Does Kinesis guarantee delivery?

Question

I am searching the documentation of Kinesis Data Streams but I can't find a clear statement like

Kinesis guarantees at least once delivery

From the producer side I expect that a message sent, gets propagated to more than one nodes (something like Kafka's ack=all) From the consumer side I am expecting something equivalent to Kafka commit offset on successful processing from the consumer, or something like Google Cloud's Pub/Sub message acknowledgement.

Is there a submit message guarantee for Kinesis? Is there a processing guarantee for Kinesis (mark message as read only if processed and acknowledged that it was processed)

user_10186808 · Answer

From the comments, if you are really after server-less approach, why not use Lambda Event source mapping to consume Kinesis messages?

Lambda/AWS will poll for messages, batch them up and call our lambda function for a batch of records(batch size as small as 1).
If Lambda function returns success, batch will be committed and function will be called for next batch.
If Lambda function returns failure, batch can be retired.
Unlike Kafka consumers where we need to handle retrying exceptions separately with another kafka topic or a database, to keep consumer moving forward, AWS Lambda supports asynchronous retires as well.

For a simple use case, Kinesis + Lambda is as good as Kafka + Consumer.

user_1943856 · Answer

Based on the documentation consumers from Kinesisread the entire data from the Stream. Thus consumers are responsible to commit their history and pick up from where they left. https://docs.aws.amazon.com/streams/latest/dev/building-consumers.html

Does Kinesis guarantee delivery?

2 Answers