perf(storage): optimize internode RPC transfer path (#2262)

Co-authored-by: momoda693 <momoda693@gmail.com>
This commit is contained in:
weisd
2026-03-23 17:09:49 +08:00
committed by GitHub
parent 236142a682
commit 05dc131a49
43 changed files with 1846 additions and 566 deletions
+164 -6
View File
@@ -15,6 +15,7 @@
use bytes::Bytes;
use pin_project_lite::pin_project;
use rustfs_utils::HashAlgorithm;
use std::io::IoSlice;
use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, AsyncWriteExt};
use tracing::error;
use uuid::Uuid;
@@ -155,23 +156,49 @@ where
error!("bitrot writer write hash error: hash is empty");
return Err(std::io::Error::new(std::io::ErrorKind::InvalidInput, "hash is empty"));
}
self.inner.write_all(hash.as_ref()).await?;
write_all_vectored(&mut self.inner, hash.as_ref(), buf).await?;
} else {
self.inner.write_all(buf).await?;
}
self.inner.write_all(buf).await?;
self.inner.flush().await?;
let n = buf.len();
Ok(n)
}
pub async fn shutdown(&mut self) -> std::io::Result<()> {
self.inner.flush().await?;
self.inner.shutdown().await
}
}
async fn write_all_vectored<W>(writer: &mut W, hash: &[u8], data: &[u8]) -> std::io::Result<()>
where
W: AsyncWrite + Unpin,
{
let mut hash_offset = 0;
let mut data_offset = 0;
while hash_offset < hash.len() || data_offset < data.len() {
let slices = [IoSlice::new(&hash[hash_offset..]), IoSlice::new(&data[data_offset..])];
let written = writer.write_vectored(&slices).await?;
if written == 0 {
return Err(std::io::Error::new(std::io::ErrorKind::WriteZero, "failed to write hash and data"));
}
let hash_remaining = hash.len() - hash_offset;
if written < hash_remaining {
hash_offset += written;
continue;
}
hash_offset = hash.len();
data_offset += written - hash_remaining;
}
Ok(())
}
pub fn bitrot_shard_file_size(size: usize, shard_size: usize, algo: HashAlgorithm) -> usize {
if algo != HashAlgorithm::HighwayHash256S && algo != HashAlgorithm::HighwayHash256SLegacy {
return size;
@@ -292,6 +319,33 @@ impl AsyncWrite for CustomWriter {
}
}
}
fn poll_write_vectored(
self: std::pin::Pin<&mut Self>,
cx: &mut std::task::Context<'_>,
bufs: &[IoSlice<'_>],
) -> std::task::Poll<std::io::Result<usize>> {
match self.get_mut() {
Self::InlineBuffer(data) => {
let total = bufs.iter().map(|buf| buf.len()).sum::<usize>();
for buf in bufs {
data.extend_from_slice(buf);
}
std::task::Poll::Ready(Ok(total))
}
Self::Other(writer) => {
let pinned_writer = std::pin::Pin::new(writer.as_mut());
pinned_writer.poll_write_vectored(cx, bufs)
}
}
}
fn is_write_vectored(&self) -> bool {
match self {
Self::InlineBuffer(_) => true,
Self::Other(writer) => writer.is_write_vectored(),
}
}
}
/// Wrapper around BitrotWriter that uses our custom writer
@@ -361,7 +415,74 @@ mod tests {
use super::BitrotReader;
use super::BitrotWriter;
use rustfs_utils::HashAlgorithm;
use std::io::Cursor;
use std::io::{Cursor, IoSlice};
use std::sync::{
Arc,
atomic::{AtomicUsize, Ordering},
};
use std::task::{Context, Poll};
use tokio::io::AsyncWrite;
#[derive(Default)]
struct VectoredCountingWriter {
vectored_writes: Arc<AtomicUsize>,
writes: Vec<u8>,
}
impl AsyncWrite for VectoredCountingWriter {
fn poll_write(self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>, _buf: &[u8]) -> Poll<std::io::Result<usize>> {
Poll::Ready(Err(std::io::Error::other("poll_write should not be used")))
}
fn poll_flush(self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
Poll::Ready(Ok(()))
}
fn poll_shutdown(self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
Poll::Ready(Ok(()))
}
fn poll_write_vectored(
mut self: std::pin::Pin<&mut Self>,
_cx: &mut Context<'_>,
bufs: &[IoSlice<'_>],
) -> Poll<std::io::Result<usize>> {
self.vectored_writes.fetch_add(1, Ordering::SeqCst);
let total = bufs.iter().map(|buf| buf.len()).sum::<usize>();
for buf in bufs {
self.writes.extend_from_slice(buf);
}
Poll::Ready(Ok(total))
}
fn is_write_vectored(&self) -> bool {
true
}
}
#[derive(Default)]
struct CountingWriter {
flushes: Arc<AtomicUsize>,
shutdowns: Arc<AtomicUsize>,
writes: Vec<u8>,
}
impl AsyncWrite for CountingWriter {
fn poll_write(mut self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>, buf: &[u8]) -> Poll<std::io::Result<usize>> {
self.writes.extend_from_slice(buf);
Poll::Ready(Ok(buf.len()))
}
fn poll_flush(self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
self.flushes.fetch_add(1, Ordering::SeqCst);
Poll::Ready(Ok(()))
}
fn poll_shutdown(self: std::pin::Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
self.shutdowns.fetch_add(1, Ordering::SeqCst);
Poll::Ready(Ok(()))
}
}
#[tokio::test]
async fn test_bitrot_read_write_ok() {
@@ -471,4 +592,41 @@ mod tests {
assert_eq!(n, data_size);
assert_eq!(data, &out[..]);
}
#[tokio::test]
async fn test_bitrot_writer_flushes_once_on_shutdown() {
let flushes = Arc::new(AtomicUsize::new(0));
let shutdowns = Arc::new(AtomicUsize::new(0));
let writer = CountingWriter {
flushes: flushes.clone(),
shutdowns: shutdowns.clone(),
writes: Vec::new(),
};
let mut bitrot_writer = BitrotWriter::new(writer, 8, HashAlgorithm::None);
bitrot_writer.write(b"12345678").await.unwrap();
bitrot_writer.write(b"abc").await.unwrap();
assert_eq!(flushes.load(Ordering::SeqCst), 0);
assert_eq!(shutdowns.load(Ordering::SeqCst), 0);
bitrot_writer.shutdown().await.unwrap();
assert_eq!(flushes.load(Ordering::SeqCst), 1);
assert_eq!(shutdowns.load(Ordering::SeqCst), 1);
}
#[tokio::test]
async fn test_bitrot_writer_uses_vectored_write_for_hash_and_data() {
let vectored_writes = Arc::new(AtomicUsize::new(0));
let writer = VectoredCountingWriter {
vectored_writes: vectored_writes.clone(),
writes: Vec::new(),
};
let mut bitrot_writer = BitrotWriter::new(writer, 8, HashAlgorithm::HighwayHash256);
bitrot_writer.write(b"payload").await.unwrap();
assert!(vectored_writes.load(Ordering::SeqCst) > 0);
}
}
+122 -5
View File
@@ -112,11 +112,66 @@ impl<'a> MultiWriter<'a> {
)))
}
pub async fn _shutdown(&mut self) -> std::io::Result<()> {
for writer in self.writers.iter_mut().flatten() {
writer.shutdown().await?;
async fn shutdown_writer(writer_opt: &mut Option<BitrotWriterWrapper>, err: &mut Option<Error>) {
match writer_opt {
Some(writer) => match writer.shutdown().await {
Ok(()) => {
*err = None;
}
Err(e) => {
*err = Some(Error::from(e));
*writer_opt = None;
}
},
None => {
*err = Some(Error::DiskNotFound);
}
}
Ok(())
}
pub async fn shutdown(&mut self) -> std::io::Result<()> {
{
let mut futures = FuturesUnordered::new();
for (writer_opt, err) in self.writers.iter_mut().zip(self.errs.iter_mut()) {
if err.is_some() {
continue;
}
futures.push(Self::shutdown_writer(writer_opt, err));
}
while let Some(()) = futures.next().await {}
}
let nil_count = self.errs.iter().filter(|&e| e.is_none()).count();
if nil_count >= self.write_quorum {
return Ok(());
}
if let Some(write_err) = reduce_write_quorum_errs(&self.errs, OBJECT_OP_IGNORED_ERRS, self.write_quorum) {
error!(
"reduce_write_quorum_errs during shutdown: {:?}, offline-disks={}/{}, errs={:?}",
write_err,
count_errs(&self.errs, &Error::DiskNotFound),
self.writers.len(),
self.errs
);
return Err(std::io::Error::other(format!(
"Failed to shutdown writers: {} (offline-disks={}/{})",
write_err,
count_errs(&self.errs, &Error::DiskNotFound),
self.writers.len()
)));
}
Err(std::io::Error::other(format!(
"Failed to shutdown writers: (offline-disks={}/{}): {}",
count_errs(&self.errs, &Error::DiskNotFound),
self.writers.len(),
self.errs
.iter()
.map(|e| e.as_ref().map_or("<nil>".to_string(), |e| e.to_string()))
.collect::<Vec<_>>()
.join(", ")
)))
}
}
@@ -176,7 +231,69 @@ impl Erasure {
}
let (reader, total) = task.await??;
// writers.shutdown().await?;
writers.shutdown().await?;
Ok((reader, total))
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::erasure_coding::{BitrotWriterWrapper, CustomWriter};
use rustfs_utils::HashAlgorithm;
use std::pin::Pin;
use std::sync::{Arc, Mutex};
use std::task::{Context, Poll};
use tokio::io::AsyncWrite;
#[derive(Clone, Default)]
struct DeferredCommitWriter {
buffered: Vec<u8>,
committed: Arc<Mutex<Vec<u8>>>,
}
impl DeferredCommitWriter {
fn new(committed: Arc<Mutex<Vec<u8>>>) -> Self {
Self {
buffered: Vec::new(),
committed,
}
}
}
impl AsyncWrite for DeferredCommitWriter {
fn poll_write(mut self: Pin<&mut Self>, _cx: &mut Context<'_>, buf: &[u8]) -> Poll<std::io::Result<usize>> {
self.buffered.extend_from_slice(buf);
Poll::Ready(Ok(buf.len()))
}
fn poll_flush(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
Poll::Ready(Ok(()))
}
fn poll_shutdown(mut self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
let buffered = std::mem::take(&mut self.buffered);
let mut committed = self.committed.lock().unwrap();
committed.extend_from_slice(&buffered);
Poll::Ready(Ok(()))
}
}
#[tokio::test]
async fn encode_shutdowns_writers_after_small_shards() {
let committed = Arc::new(Mutex::new(Vec::new()));
let writer = DeferredCommitWriter::new(committed.clone());
let mut writers = vec![Some(BitrotWriterWrapper::new(
CustomWriter::new_tokio_writer(writer),
16,
HashAlgorithm::HighwayHash256S,
))];
let erasure = Arc::new(Erasure::new(1, 0, 16));
let reader = tokio::io::BufReader::new(std::io::Cursor::new(b"small payload".to_vec()));
let (_reader, written) = erasure.encode(reader, &mut writers, 1).await.unwrap();
assert_eq!(written, b"small payload".len());
assert!(!committed.lock().unwrap().is_empty());
}
}