lnd force-closes in response to warning messages

[TheBlueMatt]

A while back we added warning messages to the spec with the purpose of having a way to signal to nodes that something has gone wrong but that we do not want to force close any channels.

We started using warning messages to tell peers when we ignored a gossip message because it was bogus, with the message "Unreadable/bogus gossip message".

However, it looks like lnd interprets warning messages as critical errors and responds by force-closing channels - ignoring the entire purpose of warning messages?

A peer of mine pulled this from their log:

2022-08-06 12:14:48.864 [ERR] HSWC: ChannelLink(0db90ff58728d7698dbf9e478eca33109b9777898acf28b08bd9352a75a7a74c:0): failing link: ChannelPoint(0db90ff58728d7698dbf9e478eca33109b9777898acf28b08bd9352a75a7a74c:0): received error from peer: chan_id=0000000000000000000000000000000000000000000000000000000000000000, err=Unreadable/bogus gossip message with error: remote error

On my end I just received an error message that said "remote error".

[TheBlueMatt]

I've now seen this twice, the second time the peer was running "master... something between 0.15 and 0.15.1 115b041"

[C-Otto]

I'm said peer. Let me know if you need further information.

[positiveblue]

@TheBlueMatt just checking: the type for those messages is set to 1?

[TheBlueMatt]

Yes, the type bytes written on the wire is pulled directly from https://github.com/lightningdevkit/rust-lightning/blob/main/lightning/src/ln/wire.rs#L308

[positiveblue]

I will try to replicate it locally but for the logs it looks like the message was really parsed as error and not a warning
@C-Otto do you do a quick search to see if you have any logs matching received warning message from peer for this or any other peer?

[C-Otto]

I can't find a single entry in the recent logs (~1 month).

[positiveblue]

I tried to replicate it, unsuccessfully, in the main branch + added a test for it #6805

I noticed that it is tagged for v0.16.0 BUT if @C-Otto is running master it should include those changes. I am not sure what is the process for tagging a version (v0.15.1 for example) that not includes the latest changes in master though (maybe we manually cherry pick the commits that need to be included?).

@Roasbeef 115b041 should include those changes right?

[C-Otto]

I guess you mean 99828b8? That's included in 115b041.

[C-Otto]

I can't find a single entry in the recent logs (~1 month).

I just realized that I've only started using the new code yesterday, which also means that older logs cannot contain this new log entry.

[positiveblue]

Did you started using the new code yesterday AFTER the force close? If so this solves the case. Next warning message should not trigger any force close/channel failure.

[C-Otto]

Did you started using the new code yesterday AFTER the force close? If so this solves the case. Next warning message should not trigger any force close/channel failure.

That thought also crossed my mind :) But no, I triple checked. I first started with the new code, then the quoted log message appeared.