A million view events a second must update per-video counts. Why aggregate in the stream before writing to a database?

Answer

The stream job sums views per video in memory and writes one update per video per window (say every 10 seconds) instead of one per event. Writes drop by orders of magnitude, and hot videos no longer hammer a single row. The counts lag by up to one window.

Review this in your daily deck → All cards in Streaming