-
Improvement
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
For Multi-Rail nodes, as long as a peer has not reached the uptodate state, e.g., push not ACKed or not yet timed out/failed, any message destined for that peer will be held back entirely on the `lp_dc_pendq` queue, regardless of which NI the message was originally intended to use.
This restriction is not reasonable in many scenarios. When configuration or link changes occur, the push convergence speed directly affects message processing on the peer. A push failure may cause message timeouts, and in extreme cases, consecutive push failures can further amplify errors and the duration of the fault. For example, when a new NI is added via `net add`, the initial push tends to be sent from the new NI, but the new NI's link may not have converged yet. Even though the old NI link is still healthy, messages on the peer will be held up, leading to timeouts.
This restriction can be improved to some extent by decoupling the discovery ping and push mechanisms a bit.