Is your feature request related to a problem? Please describe.
Previously, we always update PTS record if the last time we updated pts has passed by changefeed.protect_timestamp_interval and if a checkpoint happens. However, the lastProtectedTimestampUpdate is saved on frontiers, so this info is lost whenever chanegfeed restarts. This could lead to lots of PTS updates if changefeeds keep restarting.
To solve this problem, we introduced another cluster setting to allow PTS updates only if it is lagging behind high watermark by changefeed.protect_timestamp.lag in this PR https://github.com/cockroachdb/cockroach/pull/129442.
We now have two cluster settings to guard how frequent PTS updates happen. It makes things harder to rationalize. We should discuss and see if we should just keep one.
Is your feature request related to a problem? Please describe.
Previously, we always update PTS record if the last time we updated pts has passed by
changefeed.protect_timestamp_interval
and if a checkpoint happens. However, thelastProtectedTimestampUpdate
is saved on frontiers, so this info is lost whenever chanegfeed restarts. This could lead to lots of PTS updates if changefeeds keep restarting.To solve this problem, we introduced another cluster setting to allow PTS updates only if it is lagging behind high watermark by changefeed.protect_timestamp.lag in this PR https://github.com/cockroachdb/cockroach/pull/129442.
We now have two cluster settings to guard how frequent PTS updates happen. It makes things harder to rationalize. We should discuss and see if we should just keep one.
Jira issue: CRDB-41575