A stream-processing job calculates five-minute activity totals. Mobile clients can upload events several minutes late. Totals must follow the event timestamp rather than the server arrival time. Which design is most appropriate?

Published
  1. Use processing-time windows only and assume network delays do not affect the totals.
  2. Use event-time windows with a suitable watermark and late-event handling policy in a stateful stream processor.
  3. Discard every record whose timestamp differs from the current server clock by any amount.
  4. Sort only the final S3 filenames alphabetically and treat that as event-time ordering.

Comments — 0