Differential D16800
[chronik] better use the merkle tree cache Authored by PiRK on Sep 23 2024, 13:07. Tags None Subscribers None
Details
The current logic is wrong, we only use the cache when computing a merkle root for a block height lower than the previous highest one, i.e if the cache contains all or more intermediate hashes than we need. With this change, we now use the cache even if it provides only part of the intermediate hashes. ninja all check-all
Diff Detail
Event TimelineComment Actions Benchmark: diff --git a/chronik/chronik-indexer/src/query/blocks.rs b/chronik/chronik-indexer/src/query/blocks.rs index 01e658e2fd..f130f2e7f3 100644 --- a/chronik/chronik-indexer/src/query/blocks.rs +++ b/chronik/chronik-indexer/src/query/blocks.rs @@ -33,6 +33,7 @@ use crate::{ OutputsSpent, TxTokenData, }, }; +use std::time::{Duration, Instant}; const MAX_BLOCKS_PAGE_SIZE: usize = 500; @@ -368,6 +369,7 @@ impl<'a> QueryBlocks<'a> { ) -> Result<(Vec<u8>, Vec<Vec<u8>>)> { let mut root = Vec::<u8>::new(); let mut branch = Vec::<Vec<u8>>::new(); + let mut start = Instant::now(); if checkpoint_height > 0 { Self::check_checkpoint_height(block_height, checkpoint_height)?; let bridge = &self.node.bridge; @@ -376,9 +378,16 @@ impl<'a> QueryBlocks<'a> { .iter() .map(|raw_hash| Sha256d::from_le_bytes(raw_hash.data)) .collect(); + let duration = start.elapsed(); + println!("Time elapsed fetching headers is: {:?}", duration); + start = Instant::now(); let mut block_merkle_tree = self.block_merkle_tree.lock().await; let (root_hash, branch_hashes) = block_merkle_tree .merkle_root_and_branch(&hashes, block_height as usize); + + let duration = start.elapsed(); + println!("Time elapsed computing merkle tree is: {:?}", duration); + root = root_hash.to_le_bytes().to_vec(); branch = branch_ $ python Python 3.12.3 (main, Sep 11 2024, 14:17:37) [GCC 13.2.0] on linux Type "help", "copyright", "credits" or "license" for more information. >>> from chronik import client >>> cl = client.ChronikClient("127.0.0.1", 8331) >>> pb = cl.block_header(863480, 863480) >>> pb = cl.block_header(863482, 863482) >>> pb = cl.block_header(863484, 863484) >>> pb = cl.block_header(863486, 863486) >>> pb = cl.block_header(863488, 863488) Before: Time elapsed fetching headers is: 92.83265ms Time elapsed computing merkle tree is: 144.761108ms Time elapsed fetching headers is: 86.521631ms Time elapsed computing merkle tree is: 96.598198ms Time elapsed fetching headers is: 86.687492ms Time elapsed computing merkle tree is: 120.510652ms Time elapsed fetching headers is: 86.431005ms Time elapsed computing merkle tree is: 97.853639ms Time elapsed fetching headers is: 86.039362ms Time elapsed computing merkle tree is: 144.968584ms After Time elapsed fetching headers is: 96.03466ms Time elapsed computing merkle tree is: 144.438252ms Time elapsed fetching headers is: 87.074127ms Time elapsed computing merkle tree is: 54.741982ms Time elapsed fetching headers is: 87.528585ms Time elapsed computing merkle tree is: 49.760862ms Time elapsed fetching headers is: 87.293025ms Time elapsed computing merkle tree is: 55.79699ms Time elapsed fetching headers is: 86.483147ms Time elapsed computing merkle tree is: 48.828292ms I found this issue while working on an optimization to not load all the blockhashes when we only need the last few ones. |