We've introduced ExpressionContext::checkForInterrupt() to throttle the actual checks for interrupt on every DocumentSource::getNext(), because they are expensive.

The ExpressionContext::checkForInterrupt() uses ExpressionContext::_interruptCounter as a throttling frequency counter that is shared by all stages of the pipeline.

In the context of separating query execution and query optimization (e.g., SPM-4106) we might reconsider this design and count the frequency either per stage or introduce another class, for instance, StageInterruptThrottler, shared by all pipeline stages to refactor this functionality from already-overloaded ExpressionContext.