Agressive Channeld Restart Testing

[ddustin]

Repeatedly STFU and then TX_ABORT a channel to trigger the socket/fd flake issue.

Changelog-None

[rustyrussell]

This ran 50 times in a row for me without any error?

[ddustin]

It passed through CI no problem too -- it must be a Mac only problem

[cdecker]

Shall we merge this, or would you like to dig deeper and see what is going wrong with Mac? In the latter case I'd remove the milestone.

[ddustin]

Seems net positive to have it merged into the CI as a regression test

[ddustin]

Although this does bundle in the stfu and abort RPC commands which @niftynei had some design objections to in my Splice Script PR 🤔.

#6980

[rustyrussell]

Weird one. You wrote this batch API and only ever use it for one channel. Maybe just have a simpler, single-channel API? Seems clearer, too...

[rustyrussell]

These are definitely dev commands!!

[rustyrussell]

And since you only ever call it with one channel, dev_quiesce would work.

[ddustin]

🤔 Discussing the trade offs of exposing stfu / abort in this comment over here: #7083 (comment)

[rustyrussell]

You don't need these wrappers, esp. for dev commands. It automatically handles unknown RPCs (we mainly add them so you get nicer documentation and error detection).

[ddustin]

Very cool!

[ddustin]

Weird one. You wrote this batch API and only ever use it for one channel. Maybe just have a simpler, single-channel API? Seems clearer, too...

The bulk commands were needed & implemented in #6980, this PR should remove its implementation and instead depend on #6980

[ddustin]

I removed the stfu_channels and abort_channels RPCs with the idea we can rebase off master once #6980 is merged.

I see two ways forward for these commands:

1. Expose low level splice commands to user: stfu_channels, abort_channels, splice_init, splice_update, splice_signed etc
2. Hide all these low level commands from user and expose a singular simple RPC like the dev-splice script command.

Doing something in between these two options doesn't make sense because a user doing a cross-splice of two channels needs a way to stfu both channels and potentially abort them. Prior to #6980 there was no way for a user to do so because splice_init entered stfu automatically.

[rustyrussell]

I removed the stfu_channels and abort_channels RPCs with the idea we can rebase off master once #6980 is merged.

I see two ways forward for these commands:

1. Expose low level splice commands to user: `stfu_channels`, `abort_channels`, `splice_init`, `splice_update`, `splice_signed` etc

I think this is the way. Though make sftu/abort singular (you can simply call it multiple times).

[rustyrussell]

Trivial rebase...