feature: Send the disk activity — bytes and operations #43

Blocker: no. This can wait until the agent stack is merged. It goes on top of the pressure fields in the same stack.

Part of contract v0.4.0, section "2026-09-24 — Peaks, pressure and disk activity": "Disk activity". After this issue, the agent conforms to contract v0.4.0.

Defect

• The agent sends how full the disk is. It does not send how busy the disk is.

Cost

• A backup, a database dump or a plugin that writes all the time slows each site. No chart shows it.
• A cloud disk has a limit of operations and of bytes each second. A person cannot see when a server is near the limit.

Benefit of the fix

• Each sample carries the bytes and the operations of the minute, and the peaks of the minute (the peak IOPS).

Fix

1. At each reading, read /proc/diskstats.
2. Count each whole disk that has a hardware device (/sys/block/<name>/device). Do not count partitions, loop devices, ram, zram, dm-* and md*.
3. Count all such disks, not only the disk that holds /.
4. Add the disks that both readings of a window have.
5. Operations are columns 4 and 8. Bytes are columns 6 and 10, times 512.
6. Add the optional fields disk_read_bytes, disk_write_bytes, disk_read_ops, disk_write_ops, and the four peaks disk_read_max_bytes_per_second, disk_write_max_bytes_per_second, disk_read_max_ops_per_second, disk_write_max_ops_per_second.
7. Send null when there is no usable previous reading, when a counter went back, or when no disk has a hardware device.
8. Change the pin line in the README to contract v0.4.0.

Completion checks

• A partition, a loop device and a device mapper disk are not counted.
• A disk that appears or goes away makes no spike.
• The eight fields are null in the first sample and when no disk has a hardware device.
• disk_read_bytes ≤ 60 × disk_read_max_bytes_per_second, and the same for the other three pairs.
• The README states contract v0.4.0.