Some recent stats again!
As mentioned above, on 2025-09-17, I switched the nodes to run with a lowered minrelayfee. That’s also visible in both the reconstruction rate getting better and having to request a lot less data from peers for the cases where we can’t reconstruct. Bob was running with lowered a lowered minrelayfee since 2025-08-22. In hindsight, it would probably have been better to not update all nodes and keep a few on v29 to see how big the difference is (but these nodes aren’t solely for measuring block reconstruction, also for testing release candidates, which I felt was more important).
I’ve also been running a v29.1 Bitcoin Knots node nico with the default configuration. This node doesn’t include the logging added in https://github.com/bitcoin/bitcoin/pull/32582, so no data on requested transaction size.
I haven’t looked into what could caused the low reconstruction rates on bob, dave, luke and nico on 2025-10-18.