* Fix how major and minor identifiers of block devices are read.
The current implementation assure that the major and the minor are
coded on one byte. But they are not:
```
brw-rw---- 1 root disk 252, 290 Feb 25 11:36 dm-290
```
290 as minor in this example is over 1 byte.
So after wondering why all my devices iops weren't correctly stored,
I found out that several points were added for some disks. For `dm-290`
it was overriding `252:34`, instead of getting udev stats for `252:290`.
The solution is here:
https://sites.uclouvain.be/SystInfo/usr/include/sys/sysmacros.h.html
The implementation is directly taken from this, fixing my bug.
* Use unix.Major/unix.Minor instead of custom implementation
* [diskio] Force type of stat.Rdev uint64 for mipsle compatibility
For most platforms, stat.Rdev is already a uint64 so this is without any effect
for linux,mipsle, unix.Stat_t.Rdev is a uint32, but the way to compute major and minor doesn't change, casting the uint32 has no impact either
For most platforms, stat.Rdev is already a uint64 so this is without any effect
for linux,mipsle, unix.Stat_t.Rdev is a uint32, but the way to compute major and minor doesn't change, casting the uint32 has no impact either
If an empty column is present in parsed csv data, a `recordFields` is
set to the key of "". If no `MeasurementColumn` is defined, this empty
value was being used as the measurement name. By only setting the
measurementName if the value is not empty, we avoid this error.
Since an empty column is a valid column, skip values must account for
them.
Daniel Nelson
Pierre Tessier
liispon
Olli-Pekka Lehto
Daniel Nelson
John Hu
scier
Soulou
Adrián López
Greg
Dimitri Saingre