Pool overloading in benchmark

Question

I am working on substrate and trying to benchmark my node. I'm sending unsigned extrinsics to the pool and implemented the ValidateUnsigned trait where I add priority, longevity, provide and require tags. The benchmark launches about 100k extrinsics, using the author_submitExtrinsic endpoint. While sending these extrinsics, I check the status of the pool by querying the author_pendingExtrinsics endpoint.

After about 14k transactions have entered the pool, the amount of extrinsics stagnates, with sometimes ~100 extra extrinsics entering. Also looking at the extrinsic data, it looks like even after 4-5 blocks, although transactions are being executed (I see extrinsics being added to blocks in the logs), it seems like no extrinsics have been removed from the pool and I start getting nonce errors (which confirmes that they have been executed but are now being reexecuted?).

What I tried so far:

• confirming that sending a lower amount of extrinsics (10k) allows them to be processed correctly and that they are removed from the pool
• replacing my implementation of the ValidateUnsigned yields the same result

Answer 1

So I can't be 100% sure of the issue, my guess is that my validate_unsigned logic was too heavy, causing the pool to spend too much time validating and preventing it from pruning executed transactions. Reduced the workload on the validate_unsigned and moved most of this heavy logic to pre_dispatch, might not be a perfect solution, if anyone has some better idea would love to hear it !