You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The route-propagation logic is elegant in the sense that if a new server starts up, it can reach out to any peers with receiveRoutes=true and ask them for routes.
However, the system is complicated, and doesn't account for connections that hang -- e.g., in order to "receive routes", a connector has to send a "route control request" to the peer. If the peer doesn't send route updates after some amount of time, the connector should retry.
The text was updated successfully, but these errors were encountered:
For more context here: the way any given peer (e.g., route requestor) requests routes is that it reaches out to its peer (route-propagator) and asks it to start sending route updates. In the Java connector, the propagator emits route updates on a timer, so retry should not be necessary assuming the propagator has not crashed.
In those cases, however (i.e, the propagator crashes and restarts), then route-updates will no longer come to the requestor. In these cases, the requestor should detect this condition and send a new route-control message to trigger the propagator to start sending again.
(it's possible that this will simply happen via the scheduler on the requestor -- if that's the case, then this ticket can be closed).
The route-propagation logic is elegant in the sense that if a new server starts up, it can reach out to any peers with
receiveRoutes=true
and ask them for routes.However, the system is complicated, and doesn't account for connections that hang -- e.g., in order to "receive routes", a connector has to send a "route control request" to the peer. If the peer doesn't send route updates after some amount of time, the connector should retry.
The text was updated successfully, but these errors were encountered: