Cluster Mempool RBF Thoughts

[instagibbs]

@ajtowns

But what if you didn’t see C2, C3, C4 or C5, but had seen P11…P99? C6 alone would look like it was merging 90 clusters, which would presumably be more than the allowed cluster limit. So the rbf from C1 to C6 becomes path dependent: doing it directly fails, but doing it indirectly works.

Yeah that’s kind of annoying in how order-dependent it is, without any external actions by griefing parties.

Another metric could be number of total txs in effected clusters.

f.e. We continue to evaluate an RBF if the number of effected clusters is <= 100 AND number of total effected transactions in effected clusters is <= 2,500.

Would allow “simple” CPFP patterns like @ajtowns writes above up to cluster limits, and still bound the diagram checks? In adversarial case, someone could still package RBF/CPFP ~25 separate clusters, assuming each thing being bumped was part of a max size cluster.