Reconnection strategy

[inza4]

Hi, I am trying to make my nodes to be connected even if they disconnect due to no activity. The set of nodes in the system and adresseses are known and fixed, for the moment.

My first approach was upon a Connection closed event, I dial again, but this doesn't works always. I understand the "connection" state machine is more complex than that.

Should I dial before trying to send a request response message? When I see the disconnect? Should I schedule a dial in an exponential backoff strategy? I found a method called "add_external_address", should I call it only once? It looks like nodes "forget" the address of peers.

The question is wide, but I want to understand how this works and what are the possibilities. Thanks in advance.

[elenaf9]

I am trying to make my nodes to be connected even if they disconnect due to no activity

If the nodes should stay connected unconditionally, you can set swarm::Config::with_idle_connection_timeout to a high value / Duration::MAX.

I found a method called "add_external_address", should I call it only once? It looks like nodes "forget" the address of peers.

external_addresses refers to your own external addresses.
Peer addresses can be added with add_peer_address. Internally that notifies each behavior of a new known peer address. If you have, e.g., identify included in your behavior it will then handle the event and persist the peer address, so it is know when you do the next dial.

[inza4]

Thanks for the clarifications. What if nodes can leave and rejoin the system?

[elenaf9]

What do you mean exactly? How to reconnect to a node after it rejoins?

[inza4]

Yes, in the scenario where I can't define an "infinite" timeout time, I want to be able to reconnect to failed peers that are again online. When is the right time to dial them? should I do something else?

[elenaf9]

That depends on your application. If nodes are just very briefly offline because they, e.g., just restart then you can try to immediately dial them again after a connection closed event. You just have to make sure that the nodes have a persistent PeerId and listening address.