# TiDB Resource Group RU sizing

Use this topic when the customer asks how to choose or validate a Resource
Group `RU_PER_SEC` value. It establishes a sizing boundary, not production
approval for a particular number.

- `RU_PER_SEC` is the resource group's RU token refill rate. Values such as
  `500` or `600` in product documentation demonstrate syntax and behavior; they
  are not recommended production thresholds.
- Do not recommend a generic `RU_PER_SEC` value from a documentation example.
  Size the group from its observed normal and peak RU demand over
  representative periods, the workload's business priority and required QoS,
  the intended isolation or strict-throttling boundary, and the desired burst
  headroom.
- `BURSTABLE` allows the group to exceed its configured quota while the cluster
  has spare capacity. Without `BURSTABLE`, insufficient tokens can make
  requests wait and eventually time out. Treat this as an explicit isolation
  and latency trade-off, not a default sizing choice.
- Label any initial value as a workload-specific starting hypothesis. Validate
  it with a controlled change while observing the group's RU consumption,
  available RU or throttling/wait signals, and application latency; then adjust
  it rather than presenting the initial number as a safe production answer.
- `CALIBRATE RESOURCE` is available for TiDB Self-Managed capacity estimation
  and is not applicable to TiDB Cloud. For a TiDB Cloud sizing question, use
  applicable Cloud RU-consumption observations and do not instruct the customer
  to run `CALIBRATE RESOURCE`.

Confirm the applicable deployment type and current official documentation.
This topic does not establish incident causality or authorize a production
configuration change.