Feature Request: criticality-aware transaction throttler

[ejortegau]

Feature Description

Context

The transaction throttler allows throttling/rejecting queries based on replication lag. When it thinks that the rate at which requests are coming in is too high, it immediately fails the query with a RESOURCE EXHAUSTED error.

Feature request

The transaction throttler uses criticality information to decide whether a request should be throttled. When the lag and request rate is such that throttle is necessary, It will also consider that criticality information, ranging from should not throttle at all, to should throttle always.

See dicussion in #feat-handling-overload.

Use Case(s)

This allows to provided a form of differentiated services guarantees. For example, requests being issued by end user facing web API can be marked as high criticality and always get served, whereas background jobs or other internal, less important requests can be failed. This allows to mitigate the impact of throttling.