Disk I/O limits
Disk I/O limits
Throttle an instance’s block I/O when it is idle, with the idle.blkio-* and active.blkio-* settings. Use it when an idle workload should keep running but must not compete with active workloads for disk bandwidth.
flowchart LR
active["Active<br/>blkio-weight 500"]
idle["Idle<br/>blkio-weight 100"]
active -->|session expires| idle
idle -->|new request| active
| Setting (idle / active) | Format | Example |
|---|---|---|
idle.blkio-weight / active.blkio-weight | Integer 10–1000 | 100, 500 |
idle.blkio-weight-device / active.blkio-weight-device | path:weight list | /dev/sda:100 |
idle.blkio-device-read-bps / active.blkio-device-read-bps | path:rate list | /dev/sda:10m |
idle.blkio-device-write-bps / active.blkio-device-write-bps | path:rate list | /dev/sda:10m |
idle.blkio-device-read-iops / active.blkio-device-read-iops | path:iops list | /dev/sda:100 |
idle.blkio-device-write-iops / active.blkio-device-write-iops | path:iops list | /dev/sda:100 |
blkio-weightis a relative scheduling weight from10to1000. Values outside that range, or0, are ignored.- Rates use byte units in powers of 1024:
10mis 10 MiB/s,100kis 100 KiB/s. - IOPS values are plain integers.
- Per-device values are
path:valuepairs, separated by commas. - All of them need
idle.replicasat1or more, and you should set theactivevalue of every limit you set onidle. See Scale instead of stop.
Your provider page shows how the limits are applied, and their requirements.



