Historical re-exec: bottleneck on .efi 1-st layer lookup

[AskAlexSharov]

Maybe same blocks and receipts LRU as was in RPC (or different if it’s hard).

Probably it does re-exec because Receipt requested:

[SLoeuillet]

Here is the GraphQL request responsible for this profile :

	jsonData := map[string]string{
		"query": fmt.Sprintf(`
			{	
				block(number: %d ) {
					number
					difficulty
					extraData
					gasLimit
					gasUsed
					logsBloom
					ommerHash
					ommers{
					  hash
					}
					miner {
					  address
					}
					mixHash
					nonce
					hash
					parent {
					  hash
					}
					receiptsRoot
					stateRoot
					timestamp
					transactionsRoot
					transactions {
					  createdContract {
						address
					  }
					  cumulativeGasUsed
					  from {
						address
					  }
					  gasUsed
					  gasPrice
					  inputData
					  nonce
					  index
					  value
					  logs {
						account {
						  address
						}
						data
						index
						topics
					  }
					  hash
					  index
					  to {
						address
					  }
					  status
					}
				  }
				}
        `, blockNumber),
	}

[SLoeuillet]

Everytime we launch a backfill (from first block or more generaly from first block of a pool or ERC20 token, we run that query for each block)

Rest of the time, we run this query on each block while they arrive on chain tip

[SLoeuillet]

@scorring @Ugo got that trace by using the following command

/opt/golang/bin/go tool pprof -png "http://127.0.0.1:6062/debug/pprof/profile?debug=1&seconds=20" > cpu.png

[AskAlexSharov]

did some step with ~25% improvement #12970
but probably it's still not enough. keep working on it.

[SLoeuillet]

OK, thanks