Ignore fees per channel

[rustyrussell]

We always had a global ignore-fee-limits: this adds an equivalent on a per-channel basis. This is a workaround, of course: letting peers set arbitrary fees is fraught, but there are cases where this can help. For example, there seems to be a problem where LND nodes insist on retransmitting the same fee, even if it knows it's now wrong (e.g. neutrino, or bitcoind restart). We will keep closing the connection on them.

With this, you could temporarily setchannel id=<channel> ignorefeelimits=true and it will hopefully make progress. You can also use it to "trust" peers, and not get upset if there are fee disagreements.

[vincenzopalazzo]

If you have time to rebase it, I will give a shot for the review today :)

[rustyrussell]

If you have time to rebase it, I will give a shot for the review today :)

Done, thanks!

[wtogami]

https://fedorapeople.org/~wtogami/a/2023/0001-DO-NOT-COMMIT-Temporary-workaround-for-LND-update_fe.patch
Temporary workaround for v23.05.2

Your setchannel approach of runtime switchable ignore fees is far superior to temporary hack that requires restarting CLN.

Would it be possible to add a print to let you know every time an ignore fee event happened?

What I like about my ugly hack is the print tells you exactly what is happening. It is diagnostically useful to know because it should happen JUST ONCE per channel to recover from LND's bug after their fee estimator was broken in the past. If it happens more than once the remote peer's fee estimator is currently broken. You want to know that so there is a chance to talk to the sysadmin to be able to fix it. It appears this PR currently would be silent when it ignores which is less informative?

[vincenzopalazzo]

The critical logic LGMT but I had some comments from the API point of view.

I was wondering if the following CI failure is related?

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=========================== short test summary info ============================
ERROR tests/test_opening.py::test_anchor_min_emergency - ValueError: 
Node errors:
 - lightningd-1: had bad gossip messages
Global errors:
======= 59 passed, 3 skipped, 4 warnings, 1 error in 2441.54s (0:40:41) ========

[vincenzopalazzo]

ACK 7ec9faa

[rustyrussell]

https://fedorapeople.org/~wtogami/a/2023/0001-DO-NOT-COMMIT-Temporary-workaround-for-LND-update_fe.patch Temporary workaround for v23.05.2

Your setchannel approach of runtime switchable ignore fees is far superior to temporary hack that requires restarting CLN.

Would it be possible to add a print to let you know every time an ignore fee event happened?

What I like about my ugly hack is the print tells you exactly what is happening. It is diagnostically useful to know because it should happen JUST ONCE per channel to recover from LND's bug after their fee estimator was broken in the past. If it happens more than once the remote peer's fee estimator is currently broken. You want to know that so there is a chance to talk to the sysadmin to be able to fix it. It appears this PR currently would be silent when it ignores which is less informative?

This may be possible, but the current PR works by telling channeld an infinite allowable range. I think we can do logging in lightningd however, if the actual feerate used is weird.

I will do this as a separate PR.