We have hardcoded kinsumer's leaderActionFrequency (which controls how often we poll a stream for scaling actions) and shardCheckFrequency (which controls how often each client propagates a change detected by the leader action) both to 10 seconds.
For larger streams this is problematic, as the app is too sensitive to scaling actions, and rebalancing clients too frequently can cause latency.
This should be configurable in snowbridge, and the default value for leaderActionFrequency should be change to a more sensible value in the order of minutes, not seconds.
We have hardcoded kinsumer's
leaderActionFrequency(which controls how often we poll a stream for scaling actions) andshardCheckFrequency(which controls how often each client propagates a change detected by the leader action) both to 10 seconds.For larger streams this is problematic, as the app is too sensitive to scaling actions, and rebalancing clients too frequently can cause latency.
This should be configurable in snowbridge, and the default value for leaderActionFrequency should be change to a more sensible value in the order of minutes, not seconds.