Open eyusim opened 4 years ago
in addition 2.x increased the load on our servers significantly, which would have required us to use much more compute resources. we ended up rolling back to the latest 1.x version
I was surprised by this as well. Do I understand correctly that passing an instance of PollingConfig
to the Scheduler
should disable the new fan-out behavior and eliminate additional Kinesis costs?
The defaults in the 2.x configuration to use Fan-out and detailed Cloudwatch metrics can be very surprising when migrating from 1.x, which does not do either of those things. We noticed a significant rise in Kinesis and Cloudwatch costs (we use datadog to pull custom metrics and didn't expect to all of a sudden have 100s of thousands more metrics)
These things should be spelled out as warnings in the docs (https://docs.aws.amazon.com/streams/latest/dev/kcl-migration.html)