mailbox+gbn: conn fixes

lightninglabs / lightning-node-connect

MIT License

79 stars 22 forks source link

mailbox+gbn: conn fixes #30

Closed ellemouton closed 2 years ago

ellemouton commented 2 years ago

ensure we dont hang on channel reads
add and use gbn read & write deadlines
add separate pong timeout
wrap ClientConn with a Client so that proper refresh can happen
add itest for server reconnect

ellemouton commented 2 years ago

need to rework/add to this today now that the infra fix is in. Will ping here when ready 👍

ellemouton commented 2 years ago

Ok @guggero , this is ready now. The main change is in the mailbox: ensure proper client refresh commit.

There is a race condition being caught that is as a result of us using ProxyConn at the same time that the grpc library is using it.... any ideas how to fix this?

ellemouton commented 2 years ago

ok, can confirm this race is not added in this pr. it does happen on master, just happens more here since there are way more reconnect tests here. Still havent found the issue. But the symptoms are that it either happens in the client reconnect test in the ServerHandshake function or it happens in the 'server reconnect' test in the ClientHandshake function. Looks like grpc calls the Close() method on the connection we use in the handshake functions at the same time....