Autonomously scale Compute, Storage
Autonomous features enable automatic scaling in response to changes in workload.
Auto-scale of nodes enables scaling based on load:
- In/Out auto-scaling performs horizontal scaling, decreasing (In) or increasing (Out) the node count.
- Up/Down auto-scaling performs vertical scaling, increasing (Up) or decreasing (Down) the instance size.
Auto-scale of storage enables expansion of capacity based on usage.
Autonomous features can be enabled at time of service launch. Autonomous features can be enabled or disabled after launch.
Enable Auto-Scaling of Nodes
Auto-scaling of nodes can be enabled either at time of service launch or after service launch.
During service launch:
- Check the "Enable auto-scale nodes" checkbox and set the desired scaling parameters.
After service launch, manage Autonomous settings, and enable the desired auto-scaling features.
Enable Auto-Scaling of Storage
Auto-scaling of storage can be enabled either at time of service launch or after service launch.
During Service Launch:
- Check the "Enable auto-scale storage" checkbox and set the desired maximum transactional data storage.
After service launch, manage Autonomous settings, and enable the desired auto-scaling features.
For ColumnStore Data Warehouse, object storage adjusts automatically. Autonomous scaling features for storage are not used by this topology.
Manage Autonomous Settings
To manage Autonomous settings:
- From the Portal, click the "MANAGE" button for the desired service, then choose "Autonomous" from the menu.
- Update settings as desired.
- Click "Apply Changes" when complete.
Scaling Rules
Automatic scaling occurs based on rules.
| Policy | Condition | Action |
|---|---|---|
| Auto-Scale Disk | Disk utilization > 90% sustained for 5 minutes | The disk is expected to run out of capacity in the next 24 hours (predicted based on the last 6 hours of service usage). Upgrade storage to the next available size in 100GB increments Note: you cannot downgrade storage, the upgrade is irreversible |
| Auto-Scale Nodes Out | CPU utilization > 75% over all replicas sustained for 30 minutes. Number of concurrent sessions > 90% over all replicas sustained for 1 hour. Number of concurrent sessions is expected to hit the maximum within 4 hours (predicted based on the last 2 hours of service usage) |
Add new replica or node Additional nodes will be of the same size and configuration as existing nodes |
| Auto-Scale Nodes In | CPU utilization < 50% over all replicas sustained for 1 hour. Number of concurrent sessions < 50% over all replicas sustained for 1 hour |
Remove replica or node Node count will not decrease below the initial count set at launch |
| Auto-Scale Nodes Up | Number of concurrent sessions is expected to hit the maximum within 4 hours (predicted based on the last 2 hours of service usage) | Upgrade all nodes to the next available size |
| Auto-Scale Nodes Down | CPU utilization < 50% over all replicas sustained for 1 hour. Number of concurrent sessions < 50% over all replicas sustained for 1 hour |
Downgrade nodes Node size will not decrease below the initial node size set at launch |
Autonomous actions are not instantaneous.
Cooldown periods may apply. A cooldown period is the time period after a scaling operation is completed before another scaling operation can occur. The cooldown period for storage scaling is 6 hours.