mdebray/tvl-depot
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

refactor(tvix/castore): extract concurrent blob uploader

Browse source 
The archive ingester has a mechanism for concurrently uploading small
blobs to the blob service in order to hide round trip latency with the
blob service when ingesting many small blobs.

Other ingestion sources like NARs also need a similar mechanism, this
extracts the concurrent blob uploading mechanism into its own struct to
make it more reusable.

Change-Id: I05020419ff4b9ad5829fbfb5cd08d36db983b8c0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11693
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This commit is contained in:
Connor Brewster 2024-05-18 13:25:47 -05:00
parent bc42c355cf
commit b0aaff25fa
3 changed files with 190 additions and 97 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

107
tvix/castore/src/import/archive.rs
View file

@ -1,39 +1,24 @@
//! Imports from an archive (tarballs)
use std::collections::HashMap;
use std::io::{Cursor, Write};
use std::sync::Arc;
use petgraph::graph::{DiGraph, NodeIndex};
use petgraph::visit::{DfsPostOrder, EdgeRef};
use petgraph::Direction;
use tokio::io::AsyncRead;
use tokio::sync::Semaphore;
use tokio::task::JoinSet;
use tokio_stream::StreamExt;
use tokio_tar::Archive;
use tokio_util::io::InspectReader;
use tracing::{instrument, warn, Level};
use crate::blobservice::BlobService;
use crate::directoryservice::DirectoryService;
use crate::import::{ingest_entries, IngestionEntry, IngestionError};
use crate::proto::node::Node;
use crate::B3Digest;
use super::blobs::{self, ConcurrentBlobUploader};
type TarPathBuf = std::path::PathBuf;
/// Files smaller than this threshold, in bytes, are uploaded to the [BlobService] in the
/// background.
///
/// This is a u32 since we acquire a weighted semaphore using the size of the blob.
/// [Semaphore::acquire_many_owned] takes a u32, so we need to ensure the size of
/// the blob can be represented using a u32 and will not cause an overflow.
const CONCURRENT_BLOB_UPLOAD_THRESHOLD: u32 = 1024 * 1024;
/// The maximum amount of bytes allowed to be buffered in memory to perform async blob uploads.
const MAX_TARBALL_BUFFER_SIZE: usize = 128 * 1024 * 1024;
#[derive(Debug, thiserror::Error)]
pub enum Error {
#[error("unable to construct stream of entries: {0}")]
@ -57,13 +42,6 @@ pub enum Error {
#[error("unable to read link name field for {0}: {1}")]
LinkName(TarPathBuf, std::io::Error),
#[error("unable to read blob contents for {0}: {1}")]
BlobRead(TarPathBuf, std::io::Error),
// FUTUREWORK: proper error for blob finalize
#[error("unable to finalize blob {0}: {1}")]
BlobFinalize(TarPathBuf, std::io::Error),
#[error("unsupported tar entry {0} type: {1:?}")]
EntryType(TarPathBuf, tokio_tar::EntryType),
@ -72,6 +50,9 @@ pub enum Error {
#[error("unexpected number of top level directory entries")]
UnexpectedNumberOfTopLevelEntries,
#[error(transparent)]
BlobUploadError(#[from] blobs::Error),
}
/// Ingests elements from the given tar [`Archive`] into a the passed [`BlobService`] and
@ -94,8 +75,7 @@ where
// In the first phase, collect up all the regular files and symlinks.
let mut nodes = IngestionEntryGraph::new();
let semaphore = Arc::new(Semaphore::new(MAX_TARBALL_BUFFER_SIZE));
let mut async_blob_uploads: JoinSet<Result<(), Error>> = JoinSet::new();
let mut blob_uploader = ConcurrentBlobUploader::new(blob_service);
let mut entries_iter = archive.entries().map_err(Error::Entries)?;
while let Some(mut entry) = entries_iter.try_next().await.map_err(Error::NextEntry)? {
@ -110,77 +90,14 @@ where
tokio_tar::EntryType::Regular
| tokio_tar::EntryType::GNUSparse
| tokio_tar::EntryType::Continuous => {
let header_size = header
let size = header
.size()
.map_err(|e| Error::Size(tar_path.clone(), e))?;
// If the blob is small enough, read it off the wire, compute the digest,
// and upload it to the [BlobService] in the background.
let (size, digest) = if header_size <= CONCURRENT_BLOB_UPLOAD_THRESHOLD as u64 {
let mut buffer = Vec::with_capacity(header_size as usize);
let mut hasher = blake3::Hasher::new();
let mut reader = InspectReader::new(&mut entry, |bytes| {
hasher.write_all(bytes).unwrap();
});
// Ensure that we don't buffer into memory until we've acquired a permit.
// This prevents consuming too much memory when performing concurrent
// blob uploads.
let permit = semaphore
.clone()
// This cast is safe because ensure the header_size is less than
// CONCURRENT_BLOB_UPLOAD_THRESHOLD which is a u32.
.acquire_many_owned(header_size as u32)
let digest = blob_uploader
.upload(&path, size, &mut entry)
.await
.unwrap();
let size = tokio::io::copy(&mut reader, &mut buffer)
.await
.map_err(|e| Error::Size(tar_path.clone(), e))?;
let digest: B3Digest = hasher.finalize().as_bytes().into();
{
let blob_service = blob_service.clone();
let digest = digest.clone();
async_blob_uploads.spawn({
let tar_path = tar_path.clone();
async move {
let mut writer = blob_service.open_write().await;
tokio::io::copy(&mut Cursor::new(buffer), &mut writer)
.await
.map_err(|e| Error::BlobRead(tar_path.clone(), e))?;
let blob_digest = writer
.close()
.await
.map_err(|e| Error::BlobFinalize(tar_path, e))?;
assert_eq!(digest, blob_digest, "Tvix bug: blob digest mismatch");
// Make sure we hold the permit until we finish writing the blob
// to the [BlobService].
drop(permit);
Ok(())
}
});
}
(size, digest)
} else {
let mut writer = blob_service.open_write().await;
let size = tokio::io::copy(&mut entry, &mut writer)
.await
.map_err(|e| Error::BlobRead(tar_path.clone(), e))?;
let digest = writer
.close()
.await
.map_err(|e| Error::BlobFinalize(tar_path.clone(), e))?;
(size, digest)
};
.map_err(Error::BlobUploadError)?;
let executable = entry
.header()
@ -219,9 +136,7 @@ where
nodes.add(entry)?;
}
while let Some(result) = async_blob_uploads.join_next().await {
result.expect("task panicked")?;
}
blob_uploader.join().await.map_err(Error::BlobUploadError)?;
let root_node = ingest_entries(
directory_service,

177
tvix/castore/src/import/blobs.rs Normal file
View file

@ -0,0 +1,177 @@
use std::{
io::{Cursor, Write},
sync::Arc,
};
use tokio::{
io::AsyncRead,
sync::Semaphore,
task::{JoinError, JoinSet},
};
use tokio_util::io::InspectReader;
use crate::{blobservice::BlobService, B3Digest, Path, PathBuf};
/// Files smaller than this threshold, in bytes, are uploaded to the [BlobService] in the
/// background.
///
/// This is a u32 since we acquire a weighted semaphore using the size of the blob.
/// [Semaphore::acquire_many_owned] takes a u32, so we need to ensure the size of
/// the blob can be represented using a u32 and will not cause an overflow.
const CONCURRENT_BLOB_UPLOAD_THRESHOLD: u32 = 1024 * 1024;
/// The maximum amount of bytes allowed to be buffered in memory to perform async blob uploads.
const MAX_BUFFER_SIZE: usize = 128 * 1024 * 1024;
#[derive(Debug, thiserror::Error)]
pub enum Error {
#[error("unable to read blob contents for {0}: {1}")]
BlobRead(PathBuf, std::io::Error),
// FUTUREWORK: proper error for blob finalize
#[error("unable to finalize blob {0}: {1}")]
BlobFinalize(PathBuf, std::io::Error),
#[error("unexpected size for {path} wanted: {wanted} got: {got}")]
UnexpectedSize {
path: PathBuf,
wanted: u64,
got: u64,
},
#[error("blob upload join error: {0}")]
JoinError(#[from] JoinError),
}
/// The concurrent blob uploader provides a mechanism for concurrently uploading small blobs.
/// This is useful when ingesting from sources like tarballs and archives which each blob entry
/// must be read sequentially. Ingesting many small blobs sequentially becomes slow due to
/// round trip time with the blob service. The concurrent blob uploader will buffer small
/// blobs in memory and upload them to the blob service in the background.
///
/// Once all blobs have been uploaded, make sure to call [ConcurrentBlobUploader::join] to wait
/// for all background jobs to complete and check for any errors.
pub struct ConcurrentBlobUploader<BS> {
blob_service: BS,
upload_tasks: JoinSet<Result<(), Error>>,
upload_semaphore: Arc<Semaphore>,
}
impl<BS> ConcurrentBlobUploader<BS>
where
BS: BlobService + Clone + 'static,
{
/// Creates a new concurrent blob uploader which uploads blobs to the provided
/// blob service.
pub fn new(blob_service: BS) -> Self {
Self {
blob_service,
upload_tasks: JoinSet::new(),
upload_semaphore: Arc::new(Semaphore::new(MAX_BUFFER_SIZE)),
}
}
/// Uploads a blob to the blob service. If the blob is small enough it will be read to a buffer
/// and uploaded in the background.
/// This will read the entirety of the provided reader unless an error occurs, even if blobs
/// are uploaded in the background..
pub async fn upload<R>(
&mut self,
path: &Path,
expected_size: u64,
mut r: R,
) -> Result<B3Digest, Error>
where
R: AsyncRead + Unpin,
{
if expected_size < CONCURRENT_BLOB_UPLOAD_THRESHOLD as u64 {
let mut buffer = Vec::with_capacity(expected_size as usize);
let mut hasher = blake3::Hasher::new();
let mut reader = InspectReader::new(&mut r, |bytes| {
hasher.write_all(bytes).unwrap();
});
let permit = self
.upload_semaphore
.clone()
// This cast is safe because ensure the header_size is less than
// CONCURRENT_BLOB_UPLOAD_THRESHOLD which is a u32.
.acquire_many_owned(expected_size as u32)
.await
.unwrap();
let size = tokio::io::copy(&mut reader, &mut buffer)
.await
.map_err(|e| Error::BlobRead(path.into(), e))?;
let digest: B3Digest = hasher.finalize().as_bytes().into();
if size != expected_size {
return Err(Error::UnexpectedSize {
path: path.into(),
wanted: expected_size,
got: size,
});
}
self.upload_tasks.spawn({
let blob_service = self.blob_service.clone();
let expected_digest = digest.clone();
let path = path.to_owned();
let r = Cursor::new(buffer);
async move {
let digest = upload_blob(&blob_service, &path, expected_size, r).await?;
assert_eq!(digest, expected_digest, "Tvix bug: blob digest mismatch");
// Make sure we hold the permit until we finish writing the blob
// to the [BlobService].
drop(permit);
Ok(())
}
});
return Ok(digest);
}
upload_blob(&self.blob_service, path, expected_size, r).await
}
/// Waits for all background upload jobs to complete, returning any upload errors.
pub async fn join(mut self) -> Result<(), Error> {
while let Some(result) = self.upload_tasks.join_next().await {
result??;
}
Ok(())
}
}
async fn upload_blob<BS, R>(
blob_service: &BS,
path: &Path,
expected_size: u64,
mut r: R,
) -> Result<B3Digest, Error>
where
BS: BlobService,
R: AsyncRead + Unpin,
{
let mut writer = blob_service.open_write().await;
let size = tokio::io::copy(&mut r, &mut writer)
.await
.map_err(|e| Error::BlobRead(path.into(), e))?;
let digest = writer
.close()
.await
.map_err(|e| Error::BlobFinalize(path.into(), e))?;
if size != expected_size {
return Err(Error::UnexpectedSize {
path: path.into(),
wanted: expected_size,
got: size,
});
}
Ok(digest)
}

1
tvix/castore/src/import/mod.rs
View file

@ -24,6 +24,7 @@ mod error;
pub use error::IngestionError;
pub mod archive;
pub mod blobs;
pub mod fs;
/// Ingests [IngestionEntry] from the given stream into a the passed [DirectoryService].