Merge branch 'master' into batch-size-moving-average

arbnode/dataposter/data_poster.go

@@ -410,6 +410,8 @@ func (p *DataPoster) evalMaxFeeCapExpr(backlogOfBatches uint64, elapsed time.Dur
 	return resultBig, nil
 }
 
+var big4 = big.NewInt(4)
+
 // The dataPosterBacklog argument should *not* include extraBacklog (it's added in in this function)
 func (p *DataPoster) feeAndTipCaps(ctx context.Context, nonce uint64, gasLimit uint64, lastFeeCap *big.Int, lastTipCap *big.Int, dataCreatedAt time.Time, dataPosterBacklog uint64) (*big.Int, *big.Int, error) {
 	config := p.config()
@@ -468,17 +470,35 @@ func (p *DataPoster) feeAndTipCaps(ctx context.Context, nonce uint64, gasLimit u
 	latestBalance := p.balance
 	balanceForTx := new(big.Int).Set(latestBalance)
 	if config.AllocateMempoolBalance && !p.usingNoOpStorage {
-		// We reserve half the balance for the first transaction, and then split the remaining balance for all after that.
+		// We split the transactions into three groups:
+		// - The first transaction gets 1/2 of the balance.
+		// - The first half of transactions get 1/3 of the balance split among them.
+		// - The remaining transactions get the remaining 1/6 of the balance split among them.
+		// This helps ensure batch posting is reliable under a variety of fee conditions.
 		// With noop storage, we don't try to replace-by-fee, so we don't need to worry about this.
 		balanceForTx.Div(balanceForTx, common.Big2)
 		if nonce != softConfNonce && config.MaxMempoolTransactions > 1 {
+			// Compared to dividing the remaining transactions by balance equally,
+			// the first half of transactions should get a 4/3 weight,
+			// and the remaining half should get a 2/3 weight.
+			// This makes sure the average weight is 1, and the first half of transactions
+			// have twice the weight of the second half of transactions.
+			// The +1 and -1 here are to account for the first transaction being handled separately.
+			if nonce > softConfNonce && nonce < softConfNonce+1+(config.MaxMempoolTransactions-1)/2 {
+				balanceForTx.Mul(balanceForTx, big4)
+			} else {
+				balanceForTx.Mul(balanceForTx, common.Big2)
+			}
+			balanceForTx.Div(balanceForTx, common.Big3)
+			// After weighting, split the balance between each of the transactions
+			// other than the first tx which already got half.
 			// balanceForTx /= config.MaxMempoolTransactions-1
 			balanceForTx.Div(balanceForTx, arbmath.UintToBig(config.MaxMempoolTransactions-1))
 		}
 	}
 	balanceFeeCap := arbmath.BigDivByUint(balanceForTx, gasLimit)
 	if arbmath.BigGreaterThan(newFeeCap, balanceFeeCap) {
-		log.Error(
+		log.Warn(
 			"lack of L1 balance prevents posting transaction with desired fee cap",
 			"balance", latestBalance,
 			"maxTransactions", config.MaxMempoolTransactions,