First of all, middleware for log collection can be divided into Push type and Pull type according to the data acquisition method.
Systems that transfer logs to a destination for which middleware is configured are classified as Push. Famously speaking, Apache Flume, Fluent, Scribe, Logstash etc are classified as Push type.
On the other hand, it is Pull type that the client side explicitly specifies and acquires necessary data for middleware that centrally manages logs. Apache Kafka and so-called PubSub models such as Google Cloud Pub/Sub and Kinesis Stream, which are not middleware, are classified as Pull types.
—
Source => Push type middleware that receives records. Corresponds to input in Fluentd.
Channel => An area to save until transmission of a record is completed in Push type middleware. Equivalent to buffer in Fluentd.
Sink => Push destination of record in Push type middleware. Equivalent to output in Fluentd.
Upstream => The side that sends the record.
Downstream => The side that receives the record.