Which property of Kafka most directly enables 'replay yesterday'?

Records are durably retained in the log independent of consumption

Consumers acknowledge each message individually

Records are durably retained in the log independent of consumption

Producers wait for every consumer before sending

Brokers delete records as soon as one consumer reads them

Why Streaming — Semantic Web Academy

What streaming buys you

Stop polling, start streaming

A request/response API answers what is true right now?. A streaming platform answers what just happened, in order, and who else needs to know?. The shift from periodic batch ETL to event streams unlocks:

sub-second analytics dashboards,
event-sourced microservices (the log is the source of truth),
CDC into search / vector / graph indexes,
fan-out integration without point-to-point spaghetti.

Apache Kafka is the de-facto partitioned, replicated commit log that made this affordable.

Phone calls vs the PA system

Polling a REST API every minute is like phoning the warehouse every minute to ask 'anything new?'. A Kafka topic is the PA system: the warehouse announces every shipment exactly once, everyone who cares is already listening, and a recording is kept so a newcomer can replay yesterday's announcements.

Kafka landscape map

LayoutLabelsClick a node to focus its neighbourhood · drag to pan · scroll to zoom

client
core
platform
ops

The Kafka ecosystem at a glance — clients, core, platform and ops.

Reflect

Pick one polling integration in your stack.

›What's the smallest replay window that would have prevented your last incident?
›Which downstream system is your worst point-to-point coupling, and why?
›If the source emitted events instead, how many of today's batch jobs disappear?

Reading in progress · 0 of 1 activity done