Nice, this is a good reliability improvement and prevents users from having to implement their own out-of-band discovery/drain mechanism.
+1 on the CEP in general. Just a few public interface questions/comments (doesn't block the CEP imo, just cleaner interfaces imo), happy to discuss async or on slack too: * Do we have a reason to diverge from HTTP's standard GOAWAY nomenclature (used by both http2 and http3 for this purpose)? * The cassandra.yaml options could match the other native_transport options and skip the unit suffixes, e.g. these could slot right under native_transport_idle_timeout and perhaps be called native_transport_drain_period instead of graceful_disconnect_grace_period_ms? * Enabled flag plus a duration are likely redundant, a duration of -1 could mean disabled (and 0 means immediately close, the current behavior). Overall excited for more resilient clients! -Joey On Wed, Mar 4, 2026 at 3:16 PM Jane H <[email protected]> wrote: > Hello all, > > I would like to call a vote for CEP-59: Graceful Disconnect. > > Proposal: > https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=406619103 > > Discussion thread: > https://lists.apache.org/thread/wyswxql0zhj10nthn8pmjz40v03qfd5v > > Voting guidelines > https://www.apache.org/foundation/voting.html > > The vote will be open for 72 hours (longer if needed). > Please let me know if you have any questions or need more time to consider. > > Thank you, > Jane >
