Add IO stats and ext4 FS stats through new ext4 collector

[john-morales]

This PR attempts to revive Issue #3005 / Pull Request #3295 via implementing support for ext4 and fs error statistics through use of the procfs library.

Work here builds on the progress from #3295. I'm unclear myself if/how to properly attribute @mshahzeb's efforts, but happy to do so.

Re-iterating the original goals, this PR implements support for:

# HELP node_disk_iodone_total Number of completed or rejected IO commands.
# TYPE node_disk_iodone_total counter
node_disk_iodone_total{device="sda"} 775
node_disk_iodone_total{device="sr0"} 1.29433517e+08
# HELP node_disk_ioerr_total Number of IO commands that completed with an error.
# TYPE node_disk_ioerr_total counter
node_disk_ioerr_total{device="sda"} 11
node_disk_ioerr_total{device="sr0"} 41

...from:

/sys/block/<disk>/device/ioerr_cnt: number of SCSI commands that completed with an error
/sys/block/<disk>/device/iodone_cnt: number of completed or rejected SCSI commands

And implements support for:

# HELP node_ext4_errors_total Number of ext4 filesystem errors.
# TYPE node_ext4_errors_total counter
node_ext4_errors_total{partition="nvme0n1p5"} 0
# HELP node_ext4_messages_total Number of ext4 filesystem log messages.
# TYPE node_ext4_messages_total counter
node_ext4_messages_total{partition="nvme0n1p5"} 2
# HELP node_ext4_warnings_total Number of ext4 filesystem warnings.
# TYPE node_ext4_warnings_total counter
node_ext4_warnings_total{partition="nvme0n1p5"} 0

...from:

/sys/fs/ext4/<partition>/errors_count: number of ext4 errors
/sys/fs/ext4/<partition>/warning_count: number of ext4 warning log messages
/sys/fs/ext4/<partition>/msg_count: number of other ext4 log messages

The new ext4 collector is enabled by default. It also has support for (mutually exclusive) partition-exclude and partition-include regexp flags, which seemed to be a standard practice just from my looking around at other disk-related collectors. I also elected to make the default partition-exclude equal to ^features$, to intentionally avoid the built-in "features" directory at /sys/fs/ext4/features. It is not a real partition, and does not contain the 3 errors_count, warning_count, msg_count metrics the collector emits.

Corresponding procfs change that enabled this work: prometheus/procfs#651

Thank you for your consideration!

[anarcat]

given that sda is a device in node_disk_ioerr_total, perhaps sdb1 in node_ext4_warnings_total should be a partition?

[john-morales]

Good point - will do.

[john-morales]

Updated ext4 metrics label from device to partition. For consistency, also updated the corresponding command line flags from device-exclude / device-include to partition-exclude / partition-include. (Edited PR description above as well.)