cscotta commented on PR #4707: URL: https://github.com/apache/cassandra/pull/4707#issuecomment-4189303521
Thanks, was thinking about this a bit. The decay is pretty quick (on the basis of the last 1000 LocalDataResponses generated), which probably helps. But a workload that's 99.9% 128-byte responses and 0.01% 10MB responses would still be degenerate. If the table name/ID were available in this scope, that would make it easier to vary by table - but still not perfect. One simple approach might be to track a histogram of generated response sizes and calculate whether the old behavior or new behavior would be preferable. But I also don't mean to go overkill on it and introduce stats tracking/comparison in every response generation. I may look at deploying this for a variety of workloads with additional instrumentation that measures "better or worse" as well. Interested in your + others' thoughts on this. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
