how are you guys handling dropped transactions when the mempool gets chaotic?
been messing around with pending transaction tracking across a few EVM chains and one thing that keeps annoying me is how differently nodes seem to behave once activity suddenly spikes
a tx can sit pending looking completely normal, then disappear from one node’s mempool while another provider still reports it. replacements make it even messier because now you’re trying to figure out whether the original tx was actually dropped, replaced with the same nonce or your provider just stopped propagating it
right now I’m basically keeping a local state for every pending tx and comparing multiple RPC streams before marking anything as dead, but it feels heavier than it should be
for anyone running mempool infrastructure at decent volume, how long do you keep a missing transaction around before considering it dropped?
also curious if people trust a single low latency provider for this or if multiple node views are basically mandatory once you care about what happens before confirmation