Fix test_peer_anchor_push error due to variable signature length

[Andezion]

(Fixes #8493)

Problem

The test_peer_anchor_push test was failing with AssertionError when the actual feerate exceeded the expected by more than 2 sat/kw. This occurred because ECDSA signatures have variable length, typically 71-72 bytes, but can be shorter. A shorter signature means less transaction weight, which bumps up the feerate even though the fee stays the same!

Solution

• Increased tolerance in check_feerate() from 2 to 10 sat/kw to account for signature length variability!
• Improved did_short_sig() detection by increasing wait time from 1 to 2 seconds for more reliable log message detection!

Important

26.04 FREEZE March 11th: Non-bugfix PRs not ready by this date will wait for 26.06.

RC1 is scheduled on March 23rd

The final release is scheduled for April 15th.

Checklist

Before submitting the PR, ensure the following tasks are completed. If an item is not applicable to your PR, please mark it as checked:

• The changelog has been updated in the relevant commit(s) according to the guidelines.
• Tests have been added or modified to reflect the changes.
• Documentation has been reviewed and updated as needed.
• Related issues have been listed and linked, including any that this PR closes.
• Important All PRs must consider how to reverse any persistent changes for tools/lightning-downgrade

[cdecker]

This seems to be unrelated. The signal that we can check is rather fuzzy, being a timeout. It'd be more stable if we had something concrete to wait for, such as a log, or we could poll rather than checking once.

[Andezion]

I increased it from 1s to 2s because the log message "overgrind: short signature length" sometimes appears with a small delay, and 1s wasn't always enough to catch it reliably.

Would it be better to use wait_for() to poll the log instead? Something like:

def did_short_sig(node): return wait_for()

[cdecker]

There is pytest.approx() we could use to check if a number if close-ish to another number. The widening to tolerance=10 seems a bit much, and may actually be covering a real issue. Can you share what you are seeing and why this is addressing it?

[Andezion]

The test failed with - actual_feerate (14006.1) > expected_feerate (14000) + 2, this happens because ECDSA signatures have variable byte length 71-72 bytes, but sometimes shorter. And when we get a shorter signature - the transaction weight decreases - and that causes higher feerate for the same fee!

For example with fee = 32150 sat:
Normal weight 2294 - 32150 * 1000 / 2294 = 14000 sat/kw, but
short sig weight 2293 - 32150 * 1000 / 2293 = 14006.1 sat/kw!

So the 1-2 weight difference can cause 4-8 sat.kw difference, and i've thought that tolerance = 10 can cover this natural difference.

Would pytest.approx(rel=0.001) be better? Or should I calculate the exact tolerance based on expected weight variance?