Buffer analytics — batching events

Pattern

Telemetry — clicks, views, micro-events. Hundreds per minute. Sending each as its own HTTP request is wasteful for the network and the backend. The fix: accumulate into a batch for 30 seconds, send it in one request.

The problem it solves

An imperative array + setInterval breaks quickly: you forget to flush on destroy, you can't handle send errors (what do you do with a lost batch?), and it tests poorly.

Operators and why they matter

• bufferTime(N) — accumulate events over an N-ms window and emit an array.
• filter(batch => batch.length > 0) — don't send empty batches.
• mergeMap — send the batch as an HTTP request.

Gotchas

• bufferTime on a long-lived source without unsubscribe — the timer runs for the whole life of the service. That's fine, but on page hide you must flush explicitly.
• concatMap instead of mergeMap — if batch order matters. Otherwise mergeMap is faster.
• Don't forget a batch-size limit. If 10,000 events pile up in 30 seconds, split them across several HTTP requests rather than one giant payload.