Rishabh Saraswat created CASSANDRA-21207:
--------------------------------------------
Summary: CEP-59: Drain in-flight requests on subscribed
connections before shutdown
Key: CASSANDRA-21207
URL: https://issues.apache.org/jira/browse/CASSANDRA-21207
Project: Apache Cassandra
Issue Type: Sub-task
Components: Messaging/Client
Reporter: Rishabh Saraswat
Assignee: Rishabh Saraswat
Depends on: CASSANDRA-21206
After emitting GRACEFUL_DISCONNECT, the server now waits for in-flight
requests to complete before proceeding with drain. This ensures clients
can finish processing their outstanding requests without client-visible
timeouts or errors.
The draining logic respects the following timeouts:
- graceful_disconnect_grace_period_ms — time given to clients to stop
sending new requests after GRACEFUL_DISCONNECT is emitted
- graceful_disconnect_max_drain_ms — hard timeout after which connections
are force-closed regardless of in-flight requests
Flow:
1. GRACEFUL_DISCONNECT event emitted to subscribed connections
2. Wait grace_period_ms for clients to stop sending new requests
3. Wait for all in-flight requests to complete on subscribed connections
4. If max_drain_ms exceeded, force-close remaining connections
5. Proceed with drain(false)
Non-subscribed connections continue with existing behavior. When both
subscribed and non-subscribed connections exist, all connections continue
processing requests until gracefully-draining connections are closed or
timed out.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
