I don't think that's true/maybe that comment is misleading. Tombstones AFAIK will be propagated by hints, and the hint system doesn't do anything to check if a particular row has been tombstoned. To the node receiving the hints it just looks like it's receiving a bunch of writes, it doesn't know they are hints.
On 12 April 2018 at 13:51, Jinhua Luo <luajit...@gmail.com> wrote: > Hi All, > > In the doc: > https://docs.datastax.com/en/cassandra/3.0/cassandra/dml/dml > AboutDeletes.html > > It said "When an unresponsive node recovers, Cassandra uses hinted > handoff to replay the database mutationsthe node missed while it was > down. Cassandra does not replay a mutation for a tombstoned record > during its grace period.". > > The tombstone here is on the recovered node or coordinator? > The tombstone is a special write record, so it must have writetime. > We could compare the writetime between the version in the hint and the > version of the tombstone, which is enough to make choice, so why we > need to wait for gc_grace_seconds here? > > --------------------------------------------------------------------- > To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org > For additional commands, e-mail: user-h...@cassandra.apache.org > >
