Alexander Witte
Sep 8, 2018 · 1 min read

Hey Yan thanks for the great articles!!

Would love to get your thoughts on this if possible:

We have a pipeline that uses lambda to consume events from kinesis streams. The lambda function writes the events to DynamoDB. At peak times while we wait for the DynamoDB capacity to autoscale up, we experience write throttling and lose records. I was thinking to catch this error, send to SNS as you describe to retry the event and then finally (if unsuccessful) send to a DLQ that could ship that event to SQS to retry again.
Or perhaps I skip SNS and just write the record to SQS in the first place…

What do you think?