE(stats[i].time)

But I get results which are too low (in the order of the hundreds). I'm not sure how to properly convert that.

[0-wiz-0]

vmstat uses:

        struct timeval   *time; /* Time spent in disk i/o. */
...
        static struct io_sysctl *drives = NULL;
...
        mib[0] = CTL_HW;
        mib[1] = HW_IOSTATS;
        mib[2] = sizeof(struct io_sysctl);

        size = ndrive * sizeof(struct io_sysctl);
        if (sysctl(mib, 3, drives, &size, NULL, 0) < 0)
                err(1, "sysctl hw.iostats failed");
        for (i = 0; i < ndrive; i++) {
...
                cur.time[i].tv_sec = drives[i].time_sec;
                cur.time[i].tv_usec = drives[i].time_usec;

Does that help?

[giampaolo]

Yeah I saw that but it doesn't help. In psutil we have to return milliseconds so we calculate them somehow. This:

#define PSUTIL_TV2DOUBLE(t) ((t).tv_sec + (t).tv_usec / 1000000.0)
PSUTIL_KPT2DOUBLE(stats[i].time);

...looks wrong because the resulting value is too small.