Thanks for the KIP, José. Given that we already have metadata.log.max.record.bytes.between.snapshots, we need to define how the two properties interact. I would expect that the time-based property would take effect only if the bytes-based property did not trigger. In other words, if you are regularly generating snapshots every hour because of your setting for metadata.log.max.record.bytes.between.snapshots, setting a time-based property for every 10 hours should have no effect. But if you would have only generated a snapshot every week by your metadata.log.max.record.bytes.between.snapshots setting, setting a time-based property for every 10 hours should result in a snapshot every 10 hours. And in that case the bytes-based property is effectively ignored.
I think we should clarify this in the KIP to avoid confusion. Also, it suggests that the configuration name should include "max" somewhere to indicate that it is a maximum only and not necessarily the duration value that will result from the sum total of all configurations. So maybe metadata.log.max.snapshot.interval.ms? best, Colin On Tue, Oct 11, 2022, at 12:15, José Armando García Sancio wrote: > Hey all, > > I am interested in allowing brokers and controllers in KRaft to > generate snapshots for the cluster metadata partition on a timely > basis. This would better allow Kafka users to use cluster metadata > snapshots as a solution for backing up the cluster's metadata. > > Let's use this thread to discuss KIP-876: > https://cwiki.apache.org/confluence/x/MY3GDQ > > Thanks! > -- > -José