Add getRecentPrioritizationFees RPC endpoint (backport #27278)

[CriesofCarrots]

#27278 for v1.14

[mvines]

🎉

[CriesofCarrots]

Made it through all CI jobs first try 🕺

[GCrispino]

Sorry to make this comment out of the blue, but I'm trying to understand the implementation of the getRecentPrioritizationFees JSON-RPC method (also created a question in the Solana Stack Exchange page to address it).

The actual question is: at this point why should the cache contain data for write_account_c, since just a transaction that involves write_account_a and write_account_b has been submitted? Shouldn't it return an empty result?

[CriesofCarrots]

at this point why should the cache contain data for write_account_c

The cache doesn't in fact contain any explicit data for that account. If a key isn't found in the cache, get_prioritization_fees returns the minimum fee for any transaction in that slot:

solana/runtime/src/prioritization_fee_cache.rs

Lines 372 to 382 in b0112a5

	let mut fee = prioritization_fee_read
	.get_min_transaction_fee()
	.unwrap_or_default();
	for account_key in account_keys {
	if let Some(account_fee) =
	prioritization_fee_read.get_writable_account_fee(account_key)
	{
	fee = std::cmp::max(fee, account_fee);
	}
	}
	Some((*slot, fee))

[GCrispino]

@CriesofCarrots thanks for answering!

I guess that this also explain why I basically always get 0 values when not passing accounts as parameter to this method, because it means that there is always a transaction in the slot with 0 as the value for the priority fee and, since the minimum value is fetched from the cache, 0 is going to be returned. Is this explanation correct?

[CriesofCarrots]

Yes, correct.

[GCrispino]

That makes more sense now :) . Don't you think that the docs for this method should specify that the values returned are minimum values? It currently mentions that the returned value for each slot is the "per-compute-unit fee paid by at least one successfully landed transaction, specified in increments of 0.000001 lamports".

Although this is not wrong, having "minimum per-compute-unit fee paid by a landed transaction" instead would sound more representative to what happens for this method, in my opinion.