Failed tx stuck in cache

[schuyler]

I'm working on an application based on basecoin. I attempted to commit two transactions out of order. Tx seq #20 was rejected, then seq #19 was accepted, but now I'm getting ERROR: broadcast_tx_commit: Response error: RPC error -32603 - Internal error: Error on broadcastTxCommit: Tx already exists in cache when I try to re-run #20.

I can't just skip to seq #21 because the backend is waiting for #20.

Restarting the peer fixes the issue.

How do I force this transaction either to be completed or flushed from the peer's cache, without restarting?

(filing this as a ticket per @zmanian's request)

[schuyler]

Incidentally I have mempool.recheck and mempool.recheck_empty both set to true

[ebuchman]

Invalid txs should be removed from the cache - ie the app needs to return non-zero ABCI code.

You can also use --rpc.unsafe to expose the unsafe_flush_mempool endpoint ...

Going to close this, but please open a new issue if you're still having troubles (though it might fit better on Tendermint ...). Thanks!