[collectd] disk_time broken for Linux

[Thorsten von Eicken]

I've been looking into the disk plugin disk_time metric under Linux and
it looks to me like it's fundamentally broken. The metric is defined as
DERIVE and, for example, if HAVE_IOKIT_IOKITLIB_H the values generated
come from something called
kIOBlockStorageDriverStatisticsTotalReadTimeKey, which makes sense to
me; and if HAVE_LIBKSTAT then there are calls to get KIO_RTIME and
kio.KIO_WTIME, again sounds plausible. But under Linux, the code very
clearly calculates the average time spent per I/O read and write, which
is nice, but which must be represented as a GAUGE, not DERIVE. Am I
missing something here or is it indeed broken?
Thanks!
Thorsten