Auto scaling overview
Auto scaling raises a database's resources when usage crosses a threshold you define, so you do not have to watch utilization and resize by hand. It is vertical scaling: the same database gets a bigger flavor or a bigger disk, rather than gaining nodes.
Compute and storage scale separately
The two are configured independently and can be used on their own.
| Compute scaling | Storage scaling | |
|---|---|---|
| What grows | vCPU and RAM, by moving to a larger flavor | Data disk capacity |
| Trigger | Average CPU or RAM above your threshold | Storage usage above your threshold |
| How much it adds | The next flavor in a list you order | A percentage of current capacity |
| Service impact | None on most engines. PostgreSQL, OpenSearch, and ClickHouse need a manual restart before the change takes effect | None. Expansion runs online |
| Reversible | Yes, by resizing back down | No. Expansion is permanent |
How the decision is made
Compute scaling samples usage on a tracing interval you set, and compares the average over that interval against your CPU and RAM thresholds. It is the average that matters, not a momentary spike, so a short burst does not trigger a resize.
When the threshold is exceeded, the system moves to the next flavor in your List Flavor Scaling, in the order you listed them. If usage is still above the threshold after that, it moves to the next one. The list is the ceiling — scaling stops at the end of it.
Scaling also stops at your quota. Once the VPC's available quota is consumed, no further scale-up happens regardless of thresholds.
Before you turn it on
A notification recipient has to exist first. The backup service has the same requirement, and the console blocks activation with a dialog until at least one email or Telegram recipient is configured. See Manage alert rules.
Choosing thresholds
- Keep thresholds at 60% or above for compute. Lower values make the system scale on ordinary traffic variation, and each scale-up costs money you do not get back by scaling down later.
- Set storage thresholds at 70% or above. Storage cannot shrink, so a threshold that trips on a temporary spike permanently raises your bill.
- Give the flavor list room. A one-entry list means the database stops scaling at the first step and you are back to manual resizing.
Storage expansion is one-way. There is no automatic or manual path back to a smaller disk, and the larger capacity is billed from then on.
Next steps
- Configure auto scaling for the procedure
- Change database resource configuration to resize manually instead
- Monitor database with FMON to see the utilization trend your thresholds should follow