Currently, we publish rangefeed closed timestamp updates every 200ms. This is controlled by two settings:

• kv.rangefeed.closed_timestamp_refresh_interval = 0 (defaults to kv.closed_timestamp.side_transport_interval)
• kv.closed_timestamp.side_transport_interval = '200ms'

For the purposes of rangefeeds, this is overly aggressive, and can lead to excessive load with large numbers of ranges (see #96395). The target duration kv.closed_timestamp.target_duration = '3s' is one order of magnitude larger than the refresh interval, and the default changefeed min_checkpoint_frequency = '30s' is two orders of magnitude larger.

We should increase the default of either kv.rangefeed.closed_timestamp_refresh_interval or kv.closed_timestamp.side_transport_interval to e.g. 1 second, which is still sufficient for rangefeeds, and can significantly reduce CPU usage on nodes with many replicas.

Preferably kv.closed_timestamp.side_transport_interval, since users may already have set this higher (e.g. to 30 seconds), and defaulting kv.rangefeed.closed_timestamp_refresh_interval to e.g. 1 second instead of the side transport interval can cause significant load increases when the effective default changes from 30 seconds to 1 second.

Jira issue: CRDB-30568

Epic CRDB-26372