Major Description
When `compression.type` is set to `producer`, we would accept produce request and use its encoded compression to apply to the logs; otherwise we would recompress the message according to the configured compression type before appending. There are pros and cons to recompress the data: you pay more CPU to recompress, but you reduce the storage cost.
In practice, if the incoming produce requests are not compressed, then compressing before appending maybe more beneficial, otherwise just keep them as if `producer` config maybe better. Adding a metric to expose the incoming requests' compression in percentage would be a helpful data point to help operators selecting their compression policy.
Attachments
Issue Links
- links to