feat(reth-bench): substract block fetch waiting time from benchmark duration #14299
Conversation
fgimenez
added
C-enhancement
| let fetch_start = Instant::now(); | ||
|
|
||
| let block_res = | ||
| block_provider.get_block_by_number(next_block.into(), true.into()).await; | ||
| let fetch_duration = fetch_start.elapsed(); | ||
| let block = block_res.unwrap().unwrap(); | ||
| let block = from_any_rpc_block(block); | ||
|
|
||
| next_block += 1; | ||
| sender.send(block).await.unwrap(); | ||
| sender.send((block, fetch_duration)).await.unwrap(); | ||
| } | ||
| }); | ||
|
|
||
| // put results in a summary vec so they can be printed at the end | ||
| let mut results = Vec::new(); | ||
| let total_benchmark_duration = Instant::now(); | ||
|
|
||
| while let Some(block) = receiver.recv().await { | ||
| let mut total_fetch_duration = Duration::ZERO; | ||
| while let Some((block, fetch_duration)) = receiver.recv().await { | ||
| // just put gas used here | ||
| let gas_used = block.gas_used; |
There was a problem hiding this comment.
given this is happening in a separate task, is this accurate? it seems like we might subtract too much
There was a problem hiding this comment.
can you pls elaborate? happens in a separate task but we record and return the Duration that the fetch operations per block took, in the other task we aggregate these durations and subtract them from the total benchmark time so far, is this correct?
There was a problem hiding this comment.
oh i see what you mean, the channel size is 1000, the block downloader task can progress without blocking the engine interaction task, it is not that easy to know if the block downloads are actually affecting the engine interactions
There was a problem hiding this comment.
the block downloader task can progress without blocking the engine interaction task, it is not that easy to know if the block downloads are actually affecting the engine interactions
Yeah exactly, I think we would instead want to subtract any blocking that occurs in the main task, which currently happens in the loop condition (awaiting on the recv)
There was a problem hiding this comment.
yep that's it, pushed the changes PTAL
fgimenez
changed the title
Currently we are including the potential time spent waiting for blocks in the total benchmark duration, with these changes we record any waiting time and subtract it when we store the current benchmark duration for each block.