ddustin
commented
1 year ago ** mutual_splice_locked **
WIRE_ANNOUNCEMENT_SIGNATURES <- l2
l1 -> WIRE_ANNOUNCEMENT_SIGNATURES
l1 <- WIRE_ANNOUNCEMENT_SIGNATURES [A]
WIRE_ANNOUNCEMENT_SIGNATURES -> l2
l1 -> WIRE_ANNOUNCEMENT_SIGNATURES
l1 <- WIRE_ANNOUNCEMENT_SIGNATURES [B]
WIRE_ANNOUNCEMENT_SIGNATURES <- l2
WIRE_ANNOUNCEMENT_SIGNATURES -> l2l1 is getting an extra announcement signatures that was never sent by l2 in these logs. I'm guessing [B] is actually the valid one and [A] is the stale one. It almost looks like it's a queued up announcement_sigs that got sent earlier on, maybe the bug is announcement sigs doing something weird with STFU mode 🤔.
Can you reference the full logs? Would love to see when this extra WIRE_ANNOUNCEMENT_SIGNATURES is sent 👀.
This happened under CI. gossipd complained that the channel_announcement we fed it bad a bad signature, and indeed, I checked l2 signed it with the old scid
103x1x0and l1 signed it with109x1x0. l1 is complaining, so it obviously handed the old signature to gossipd.I can see l1 receiving 3 ANNOUNCEMENT_SIGNATURES from l2 after the splice: is l2 sending the old txid somehow?
This may be a weird corner case caused by all 6 blocks coming in at once: we go from mined to announceable immediately.