fix: revert standalone #2351 artifacts (phase 5) (#2535)

This commit is contained in:
安正超
2026-04-14 22:35:05 +08:00
committed by GitHub
parent db8ef63674
commit 68d3dba9fc
20 changed files with 138 additions and 1681 deletions
-28
View File
@@ -49,34 +49,6 @@ pub const ENV_OBJECT_ZERO_COPY_ENABLE: &str = "RUSTFS_OBJECT_ZERO_COPY_ENABLE";
/// to regular I/O without errors.
pub const DEFAULT_OBJECT_ZERO_COPY_ENABLE: bool = true;
/// Environment variable for zero-copy read operating mode.
///
/// Supported values:
/// - `off`: disable mmap-backed chunk fast path and always use the compatibility path
/// - `conservative`: allow a single mmap window per request
/// - `balanced`: allow multiple mmap windows with the default size guardrails
/// - `aggressive`: allow multi-window mmap and relax the small-object cutoff
pub const ENV_OBJECT_ZERO_COPY_MODE: &str = "RUSTFS_OBJECT_ZERO_COPY_MODE";
/// Default zero-copy read mode.
pub const DEFAULT_OBJECT_ZERO_COPY_MODE: &str = "balanced";
/// Environment variable for the maximum mmap window size used by the chunk fast path.
///
/// This controls the visible bytes per mapped chunk before the implementation emits a new window.
pub const ENV_OBJECT_ZERO_COPY_MMAP_WINDOW_BYTES: &str = "RUSTFS_OBJECT_ZERO_COPY_MMAP_WINDOW_BYTES";
/// Default mmap window size for chunk fast path reads: 8 MiB.
pub const DEFAULT_OBJECT_ZERO_COPY_MMAP_WINDOW_BYTES: usize = 8 * 1024 * 1024;
/// Environment variable for the maximum total active mmap bytes.
///
/// Requests that would exceed this active window budget fall back to the compatibility path.
pub const ENV_OBJECT_ZERO_COPY_MAX_ACTIVE_MMAP_BYTES: &str = "RUSTFS_OBJECT_ZERO_COPY_MAX_ACTIVE_MMAP_BYTES";
/// Default maximum active mmap bytes across concurrent local chunk fast-path reads: 256 MiB.
pub const DEFAULT_OBJECT_ZERO_COPY_MAX_ACTIVE_MMAP_BYTES: usize = 256 * 1024 * 1024;
// =============================================================================
// Direct I/O Configuration
// =============================================================================
-5
View File
@@ -20,11 +20,6 @@ license.workspace = true
repository.workspace = true
rust-version.workspace = true
[[bin]]
name = "small_put_bench"
path = "src/bin/small_put_bench.rs"
test = false
[lints]
workspace = true
-441
View File
@@ -1,441 +0,0 @@
// Copyright 2024 RustFS Team
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
use anyhow::{Context, Result, anyhow, bail};
use aws_sdk_s3::config::{Credentials, Region};
use aws_sdk_s3::primitives::ByteStream;
use aws_sdk_s3::types::{Delete, ObjectIdentifier};
use aws_sdk_s3::{Client, Config};
use aws_smithy_http_client::Builder as SmithyHttpClientBuilder;
use bytes::Bytes;
use clap::Parser;
use serde::Serialize;
use std::path::PathBuf;
use std::time::{Duration, Instant};
#[derive(Parser, Debug)]
#[command(name = "small_put_bench")]
#[command(about = "Rust-native small PUT benchmark for RustFS-compatible S3 endpoints")]
struct Args {
#[arg(long, env = "RUSTFS_BENCH_ENDPOINT")]
endpoint: String,
#[arg(long, env = "RUSTFS_BENCH_ACCESS_KEY", default_value = "rustfsadmin")]
access_key: String,
#[arg(long, env = "RUSTFS_BENCH_SECRET_KEY", default_value = "rustfsadmin")]
secret_key: String,
#[arg(long, env = "RUSTFS_BENCH_REGION", default_value = "us-east-1")]
region: String,
#[arg(long, env = "RUSTFS_BENCH_BUCKET", default_value = "small-put-benchmark")]
bucket: String,
#[arg(long, env = "RUSTFS_BENCH_SIZES", default_value = "4KiB,16KiB,64KiB,256KiB,1MiB")]
sizes: String,
#[arg(long, env = "RUSTFS_BENCH_CONCURRENCY", default_value_t = 8)]
concurrency: usize,
#[arg(long, env = "RUSTFS_BENCH_DURATION_SECS", default_value_t = 10)]
duration_secs: u64,
#[arg(long, env = "RUSTFS_BENCH_TIMEOUT_SECS", default_value_t = 15)]
timeout_secs: u64,
#[arg(long, env = "RUSTFS_BENCH_PREFIX")]
prefix: Option<String>,
#[arg(long)]
output_json: Option<PathBuf>,
#[arg(long, default_value_t = false)]
cleanup: bool,
}
#[derive(Clone, Debug)]
struct SizeSpec {
label: String,
slug: String,
bytes: usize,
}
#[derive(Debug)]
struct Sample {
ok: bool,
duration_ms: f64,
}
#[derive(Debug, Serialize)]
struct SizeSummary {
label: String,
bytes: usize,
total: usize,
succeeded: usize,
failed: usize,
wall_secs: f64,
object_rate: f64,
throughput_mib_per_sec: f64,
avg_ms: Option<f64>,
p50_ms: Option<f64>,
p90_ms: Option<f64>,
p99_ms: Option<f64>,
}
#[derive(Debug, Serialize)]
struct RunSummary {
run_id: String,
endpoint: String,
bucket: String,
concurrency: usize,
duration_secs: u64,
timeout_secs: u64,
sizes: Vec<SizeSummary>,
}
fn main() -> Result<()> {
let runtime = tokio::runtime::Builder::new_multi_thread()
.enable_all()
.build()
.context("failed to build tokio runtime")?;
runtime.block_on(async_main())
}
async fn async_main() -> Result<()> {
let args = Args::parse();
validate_args(&args)?;
let sizes = parse_size_list(&args.sizes)?;
let run_id = args.prefix.clone().unwrap_or_else(default_run_id);
let client = build_s3_client(&args.endpoint, &args.access_key, &args.secret_key, &args.region);
ensure_bucket(&client, &args.bucket).await?;
let mut size_summaries = Vec::with_capacity(sizes.len());
for size in &sizes {
let summary = run_size_benchmark(
client.clone(),
args.bucket.clone(),
run_id.clone(),
size.clone(),
args.concurrency,
Duration::from_secs(args.duration_secs),
Duration::from_secs(args.timeout_secs),
)
.await?;
print_size_summary(&summary);
size_summaries.push(summary);
}
if args.cleanup {
cleanup_prefix(&client, &args.bucket, &run_id).await?;
}
let summary = RunSummary {
run_id,
endpoint: args.endpoint,
bucket: args.bucket,
concurrency: args.concurrency,
duration_secs: args.duration_secs,
timeout_secs: args.timeout_secs,
sizes: size_summaries,
};
if let Some(path) = args.output_json {
let json = serde_json::to_vec_pretty(&summary).context("failed to serialize benchmark summary")?;
std::fs::write(&path, json).with_context(|| format!("failed to write benchmark summary to {}", path.display()))?;
println!("Wrote summary to {}", path.display());
}
Ok(())
}
fn validate_args(args: &Args) -> Result<()> {
if args.concurrency == 0 {
bail!("--concurrency must be greater than zero");
}
if args.duration_secs == 0 {
bail!("--duration-secs must be greater than zero");
}
if args.timeout_secs == 0 {
bail!("--timeout-secs must be greater than zero");
}
Ok(())
}
fn build_s3_client(endpoint: &str, access_key: &str, secret_key: &str, region: &str) -> Client {
let credentials = Credentials::new(access_key, secret_key, None, None, "small-put-bench");
let mut config = Config::builder()
.credentials_provider(credentials)
.region(Region::new(region.to_string()))
.endpoint_url(endpoint)
.force_path_style(true)
.behavior_version_latest();
if endpoint.starts_with("http://") {
config = config.http_client(SmithyHttpClientBuilder::new().build_http());
}
Client::from_conf(config.build())
}
async fn ensure_bucket(client: &Client, bucket: &str) -> Result<()> {
if client.head_bucket().bucket(bucket).send().await.is_ok() {
return Ok(());
}
match client.create_bucket().bucket(bucket).send().await {
Ok(_) => Ok(()),
Err(err) => {
let rendered = err.to_string();
if rendered.contains("BucketAlreadyOwnedByYou") || rendered.contains("BucketAlreadyExists") {
Ok(())
} else {
Err(err).with_context(|| format!("failed to create benchmark bucket {bucket}"))
}
}
}
}
async fn run_size_benchmark(
client: Client,
bucket: String,
run_id: String,
size: SizeSpec,
concurrency: usize,
duration: Duration,
timeout: Duration,
) -> Result<SizeSummary> {
let payload = Bytes::from(vec![0_u8; size.bytes]);
let deadline = Instant::now() + duration;
let wall_start = Instant::now();
let mut handles = Vec::with_capacity(concurrency);
for worker in 0..concurrency {
let client = client.clone();
let bucket = bucket.clone();
let payload = payload.clone();
let prefix = format!("{run_id}/{}/worker-{worker}", size.slug);
handles.push(tokio::spawn(async move {
let mut samples = Vec::new();
let mut idx = 0usize;
while Instant::now() < deadline {
let key = format!("{prefix}/obj-{idx}.bin");
let started_at = Instant::now();
let request = client
.put_object()
.bucket(&bucket)
.key(key)
.body(ByteStream::from(payload.clone()))
.content_type("application/octet-stream");
let ok = matches!(tokio::time::timeout(timeout, request.send()).await, Ok(Ok(_)));
samples.push(Sample {
ok,
duration_ms: started_at.elapsed().as_secs_f64() * 1000.0,
});
idx += 1;
}
samples
}));
}
let mut samples = Vec::new();
for handle in handles {
samples.extend(handle.await.map_err(|err| anyhow!("benchmark worker join error: {err}"))?);
}
Ok(build_size_summary(&size, samples, wall_start.elapsed()))
}
fn build_size_summary(size: &SizeSpec, mut samples: Vec<Sample>, wall_elapsed: Duration) -> SizeSummary {
let total = samples.len();
let succeeded = samples.iter().filter(|sample| sample.ok).count();
let failed = total.saturating_sub(succeeded);
let wall_secs = wall_elapsed.as_secs_f64();
let object_rate = if wall_secs > 0.0 { succeeded as f64 / wall_secs } else { 0.0 };
let throughput_mib_per_sec = if wall_secs > 0.0 {
((size.bytes * succeeded) as f64 / (1024.0 * 1024.0)) / wall_secs
} else {
0.0
};
let avg_ms = if total > 0 {
Some(samples.iter().map(|sample| sample.duration_ms).sum::<f64>() / total as f64)
} else {
None
};
samples.sort_by(|lhs, rhs| lhs.duration_ms.total_cmp(&rhs.duration_ms));
let durations: Vec<f64> = samples.into_iter().map(|sample| sample.duration_ms).collect();
SizeSummary {
label: size.label.clone(),
bytes: size.bytes,
total,
succeeded,
failed,
wall_secs,
object_rate,
throughput_mib_per_sec,
avg_ms,
p50_ms: percentile(&durations, 0.50),
p90_ms: percentile(&durations, 0.90),
p99_ms: percentile(&durations, 0.99),
}
}
async fn cleanup_prefix(client: &Client, bucket: &str, prefix: &str) -> Result<()> {
let mut continuation_token = None;
loop {
let response = client
.list_objects_v2()
.bucket(bucket)
.prefix(prefix)
.set_continuation_token(continuation_token.clone())
.send()
.await
.with_context(|| format!("failed to list objects for cleanup under {bucket}/{prefix}"))?;
let objects: Vec<ObjectIdentifier> = response
.contents
.unwrap_or_default()
.into_iter()
.filter_map(|object| object.key.map(|key| ObjectIdentifier::builder().key(key).build().ok()))
.flatten()
.collect();
for chunk in objects.chunks(1_000) {
if chunk.is_empty() {
continue;
}
client
.delete_objects()
.bucket(bucket)
.delete(
Delete::builder()
.set_objects(Some(chunk.to_vec()))
.quiet(true)
.build()
.context("failed to build delete request")?,
)
.send()
.await
.with_context(|| format!("failed to delete cleanup batch under {bucket}/{prefix}"))?;
}
if response.is_truncated.unwrap_or(false) {
continuation_token = response.next_continuation_token;
} else {
break;
}
}
Ok(())
}
fn parse_size_list(input: &str) -> Result<Vec<SizeSpec>> {
input
.split(',')
.map(str::trim)
.filter(|item| !item.is_empty())
.map(parse_size_spec)
.collect()
}
fn parse_size_spec(input: &str) -> Result<SizeSpec> {
let normalized = input.trim();
let lower = normalized.to_ascii_lowercase();
let (number_part, multiplier) = if let Some(value) = lower.strip_suffix("kib") {
(value, 1024usize)
} else if let Some(value) = lower.strip_suffix("mib") {
(value, 1024usize * 1024usize)
} else if let Some(value) = lower.strip_suffix('b') {
(value, 1usize)
} else {
(lower.as_str(), 1usize)
};
let value = number_part
.trim()
.parse::<usize>()
.with_context(|| format!("invalid size component: {input}"))?;
let bytes = value
.checked_mul(multiplier)
.ok_or_else(|| anyhow!("size overflow for {input}"))?;
Ok(SizeSpec {
label: normalized.to_string(),
slug: normalized
.chars()
.filter(|ch| ch.is_ascii_alphanumeric())
.collect::<String>()
.to_ascii_lowercase(),
bytes,
})
}
fn percentile(values: &[f64], percentile: f64) -> Option<f64> {
if values.is_empty() {
return None;
}
let index = ((values.len() - 1) as f64 * percentile).floor() as usize;
values.get(index).copied()
}
fn default_run_id() -> String {
format!("small-put-bench-{}", chrono::Utc::now().format("%Y%m%d-%H%M%S"))
}
fn print_size_summary(summary: &SizeSummary) {
println!(
"{}: success={} failed={} obj/s={:.3} MiB/s={:.3} avg={:.3?} p50={:.3?} p90={:.3?} p99={:.3?}",
summary.label,
summary.succeeded,
summary.failed,
summary.object_rate,
summary.throughput_mib_per_sec,
summary.avg_ms,
summary.p50_ms,
summary.p90_ms,
summary.p99_ms,
);
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn parse_size_spec_supports_binary_units() {
let four_kib = parse_size_spec("4KiB").expect("4KiB should parse");
assert_eq!(four_kib.bytes, 4 * 1024);
let one_mib = parse_size_spec("1MiB").expect("1MiB should parse");
assert_eq!(one_mib.bytes, 1024 * 1024);
}
#[test]
fn percentile_returns_expected_bucket() {
let values = vec![10.0, 20.0, 30.0, 40.0, 50.0];
assert_eq!(percentile(&values, 0.50), Some(30.0));
assert_eq!(percentile(&values, 0.90), Some(40.0));
}
}
+1 -275
View File
@@ -19,13 +19,9 @@
mod tests {
use crate::common::{RustFSTestEnvironment, init_logging};
use aws_sdk_s3::Client;
use aws_sdk_s3::primitives::{ByteStream, SdkBody};
use aws_sdk_s3::primitives::ByteStream;
use aws_sdk_s3::types::{ChecksumAlgorithm, ChecksumMode, CompletedMultipartUpload, CompletedPart};
use base64::Engine;
use bytes::Bytes;
use futures::StreamExt;
use http_body::Frame;
use http_body_util::StreamBody;
use rustfs_rio::{Checksum, ChecksumType as RioChecksumType};
use serial_test::serial;
use sha2::{Digest, Sha256};
@@ -68,53 +64,6 @@ mod tests {
.encoded
}
fn streamed_body_70kib_of_a() -> ByteStream {
let bytes = Bytes::from_static(&[b'a'; 1024]);
let stream = futures::stream::repeat_with(move || {
let frame = Frame::data(bytes.clone());
Ok::<_, std::io::Error>(frame)
});
let body = WithSizeHint::new(StreamBody::new(stream.take(70)), 70 * 1024);
ByteStream::new(SdkBody::from_body_1_x(body))
}
struct WithSizeHint<T> {
inner: T,
size_hint: usize,
}
impl<T> WithSizeHint<T> {
fn new(inner: T, size_hint: usize) -> Self {
Self { inner, size_hint }
}
}
impl<T> http_body::Body for WithSizeHint<T>
where
T: http_body::Body + Unpin,
{
type Data = T::Data;
type Error = T::Error;
fn poll_frame(
self: std::pin::Pin<&mut Self>,
cx: &mut std::task::Context<'_>,
) -> std::task::Poll<Option<Result<Frame<Self::Data>, Self::Error>>> {
let this = self.get_mut();
std::pin::Pin::new(&mut this.inner).poll_frame(cx)
}
fn is_end_stream(&self) -> bool {
self.inner.is_end_stream()
}
fn size_hint(&self) -> http_body::SizeHint {
let mut hint = self.inner.size_hint();
hint.set_exact(self.size_hint as u64);
hint
}
}
/// PutObject with Content-MD5: upload succeeds and GetObject returns same content.
#[tokio::test]
#[serial]
@@ -187,121 +136,6 @@ mod tests {
info!("PASSED: PutObject with checksum_sha256 and GetObject content match");
}
/// Mirrors `s3s-e2e` behavior: only request `checksum_algorithm`, then expect
/// both PutObject and GetObject(checksum_mode=enabled) to expose the same checksum.
#[tokio::test]
#[serial]
async fn test_put_object_with_checksum_algorithm_only() {
init_logging();
info!("TEST: PutObject with checksum_algorithm only");
let mut env = RustFSTestEnvironment::new().await.expect("Failed to create test environment");
env.start_rustfs_server(vec![]).await.expect("Failed to start RustFS");
let client = create_s3_client(&env);
let bucket = "test-checksum-algorithm-only";
create_bucket(&client, bucket).await.expect("Failed to create bucket");
let key = "obj-with-checksum-algorithm-only.txt";
let content = vec![b'a'; 70 * 1024];
let put_resp = client
.put_object()
.bucket(bucket)
.key(key)
.checksum_algorithm(ChecksumAlgorithm::Crc32)
.body(ByteStream::from(content.clone()))
.send()
.await
.expect("PutObject with checksum_algorithm should succeed");
let put_checksum = put_resp
.checksum_crc32()
.expect("PutObject should return checksum_crc32 when checksum_algorithm is used")
.to_string();
let mut get_resp = client
.get_object()
.bucket(bucket)
.key(key)
.checksum_mode(ChecksumMode::Enabled)
.send()
.await
.expect("GetObject should succeed");
let body_bytes = std::mem::replace(&mut get_resp.body, ByteStream::new(aws_sdk_s3::primitives::SdkBody::empty()))
.collect()
.await
.expect("collect body")
.into_bytes();
assert_eq!(body_bytes.as_ref(), content.as_slice(), "GetObject body must match uploaded content");
assert_eq!(
get_resp.checksum_crc32().map(str::to_string),
Some(put_checksum),
"GetObject(checksum_mode=enabled) should expose the stored CRC32 checksum"
);
}
/// Matches the `s3s-e2e` streaming upload shape more closely than `ByteStream::from(Vec<u8>)`.
#[tokio::test]
#[serial]
async fn test_put_object_with_checksum_algorithm_only_streaming_body() {
init_logging();
info!("TEST: PutObject with checksum_algorithm only using streaming body");
let mut env = RustFSTestEnvironment::new().await.expect("Failed to create test environment");
env.start_rustfs_server(vec![]).await.expect("Failed to start RustFS");
let client = create_s3_client(&env);
let bucket = "test-checksum-algorithm-streaming";
create_bucket(&client, bucket).await.expect("Failed to create bucket");
let key = "obj-with-checksum-algorithm-streaming.txt";
let expected_content = vec![b'a'; 70 * 1024];
let put_resp = client
.put_object()
.bucket(bucket)
.key(key)
.checksum_algorithm(ChecksumAlgorithm::Crc32)
.body(streamed_body_70kib_of_a())
.send()
.await
.expect("PutObject with streaming checksum_algorithm should succeed");
let put_checksum = put_resp
.checksum_crc32()
.expect("PutObject should return checksum_crc32 for streaming checksum_algorithm uploads")
.to_string();
let mut get_resp = client
.get_object()
.bucket(bucket)
.key(key)
.checksum_mode(ChecksumMode::Enabled)
.send()
.await
.expect("GetObject should succeed");
let body_bytes = std::mem::replace(&mut get_resp.body, ByteStream::new(SdkBody::empty()))
.collect()
.await
.expect("collect body")
.into_bytes();
assert_eq!(
body_bytes.as_ref(),
expected_content.as_slice(),
"GetObject body must match uploaded content"
);
assert_eq!(
get_resp.checksum_crc32().map(str::to_string),
Some(put_checksum),
"GetObject(checksum_mode=enabled) should expose the stored CRC32 checksum for streaming uploads"
);
}
/// Multipart upload with checksum: CreateMultipartUpload, UploadPart(s) with checksum_sha256, CompleteMultipartUpload; then GetObject verifies content.
/// Uses part size >= 5MB (server minimum) for two parts.
#[tokio::test]
@@ -400,114 +234,6 @@ mod tests {
info!("PASSED: MultipartUpload with checksum and GetObject content match");
}
/// Mirrors `s3s-e2e` multipart behavior: request checksum algorithm at MPU creation,
/// rely on auto checksum handling during UploadPart, and expect CompleteMultipartUpload to succeed.
#[tokio::test]
#[serial]
async fn test_multipart_upload_with_crc32_algorithm_only() {
init_logging();
info!("TEST: MultipartUpload with checksum_algorithm only (CRC32)");
let mut env = RustFSTestEnvironment::new().await.expect("Failed to create test environment");
env.start_rustfs_server(vec![]).await.expect("Failed to start RustFS");
let client = create_s3_client(&env);
let bucket = "test-multipart-checksum-crc32-auto";
create_bucket(&client, bucket).await.expect("Failed to create bucket");
let key = "multipart-with-crc32-auto.bin";
let part1_content = "a".repeat(5 * 1024 * 1024 + 1);
let part2_content = "b".repeat(1024);
let create_resp = client
.create_multipart_upload()
.bucket(bucket)
.key(key)
.checksum_algorithm(ChecksumAlgorithm::Crc32)
.send()
.await
.expect("CreateMultipartUpload should succeed");
let upload_id = create_resp.upload_id().expect("upload_id should be present");
let part1_resp = client
.upload_part()
.bucket(bucket)
.key(key)
.upload_id(upload_id)
.part_number(1)
.body(ByteStream::from(part1_content.clone().into_bytes()))
.send()
.await
.expect("UploadPart 1 should succeed");
let part1_checksum = part1_resp
.checksum_crc32()
.expect("UploadPart 1 should return checksum_crc32")
.to_string();
let part2_resp = client
.upload_part()
.bucket(bucket)
.key(key)
.upload_id(upload_id)
.part_number(2)
.body(ByteStream::from(part2_content.clone().into_bytes()))
.send()
.await
.expect("UploadPart 2 should succeed");
let part2_checksum = part2_resp
.checksum_crc32()
.expect("UploadPart 2 should return checksum_crc32")
.to_string();
let completed_upload = CompletedMultipartUpload::builder()
.parts(
CompletedPart::builder()
.part_number(1)
.e_tag(part1_resp.e_tag().expect("etag part 1"))
.checksum_crc32(part1_checksum)
.build(),
)
.parts(
CompletedPart::builder()
.part_number(2)
.e_tag(part2_resp.e_tag().expect("etag part 2"))
.checksum_crc32(part2_checksum)
.build(),
)
.build();
client
.complete_multipart_upload()
.bucket(bucket)
.key(key)
.upload_id(upload_id)
.multipart_upload(completed_upload)
.send()
.await
.expect("CompleteMultipartUpload should succeed");
let body_bytes = client
.get_object()
.bucket(bucket)
.key(key)
.send()
.await
.expect("GetObject should succeed")
.body
.collect()
.await
.expect("collect body")
.into_bytes();
let expected_content = format!("{part1_content}{part2_content}");
assert_eq!(
body_bytes.as_ref(),
expected_content.as_bytes(),
"completed multipart object must match concatenated parts"
);
}
/// Regression test for issue #2282:
/// CRC64NVME full-object checksum should match between direct PutObject and multipart upload.
#[tokio::test]
-96
View File
@@ -344,9 +344,6 @@ async fn test_local_kms_multipart_upload() {
test_multipart_upload_with_sse_c(&s3_client, TEST_BUCKET)
.await
.expect("SSE-C multipart upload test failed");
test_multipart_download_with_wrong_sse_c_key_fails(&s3_client, TEST_BUCKET)
.await
.expect("SSE-C multipart wrong-key download test failed");
// Test 4: Large multipart upload (test streaming encryption with multiple blocks)
// TODO: Re-enable after fixing streaming encryption issues with large files
@@ -651,99 +648,6 @@ async fn test_multipart_upload_with_sse_c(
Ok(())
}
async fn test_multipart_download_with_wrong_sse_c_key_fails(
s3_client: &aws_sdk_s3::Client,
bucket: &str,
) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
let object_key = "multipart-sse-c-bad-download-test";
let part_size = 5 * 1024 * 1024;
let total_parts = 2;
let total_size = part_size * total_parts;
let encryption_key = "01234567890123456789012345678901";
let key_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, encryption_key);
let key_md5 = sse_customer_key_md5_base64(encryption_key);
let wrong_key = "abcdefghijklmnopqrstuvwxyz012345";
let wrong_key_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, wrong_key);
let wrong_key_md5 = sse_customer_key_md5_base64(wrong_key);
let test_data: Vec<u8> = (0..total_size).map(|i| ((i * 5) % 256) as u8).collect();
let create_multipart_output = s3_client
.create_multipart_upload()
.bucket(bucket)
.key(object_key)
.sse_customer_algorithm("AES256")
.sse_customer_key(&key_b64)
.sse_customer_key_md5(&key_md5)
.send()
.await?;
let upload_id = create_multipart_output.upload_id().unwrap();
let mut completed_parts = Vec::new();
for part_number in 1..=total_parts {
let start = (part_number - 1) * part_size;
let end = std::cmp::min(start + part_size, total_size);
let part_data = &test_data[start..end];
let upload_part_output = s3_client
.upload_part()
.bucket(bucket)
.key(object_key)
.upload_id(upload_id)
.part_number(part_number as i32)
.body(aws_sdk_s3::primitives::ByteStream::from(part_data.to_vec()))
.sse_customer_algorithm("AES256")
.sse_customer_key(&key_b64)
.sse_customer_key_md5(&key_md5)
.send()
.await?;
completed_parts.push(
aws_sdk_s3::types::CompletedPart::builder()
.part_number(part_number as i32)
.e_tag(upload_part_output.e_tag().unwrap())
.build(),
);
}
let completed_multipart_upload = aws_sdk_s3::types::CompletedMultipartUpload::builder()
.set_parts(Some(completed_parts))
.build();
s3_client
.complete_multipart_upload()
.bucket(bucket)
.key(object_key)
.upload_id(upload_id)
.multipart_upload(completed_multipart_upload)
.send()
.await?;
let err = s3_client
.get_object()
.bucket(bucket)
.key(object_key)
.sse_customer_algorithm("AES256")
.sse_customer_key(&wrong_key_b64)
.sse_customer_key_md5(&wrong_key_md5)
.send()
.await
.expect_err("multipart SSE-C download with the wrong key should fail");
let service_err = err.into_service_error();
assert_eq!(
service_err.meta().code(),
Some("InvalidRequest"),
"wrong-key multipart SSE-C download should return InvalidRequest, got {:?}",
service_err.meta().code()
);
Ok(())
}
/// Test large multipart upload to verify streaming encryption works correctly
#[allow(dead_code)]
async fn test_large_multipart_upload(
-4
View File
@@ -100,10 +100,6 @@ mod cluster_concurrency_test;
#[cfg(test)]
mod checksum_upload_test;
// Range request regression tests
#[cfg(test)]
mod range_request_test;
// Group deletion tests
#[cfg(test)]
mod group_delete_test;
-177
View File
@@ -1,177 +0,0 @@
// Copyright 2024 RustFS Team
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//! End-to-end regression tests for invalid and suffix GET object ranges.
#[cfg(test)]
mod tests {
use crate::common::{RustFSTestEnvironment, init_logging};
use aws_sdk_s3::Client;
use aws_sdk_s3::error::SdkError;
use aws_sdk_s3::primitives::ByteStream;
use serial_test::serial;
use tracing::info;
fn create_s3_client(env: &RustFSTestEnvironment) -> Client {
env.create_s3_client()
}
async fn create_bucket(client: &Client, bucket: &str) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
match client.create_bucket().bucket(bucket).send().await {
Ok(_) => Ok(()),
Err(err) => {
if err.to_string().contains("BucketAlreadyOwnedByYou") || err.to_string().contains("BucketAlreadyExists") {
Ok(())
} else {
Err(Box::new(err))
}
}
}
}
#[tokio::test]
#[serial]
async fn test_get_object_invalid_range_returns_416_issue_s3_implemented_tests() {
init_logging();
info!("TEST: GetObject invalid range should return InvalidRange/416");
let mut env = RustFSTestEnvironment::new().await.expect("Failed to create test environment");
env.start_rustfs_server(vec![]).await.expect("Failed to start RustFS");
let client = create_s3_client(&env);
let bucket = "test-invalid-range";
let key = "range.txt";
let content = b"testcontent";
create_bucket(&client, bucket).await.expect("Failed to create bucket");
client
.put_object()
.bucket(bucket)
.key(key)
.body(ByteStream::from_static(content))
.send()
.await
.expect("PutObject should succeed");
let result = client.get_object().bucket(bucket).key(key).range("bytes=40-50").send().await;
let err = result.expect_err("GetObject with an unsatisfiable range should fail");
match err {
SdkError::ServiceError(service_err) => {
assert_eq!(service_err.raw().status().as_u16(), 416, "invalid range should return HTTP 416");
let s3_err = service_err.into_err();
assert_eq!(s3_err.meta().code(), Some("InvalidRange"), "invalid range should map to InvalidRange");
}
other_err => panic!("Expected S3 service error, got: {other_err:?}"),
}
}
#[tokio::test]
#[serial]
async fn test_get_object_suffix_byte_range_returns_correct_body() {
init_logging();
info!("TEST: GetObject suffix byte-range should return correct body");
let mut env = RustFSTestEnvironment::new().await.expect("Failed to create test environment");
env.start_rustfs_server(vec![]).await.expect("Failed to start RustFS");
let client = create_s3_client(&env);
let bucket = "test-suffix-range";
let key = "range-suffix.bin";
// ~3 MB so the object spans multiple erasure blocks (block_size = 1 MB).
// Suffix ranges on single-block objects never hit the bug.
let file_size: usize = 3_095_910;
let content: Vec<u8> = (0..file_size).map(|i| (i % 256) as u8).collect();
create_bucket(&client, bucket).await.expect("Failed to create bucket");
client
.put_object()
.bucket(bucket)
.key(key)
.body(ByteStream::from(content.clone()))
.send()
.await
.expect("PutObject should succeed");
// bytes=-8 — last 8 bytes (parquet footer read)
let result = client
.get_object()
.bucket(bucket)
.key(key)
.range("bytes=-8")
.send()
.await
.expect("bytes=-8 should succeed");
let body = result.body.collect().await.expect("bytes=-8 body").into_bytes();
assert_eq!(body.len(), 8, "bytes=-8 body length");
assert_eq!(&body[..], &content[file_size - 8..], "bytes=-8 body content");
// bytes=-96 — last 96 bytes
let result = client
.get_object()
.bucket(bucket)
.key(key)
.range("bytes=-96")
.send()
.await
.expect("bytes=-96 should succeed");
let body = result.body.collect().await.expect("bytes=-96 body").into_bytes();
assert_eq!(body.len(), 96, "bytes=-96 body length");
assert_eq!(&body[..], &content[file_size - 96..], "bytes=-96 body content");
// bytes=-1 — last byte
let result = client
.get_object()
.bucket(bucket)
.key(key)
.range("bytes=-1")
.send()
.await
.expect("bytes=-1 should succeed");
let body = result.body.collect().await.expect("bytes=-1 body").into_bytes();
assert_eq!(body.len(), 1, "bytes=-1 body length");
assert_eq!(body[0], content[file_size - 1], "bytes=-1 body content");
// bytes=0-7 — absolute range (regression guard)
let result = client
.get_object()
.bucket(bucket)
.key(key)
.range("bytes=0-7")
.send()
.await
.expect("bytes=0-7 should succeed");
let body = result.body.collect().await.expect("bytes=0-7 body").into_bytes();
assert_eq!(body.len(), 8, "bytes=0-7 body length");
assert_eq!(&body[..], &content[0..8], "bytes=0-7 body content");
// Equivalent absolute range for the same last-8-bytes window
let start = file_size - 8;
let end = file_size - 1;
let range = format!("bytes={start}-{end}");
let result = client
.get_object()
.bucket(bucket)
.key(key)
.range(&range)
.send()
.await
.expect("absolute tail range should succeed");
let body = result.body.collect().await.expect("absolute tail body").into_bytes();
assert_eq!(body.len(), 8, "absolute tail body length");
assert_eq!(&body[..], &content[start..], "absolute tail body content");
}
}
-1
View File
@@ -70,7 +70,6 @@ reed-solomon-erasure = { workspace = true }
reed-solomon-simd = { workspace = true }
lazy_static.workspace = true
rustfs-lock.workspace = true
rustfs-io-core.workspace = true
rustfs-io-metrics.workspace = true
regex = { workspace = true }
path-absolutize = { workspace = true }
-4
View File
@@ -32,10 +32,6 @@
For comprehensive documentation, examples, and usage guides, please visit the main [RustFS repository](https://github.com/rustfs/rustfs).
## 📈 Benchmarks
ECStore ships Criterion benchmarks under [`crates/ecstore/benches/`](./benches/) for the remaining erasure-code and comparison workloads. Re-run benchmarks on your target machine before treating any local measurement as a regression baseline.
## 📄 License
This project is licensed under the Apache License 2.0 - see the [LICENSE](../../LICENSE) file for details.
+9 -32
View File
@@ -21,8 +21,6 @@ use std::time::Instant;
use tokio::io::AsyncRead;
use tracing::debug;
const BITROT_READ_OPERATION: &str = "bitrot_read";
/// Create a BitrotReader from either inline data or disk file stream
///
/// # Parameters
@@ -67,38 +65,20 @@ pub async fn create_bitrot_reader(
} else if let Some(disk) = disk {
// Read from disk
if use_zero_copy {
if !disk.is_local() {
rustfs_io_metrics::record_io_path_selected(BITROT_READ_OPERATION, rustfs_io_metrics::IoPath::Legacy);
rustfs_io_metrics::record_io_fallback(
rustfs_io_metrics::IoStage::ReadSetup,
rustfs_io_metrics::FallbackReason::NonLocalBackend,
);
let rd = disk.read_file_stream(bucket, path, offset, length).await?;
let reader = BitrotReader::new(rd, shard_size, checksum_algo, skip_verify);
return Ok(Some(reader));
}
// Try zero-copy read first (uses mmap on Unix)
let start = Instant::now();
match disk.read_file_zero_copy(bucket, path, offset, length).await {
Ok(bytes) => {
let duration_ms = start.elapsed().as_secs_f64() * 1000.0;
rustfs_io_metrics::record_io_path_selected(BITROT_READ_OPERATION, rustfs_io_metrics::IoPath::Fast);
// `read_file_zero_copy()` returns a shared `Bytes` view, which preserves the
// mmap-backed fast path without exposing chunk-native GET internals.
rustfs_io_metrics::record_io_copy_mode(
BITROT_READ_OPERATION,
rustfs_io_metrics::CopyMode::SharedBytes,
bytes.len(),
);
// Record zero-copy metrics
rustfs_io_metrics::record_zero_copy_read(bytes.len(), duration_ms);
// Log successful zero-copy read
debug!(
size = bytes.len(),
duration_ms,
path = %path,
"bitrot_fast_read_success"
"zero_copy_read_success"
);
// Wrap Bytes in Cursor for AsyncRead
@@ -113,16 +93,14 @@ pub async fn create_bitrot_reader(
Ok(Some(reader))
}
Err(e) => {
rustfs_io_metrics::record_io_path_selected(BITROT_READ_OPERATION, rustfs_io_metrics::IoPath::Legacy);
rustfs_io_metrics::record_io_fallback(
rustfs_io_metrics::IoStage::ReadSetup,
rustfs_io_metrics::FallbackReason::Unknown,
);
// Record zero-copy fallback
rustfs_io_metrics::record_zero_copy_fallback(&format!("{:?}", e));
// Log zero-copy fallback
debug!(
reason = %e,
reason = %format!("{:?}", e),
path = %path,
"bitrot_fast_read_fallback"
"zero_copy_fallback"
);
// Fall back to regular stream read on error
@@ -139,7 +117,6 @@ pub async fn create_bitrot_reader(
}
}
} else {
rustfs_io_metrics::record_io_path_selected(BITROT_READ_OPERATION, rustfs_io_metrics::IoPath::Legacy);
// Use regular stream read
match disk.read_file_stream(bucket, path, offset, length).await {
Ok(rd) => {
+1 -5
View File
@@ -24,7 +24,7 @@ use rustfs_utils::http::headers::{
AMZ_STORAGE_CLASS,
};
use rustfs_utils::http::{
AMZ_BUCKET_REPLICATION_STATUS, SUFFIX_CRC, SUFFIX_DATA_MOV, SUFFIX_HEALING, SUFFIX_PURGESTATUS, SUFFIX_REPLICA_STATUS,
AMZ_BUCKET_REPLICATION_STATUS, SUFFIX_DATA_MOV, SUFFIX_HEALING, SUFFIX_PURGESTATUS, SUFFIX_REPLICA_STATUS,
SUFFIX_REPLICA_TIMESTAMP, SUFFIX_REPLICATION_STATUS, SUFFIX_REPLICATION_TIMESTAMP, has_internal_suffix, insert_bytes,
is_internal_key,
};
@@ -230,10 +230,6 @@ impl FileMeta {
}
}
if let Some(checksum) = fi.checksum.as_ref() {
insert_bytes(&mut obj.meta_sys, SUFFIX_CRC, checksum.to_vec());
}
if let Some(mod_time) = fi.mod_time {
obj.mod_time = Some(mod_time);
}
+2 -2
View File
@@ -381,7 +381,7 @@ where
let hash_reader = HashReader::from_stream(buf_reader, content_length, content_length, None, None, false)
.map_err(|e| sanitize_storage_error("Hash reader creation", e))?;
// 15. Hand the hash reader to the chunk-native PUT data wrapper
// 15. Wrap in PutObjReader as expected by storage layer
let mut put_reader = PutObjReader::new(hash_reader);
// 16. Upload object to storage
@@ -464,7 +464,7 @@ where
let hash_reader = HashReader::from_stream(buf_reader, content_length, content_length, None, None, false)
.map_err(|e| sanitize_storage_error("Hash reader creation", e))?;
// Hand the hash reader to the chunk-native PUT data wrapper
// Wrap in PutObjReader
let mut put_reader = PutObjReader::new(hash_reader);
// Upload object to storage
+5 -41
View File
@@ -29,21 +29,11 @@ pub const RUSTFS_MULTIPART_CHECKSUM: &str = "x-rustfs-multipart-checksum";
/// RustFS multipart checksum type metadata key
pub const RUSTFS_MULTIPART_CHECKSUM_TYPE: &str = "x-rustfs-multipart-checksum-type";
const AMZ_CHECKSUM_ALGORITHM: &str = "x-amz-checksum-algorithm";
const AMZ_SDK_CHECKSUM_ALGORITHM: &str = "x-amz-sdk-checksum-algorithm";
/// Checksum type enumeration with flags
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub struct ChecksumType(pub u32);
impl ChecksumType {
fn algorithm_from_headers(headers: &HeaderMap) -> Option<&str> {
headers
.get(AMZ_CHECKSUM_ALGORITHM)
.and_then(|v| v.to_str().ok())
.or_else(|| headers.get(AMZ_SDK_CHECKSUM_ALGORITHM).and_then(|v| v.to_str().ok()))
}
/// Checksum will be sent in trailing header
pub const TRAILING: ChecksumType = ChecksumType(1 << 0);
@@ -166,7 +156,10 @@ impl ChecksumType {
pub fn from_header(headers: &HeaderMap) -> Self {
Self::from_string_with_obj_type(
Self::algorithm_from_headers(headers).unwrap_or(""),
headers
.get("x-amz-checksum-algorithm")
.and_then(|v| v.to_str().ok())
.unwrap_or(""),
headers.get("x-amz-checksum-type").and_then(|v| v.to_str().ok()).unwrap_or(""),
)
}
@@ -580,7 +573,7 @@ pub fn get_content_checksum(headers: &HeaderMap) -> Result<Option<Checksum>, std
fn get_content_checksum_direct(headers: &HeaderMap) -> (ChecksumType, String) {
let mut checksum_type = ChecksumType::NONE;
if let Some(alg) = ChecksumType::algorithm_from_headers(headers) {
if let Some(alg) = headers.get("x-amz-checksum-algorithm").and_then(|v| v.to_str().ok()) {
checksum_type = ChecksumType::from_string_with_obj_type(
alg,
headers.get("x-amz-checksum-type").and_then(|s| s.to_str().ok()).unwrap_or(""),
@@ -1139,35 +1132,6 @@ fn crc64_combine(poly: u64, crc1: u64, crc2: u64, len2: i64) -> u64 {
#[cfg(test)]
mod tests {
use super::{Checksum, ChecksumType};
use http::HeaderMap;
#[test]
fn algorithm_from_headers_parses_amz_checksum_algorithm() {
let mut headers = HeaderMap::new();
headers.insert("x-amz-checksum-algorithm", "SHA256".parse().expect("valid header value"));
assert_eq!(ChecksumType::algorithm_from_headers(&headers), Some("SHA256"));
}
#[test]
fn algorithm_from_headers_falls_back_to_sdk_header() {
let mut headers = HeaderMap::new();
headers.insert("x-amz-sdk-checksum-algorithm", "CRC32C".parse().expect("valid header value"));
assert_eq!(ChecksumType::algorithm_from_headers(&headers), Some("CRC32C"));
}
#[test]
fn algorithm_from_headers_prefers_amz_over_sdk() {
let mut headers = HeaderMap::new();
headers.insert("x-amz-checksum-algorithm", "SHA256".parse().expect("valid header value"));
headers.insert("x-amz-sdk-checksum-algorithm", "CRC32C".parse().expect("valid header value"));
assert_eq!(ChecksumType::algorithm_from_headers(&headers), Some("SHA256"));
}
#[test]
fn algorithm_from_headers_returns_none_when_missing() {
let headers = HeaderMap::new();
assert_eq!(ChecksumType::algorithm_from_headers(&headers), None);
}
#[test]
fn crc64_nvme_add_part_matches_full_object_checksum() {