Compare commits

..

1 Commits

Author SHA1 Message Date
唐小鸭 741a1cac59 fix(filemeta): keep data dir of a version awaiting purge replication 2026-09-07 03:29:15 +08:00
6 changed files with 140 additions and 265 deletions
@@ -3908,89 +3908,82 @@ async fn test_bucket_replication_replicates_directory_marker_in_versioned_bucket
Ok(())
}
/// Regression for rustfs/backlog#2340 (not Wasabi specific): a single-part
/// object uploaded with `x-amz-checksum-*` must reach the target with the same
/// checksum. The outbound options keyed the stored record by algorithm name,
/// which the target client sent as `x-amz-meta-*` user metadata, so a replica
/// never carried a checksum although the source HEAD returned one.
/// Regression for rustfs/backlog#2340 (not Wasabi specific): permanently
/// deleting a version whose payload lives in a data dir must leave the source
/// clean once the purge replicates. Managed-SSE objects are never inlined and a
/// plain object above the inline threshold takes the same layout. The version
/// retained with a pending purge used to lose its data dir, so the purge state
/// could never be applied (`VersionNotFound` on every retry) and the bucket
/// stayed `BucketNotEmpty` while `ListObjectVersions` was already empty.
#[tokio::test]
async fn test_bucket_replication_forwards_single_part_object_checksums() -> TestResult {
async fn test_bucket_replication_version_purge_of_non_inline_object_releases_source_bucket() -> TestResult {
init_logging();
let mut source_env = RustFSTestEnvironment::new().await?;
let mut source_env_vars = replication_fast_env();
source_env_vars.extend_from_slice(LOOPBACK_REPLICATION_TARGET_ENV);
source_env.start_rustfs_server_with_env(vec![], &source_env_vars).await?;
let mut target_env = RustFSTestEnvironment::new().await?;
target_env.start_rustfs_server_without_cleanup(vec![]).await?;
let source_bucket = "replication-checksum-src";
let target_bucket = "replication-checksum-dst";
let (source_env, target_env, source_bucket, target_bucket) = build_sse_replication_pair("purge-datadir", true, true).await?;
let target_arn = wait_for_remote_target_arn(&source_env, &source_bucket).await?;
put_bucket_replication_with_delete_statuses(&source_env, &source_bucket, &target_arn, "Enabled", Some("Enabled")).await?;
let source_client = source_env.create_s3_client();
let target_client = target_env.create_s3_client();
source_client.create_bucket().bucket(source_bucket).send().await?;
target_client.create_bucket().bucket(target_bucket).send().await?;
enable_bucket_versioning(&source_env, source_bucket).await?;
enable_bucket_versioning(&target_env, target_bucket).await?;
let target_arn = set_replication_target(&source_env, source_bucket, &target_env, target_bucket).await?;
put_bucket_replication(&source_env, source_bucket, &target_arn).await?;
let body = b"123456789";
let crc32_key = "checksum-crc32.txt";
let sha256_key = "checksum-sha256.txt";
let crc32_put = source_client
let sse_key = "sse-object.bin";
let large_key = "large-object.bin";
let sse_put = source_client
.put_object()
.bucket(source_bucket)
.key(crc32_key)
.body(ByteStream::from_static(body))
.checksum_algorithm(aws_sdk_s3::types::ChecksumAlgorithm::Crc32)
.bucket(&source_bucket)
.key(sse_key)
.body(ByteStream::from_static(b"encrypted source payload"))
.server_side_encryption(ServerSideEncryption::Aes256)
.send()
.await?;
let expected_crc32 = crc32_put.checksum_crc32().ok_or("source PUT omitted CRC32")?.to_string();
let sha256_put = source_client
let large_put = source_client
.put_object()
.bucket(source_bucket)
.key(sha256_key)
.body(ByteStream::from_static(body))
.checksum_algorithm(aws_sdk_s3::types::ChecksumAlgorithm::Sha256)
.bucket(&source_bucket)
.key(large_key)
.body(ByteStream::from(vec![0x5a; 2 * 1024 * 1024]))
.send()
.await?;
let expected_sha256 = sha256_put.checksum_sha256().ok_or("source PUT omitted SHA256")?.to_string();
let purged = [
(sse_key, sse_put.version_id().ok_or("SSE PUT omitted version ID")?.to_string()),
(large_key, large_put.version_id().ok_or("large PUT omitted version ID")?.to_string()),
];
assert_replication_converged(&source_client, &source_bucket, &target_client, &target_bucket).await?;
for key in [crc32_key, sha256_key] {
wait_for_source_replication_status(&source_client, source_bucket, key, "COMPLETED", false).await?;
for (key, version_id) in &purged {
source_client
.delete_object()
.bucket(&source_bucket)
.key(*key)
.version_id(version_id)
.send()
.await?;
}
assert_replication_converged(&source_client, &source_bucket, &target_client, &target_bucket).await?;
let target_state = list_replication_state(&target_client, &target_bucket).await?;
assert!(target_state.is_empty(), "target retained an explicitly purged version: {target_state:?}");
let replica = target_client
.head_object()
.bucket(target_bucket)
.key(crc32_key)
.checksum_mode(aws_sdk_s3::types::ChecksumMode::Enabled)
.send()
.await?;
assert_eq!(replica.checksum_crc32(), Some(expected_crc32.as_str()), "replica lost the CRC32 checksum");
let replica = target_client
.head_object()
.bucket(target_bucket)
.key(sha256_key)
.checksum_mode(aws_sdk_s3::types::ChecksumMode::Enabled)
.send()
.await?;
assert_eq!(
replica.checksum_sha256(),
Some(expected_sha256.as_str()),
"replica lost the SHA256 checksum"
);
// The bare algorithm name must not leak as user metadata either.
assert!(
replica
.metadata()
.is_none_or(|meta| !meta.keys().any(|k| k.eq_ignore_ascii_case("sha256"))),
"replica carries the checksum as user metadata: {:?}",
replica.metadata()
);
// The purge state is applied on the source asynchronously after the target
// acknowledges the delete; only then does the retained version go away and
// the bucket become deletable. A listing that is empty while DeleteBucket
// keeps answering BucketNotEmpty is exactly the regression.
let deadline = tokio::time::Instant::now() + Duration::from_secs(60);
loop {
let listing = source_client.list_object_versions().bucket(&source_bucket).send().await?;
let listed = listing.versions().len() + listing.delete_markers().len();
match source_client.delete_bucket().bucket(&source_bucket).send().await {
Ok(_) => break,
Err(err) if err.code() == Some("BucketNotEmpty") => {
if tokio::time::Instant::now() >= deadline {
return Err(format!(
"source bucket stayed BucketNotEmpty after the version purge replicated; \
ListObjectVersions shows {listed} entries"
)
.into());
}
sleep(Duration::from_millis(500)).await;
}
Err(err) => return Err(err.into()),
}
}
Ok(())
}
@@ -42,8 +42,7 @@ use crate::replication_extension_test::{
use aws_sdk_s3::Client;
use aws_sdk_s3::primitives::{ByteStream, DateTime};
use aws_sdk_s3::types::{
Checksum, ChecksumAlgorithm, CompletedMultipartUpload, CompletedPart, ObjectAttributes, ObjectLockLegalHoldStatus,
ObjectLockMode,
Checksum, CompletedMultipartUpload, CompletedPart, ObjectAttributes, ObjectLockLegalHoldStatus, ObjectLockMode,
};
use bytes::Bytes;
use std::error::Error;
@@ -115,13 +114,10 @@ enum ObjectShape {
LockedMultipart,
/// ODM stores two local parts while preserving a single-PUT source's MD5 ETag.
OdmPreservedMd5Multipart,
/// Single-part object uploaded with `x-amz-checksum-sha256`; the replica
/// must carry the same header (rustfs/backlog#2340).
Checksummed,
}
impl ObjectShape {
const ALL: [ObjectShape; 8] = [
const ALL: [ObjectShape; 7] = [
ObjectShape::Empty,
ObjectShape::Plain,
ObjectShape::Retention,
@@ -129,7 +125,6 @@ impl ObjectShape {
ObjectShape::Multipart,
ObjectShape::LockedMultipart,
ObjectShape::OdmPreservedMd5Multipart,
ObjectShape::Checksummed,
];
fn key(self) -> &'static str {
@@ -141,16 +136,6 @@ impl ObjectShape {
ObjectShape::Multipart => "matrix/multipart.bin",
ObjectShape::LockedMultipart => "matrix/locked-multipart.bin",
ObjectShape::OdmPreservedMd5Multipart => "matrix/odm-preserved-md5.bin",
ObjectShape::Checksummed => "matrix/checksummed.bin",
}
}
/// The `x-amz-checksum-*` header the source stored and every upload of
/// the replica must repeat.
fn forwarded_checksum_header(self) -> Option<&'static str> {
match self {
ObjectShape::Checksummed => Some("x-amz-checksum-sha256"),
_ => None,
}
}
@@ -213,18 +198,6 @@ impl ObjectShape {
ObjectShape::Multipart => multipart_put(client, bucket, key, 0x44, false).await,
ObjectShape::LockedMultipart => multipart_put(client, bucket, key, 0x55, true).await,
ObjectShape::OdmPreservedMd5Multipart => odm_preserved_md5_multipart(env, bucket, key).await,
ObjectShape::Checksummed => {
let body = payload(40 * 1024, 0x66);
client
.put_object()
.bucket(bucket)
.key(key)
.body(ByteStream::from(body.clone()))
.checksum_algorithm(ChecksumAlgorithm::Sha256)
.send()
.await?;
Ok(body)
}
}
}
}
@@ -477,19 +450,6 @@ async fn check_completed_cell(
}) {
return Err(format!("a locked PutObject went out without any integrity header (rustfs#7082): {bare:?}").into());
}
// rustfs/backlog#2340 contract: a source checksum reaches the target as
// the `x-amz-checksum-*` header, not as user metadata; every PutObject of
// the shape carries it.
if let Some(header) = shape.forwarded_checksum_header()
&& let Some(missing) = uploads.iter().find(|record| {
record.operation == FakeTargetOperation::PutObject
&& !record.transport.checksum_headers.iter().any(|name| name == header)
})
{
return Err(
format!("a PutObject went out without the source's {header} header (rustfs/backlog#2340): {missing:?}").into(),
);
}
Ok(())
}
+7 -76
View File
@@ -2064,15 +2064,7 @@ impl TargetClient {
}
}
// A forwarded source checksum is this PUT's integrity header. In
// streaming-checksum mode (`RUSTFS_REPLICATION_STREAMING_CHECKSUMS`)
// the SDK would still add its default CRC32 trailer, and a target that
// receives both keeps the trailer's algorithm: a forwarded SHA256
// vanished from the replica while the source reported COMPLETED. Pin
// this request to WhenRequired so nothing is sent beside the source's
// own checksum.
let forwards_source_checksum = headers.keys().any(|name| name.as_str().starts_with("x-amz-checksum-"));
let mut operation = builder
match builder
.bucket(bucket)
.key(object)
.content_length(size)
@@ -2092,14 +2084,10 @@ impl TargetClient {
}
Result::<_, aws_smithy_types::error::operation::BuildError>::Ok(req)
});
if forwards_source_checksum {
operation = operation.config_override(
aws_sdk_s3::config::Builder::new()
.request_checksum_calculation(aws_sdk_s3::config::RequestChecksumCalculation::WhenRequired),
);
}
match operation.send().await {
})
.send()
.await
{
Ok(output) => {
// Under SSE-KMS/DSSE or SSE-C the target's ETag is not the MD5
// of the stored plaintext, so it cannot be compared against the
@@ -2519,21 +2507,13 @@ mod tests {
}
fn header_recording_target_client(response_headers: Vec<(String, String)>) -> (TargetClient, RecordedHeaders) {
header_recording_target_client_with_checksums(response_headers, replication_request_checksum_calculation())
}
fn header_recording_target_client_with_checksums(
response_headers: Vec<(String, String)>,
checksums: RequestChecksumCalculation,
) -> (TargetClient, RecordedHeaders) {
let request_headers: RecordedHeaders = Arc::new(std::sync::Mutex::new(Vec::new()));
let connector = SharedHttpConnector::new(RecordingHeaderConnector {
request_headers: Arc::clone(&request_headers),
response_headers,
});
let http_client = http_client_fn(move |_settings, _components| connector.clone());
let client =
s3_client_for_endpoint_test_with_checksums("https://localhost:443".to_string(), Some(http_client), checksums);
let client = s3_client_for_test(443, Some(http_client));
(
TargetClient {
endpoint: "https://localhost:443".to_string(),
@@ -2700,47 +2680,6 @@ mod tests {
}
}
/// With streaming checksums enabled the SDK adds a CRC32 trailer to every
/// upload. A PUT that forwards the source's checksum must not get that
/// second algorithm: a target that receives both keeps the trailer's and
/// the forwarded SHA256 never reaches the replica (rustfs/backlog#2340).
#[tokio::test]
async fn streaming_put_object_with_forwarded_checksum_sends_no_sdk_checksum() {
let (client, recorded) =
header_recording_target_client_with_checksums(Vec::new(), RequestChecksumCalculation::WhenSupported);
let mut forwarded = PutObjectOptions::default();
forwarded.user_metadata.insert(
"x-amz-checksum-sha256".to_string(),
"OoJ3yNhRwv3wwtZoGqEIPrPX9xwTnfLl+ka0wStN1g0=".to_string(),
);
client
.put_object("target-bucket", "object", 4, streaming_test_body(b"data"), &forwarded)
.await
.expect("recorded put_object should succeed");
client
.put_object("target-bucket", "object", 4, streaming_test_body(b"data"), &PutObjectOptions::default())
.await
.expect("recorded put_object should succeed");
let recorded = recorded.lock().expect("recorded header lock should not be poisoned");
let with_forwarded = &recorded[0];
assert_eq!(
recorded_header(with_forwarded, "x-amz-checksum-sha256"),
Some("OoJ3yNhRwv3wwtZoGqEIPrPX9xwTnfLl+ka0wStN1g0=")
);
assert_eq!(
recorded_header(with_forwarded, "x-amz-trailer"),
None,
"the SDK must not add a trailer checksum"
);
assert_eq!(recorded_header(with_forwarded, "x-amz-sdk-checksum-algorithm"), None);
// Control: the same client still streams a trailer when nothing is forwarded.
let without_forwarded = &recorded[1];
assert!(
recorded_header(without_forwarded, "x-amz-trailer").is_some(),
"streaming mode must still apply to uploads without a forwarded checksum: {without_forwarded:?}"
);
}
/// A forwarded source checksum already satisfies the rule; nothing is added.
#[tokio::test]
async fn locked_put_object_keeps_a_forwarded_source_checksum() {
@@ -3106,14 +3045,6 @@ mod tests {
}
fn s3_client_for_endpoint_test(endpoint: String, http_client: Option<SharedHttpClient>) -> S3Client {
s3_client_for_endpoint_test_with_checksums(endpoint, http_client, replication_request_checksum_calculation())
}
fn s3_client_for_endpoint_test_with_checksums(
endpoint: String,
http_client: Option<SharedHttpClient>,
checksums: RequestChecksumCalculation,
) -> S3Client {
let credentials = SdkCredentials::builder()
.access_key_id("test-access")
.secret_access_key("test-secret")
@@ -3127,7 +3058,7 @@ mod tests {
.behavior_version(aws_sdk_s3::config::BehaviorVersion::latest())
// Mirror the production remote-target builder so recorded requests
// exercise the same checksum/framing behavior (#6853).
.request_checksum_calculation(checksums);
.request_checksum_calculation(replication_request_checksum_calculation());
if let Some(http_client) = http_client {
config = config.http_client(http_client);
}
@@ -281,6 +281,12 @@ pub(crate) fn replication_put_object_options(sc: &str, object_info: &ObjectInfo)
// record may only add multipart-ness, never take it away.
is_multipart = base_is_multipart || checksum_record_is_multipart;
for (key, value) in checksum_meta.iter() {
if key != AMZ_CHECKSUM_TYPE {
meta.insert(key.clone(), value.clone());
}
}
if !base_is_multipart
&& checksum_meta
.get(AMZ_CHECKSUM_TYPE)
@@ -288,26 +294,6 @@ pub(crate) fn replication_put_object_options(sc: &str, object_info: &ObjectInfo)
{
is_multipart = false;
}
// The record keys each checksum by algorithm name ("CRC32"); the
// target only reads `x-amz-checksum-<algorithm>`. Inserting the bare
// name here made `PutObjectOptions::header()` send it as user
// metadata (`x-amz-meta-crc32`), so no replica ever carried the
// source checksum (rustfs/backlog#2340). The object-level record
// describes one PUT body: a multipart replica is rebuilt part by
// part, and its CreateMultipartUpload must not announce a checksum
// the parts do not carry, so the record is forwarded on the
// single-PUT route only (MinIO `getCRCMeta` parity).
if !is_multipart {
for (key, value) in checksum_meta.iter() {
if key == AMZ_CHECKSUM_TYPE {
continue;
}
if let Some(header) = rustfs_rio::ChecksumType::from_string(key).key() {
meta.insert(header.to_string(), value.clone());
}
}
}
}
}
@@ -1482,63 +1468,12 @@ mod tests {
..Default::default()
};
let (opts, is_multipart) = replication_put_object_options("", &object_info).expect("build replication put options");
let (opts, _is_multipart) = replication_put_object_options("", &object_info).expect("build replication put options");
assert!(!is_multipart, "{name}: a single-part checksum record must keep the single-PUT route");
let header = ty.key().expect("every forwarded algorithm has an x-amz-checksum header");
assert_eq!(
opts.user_metadata.get(header),
opts.user_metadata.get(name),
Some(&checksum.encoded),
"replication must forward the {name} checksum as the {header} header"
);
assert!(
!opts.user_metadata.contains_key(name),
"{name}: the bare algorithm name would leave as x-amz-meta user metadata"
);
}
}
/// The object-level record of a multipart upload (composite or full-object)
/// must not become a PutObject checksum header: the replica is rebuilt
/// through CreateMultipartUpload/UploadPart, and a checksum announced there
/// that the parts do not carry would be rejected by the target.
#[test]
fn replication_put_object_options_keeps_multipart_checksum_records_off_the_wire() {
let mut composite_type = rustfs_rio::ChecksumType::from_string("crc32");
composite_type
.merge(rustfs_rio::ChecksumType::MULTIPART)
.merge(rustfs_rio::ChecksumType::INCLUDES_MULTIPART);
let mut combined = Vec::new();
for part in [b"part-one".as_slice(), b"part-two".as_slice()] {
let part_checksum =
rustfs_rio::Checksum::new_from_data(rustfs_rio::ChecksumType::from_string("crc32"), part).expect("part checksum");
combined.extend_from_slice(part_checksum.raw.as_slice());
}
let composite = rustfs_rio::Checksum::new_from_data(composite_type, &combined)
.expect("composite checksum")
.to_bytes(&combined);
for (label, checksum, etag) in [
("composite", composite, "0123456789abcdef0123456789abcdef-2"),
(
"full-object",
full_object_multipart_checksum_record(),
"0123456789abcdef0123456789abcdef-3",
),
] {
let object_info = ObjectInfo {
etag: Some(etag.to_string()),
checksum: Some(checksum),
..Default::default()
};
let (opts, is_multipart) = replication_put_object_options("", &object_info).expect("build replication put options");
assert!(is_multipart, "{label}: a multipart object must keep the multipart route");
assert!(
opts.user_metadata
.keys()
.all(|key| !key.starts_with("x-amz-checksum-") && key != "CRC32"),
"{label}: no object-level checksum may reach the target's CreateMultipartUpload: {:?}",
opts.user_metadata
"replication must forward the {name} checksum into user_metadata identically to the classic algorithms"
);
}
}
+59 -2
View File
@@ -692,10 +692,15 @@ impl FileMeta {
}
}
let old_dir = v.object.as_ref().map(|v| v.data_dir).unwrap_or_default();
// The version stays on disk while the purge replicates
// (status PENDING/FAILED); its data dir must stay with
// it. Returning the dir here made the disk layer delete
// it, which turned every non-inline retained version
// into an unreadable zombie: the purge state could never
// be applied and the bucket could never be deleted.
self.set_idx(i, v)?;
return Ok(old_dir);
return Ok(None);
}
found_index = Some(i);
}
@@ -2702,6 +2707,58 @@ mod test {
);
}
/// Regression for rustfs/backlog#2340: a version purge that still awaits
/// the replication target keeps the object version on disk with a pending
/// purge status. Its data dir must be retained with it; handing the dir
/// back here made the disk layer delete it, leaving every non-inline
/// retained version unreadable. The dir is released only once the purge
/// completes and the version itself goes away.
#[test]
fn delete_version_pending_version_purge_retains_object_data_dir() {
let version_id = Uuid::new_v4();
let data_dir = Uuid::new_v4();
let mut fm = FileMeta::new();
let mut fi = FileInfo::new("object", 2, 2);
fi.version_id = Some(version_id);
fi.data_dir = Some(data_dir);
fi.mod_time = Some(OffsetDateTime::now_utc());
fm.add_version(fi).unwrap();
let pending_purge = FileInfo {
name: "object".to_string(),
version_id: Some(version_id),
mark_deleted: true,
replication_state_internal: Some(ReplicationState {
version_purge_status_internal: Some("target=PENDING;".to_string()),
purge_targets: version_purge_statuses_map("target=PENDING;"),
..Default::default()
}),
..Default::default()
};
let freed = fm.delete_version(&pending_purge).unwrap();
assert_eq!(freed, None, "a pending purge must not release the retained version's data dir");
assert_eq!(fm.versions.len(), 1, "the version must stay until the purge replicates");
let retained = fm
.into_fileinfo("vol", "object", &version_id.to_string(), false, false, true)
.unwrap();
assert_eq!(retained.data_dir, Some(data_dir));
assert_eq!(retained.version_purge_status(), VersionPurgeStatusType::Pending);
let completed_purge = FileInfo {
name: "object".to_string(),
version_id: Some(version_id),
replication_state_internal: Some(ReplicationState {
version_purge_status_internal: Some("target=COMPLETE;".to_string()),
purge_targets: version_purge_statuses_map("target=COMPLETE;"),
..Default::default()
}),
..Default::default()
};
let freed = fm.delete_version(&completed_purge).unwrap();
assert_eq!(freed, Some(data_dir), "a completed purge removes the version and releases its data dir");
assert!(fm.versions.is_empty());
}
#[test]
fn delete_version_accepts_delete_only_marker_and_free_version_paths() {
let marker_version_id = Uuid::new_v4();
@@ -6,7 +6,7 @@
## What a replication PUT carries by default
- A plain signed body with an exact `Content-Length`. The SDK does not add a streaming trailer checksum, so the body is never wrapped in `aws-chunked` framing (rustfs#6853: a target that does not decode that framing stored the frames verbatim while RustFS recorded COMPLETED).
- For a single-part object, the checksum the source object was uploaded with, forwarded as its `x-amz-checksum-<algorithm>` header (the value the source verified on upload). A multipart replica is rebuilt through CreateMultipartUpload/UploadPart and carries no object-level checksum header. Managed-SSE objects forward none.
- Any object-level checksum the source object was uploaded with, forwarded as its `x-amz-checksum-*` header.
- On a PUT that carries Object Lock parameters and no forwarded checksum: `Content-MD5` derived from the source ETag, or an SDK CRC32 checksum when the ETag is not the MD5 of the wire bytes (rustfs#7082).
- The source ETag, mtime and version id on `x-rustfs-source-*` headers (with `x-minio-source-*` twins), and the Object Lock mode, retain-until date and legal hold of the source version when present.
- After the PUT, the target's ETag is compared with the source ETag when both are plain single-part MD5s; a mismatch fails the replication instead of reporting a corrupted replica as COMPLETED.
@@ -17,7 +17,6 @@
| --- | --- | --- |
| Rejects or mis-stores `aws-chunked` bodies (SeaweedFS 3.97) | Handled by the plain-payload default above. | Outbound target matrix, `RejectAwsChunked` mode |
| Requires `Content-MD5` or `x-amz-checksum-*` on a PutObject with Object Lock parameters (AWS S3, MinIO, Impossible Cloud, most compatible stores) | Satisfied: a locked single PUT carries `Content-MD5` derived from the source ETag (plaintext objects whose ETag is the MD5 of the wire bytes) or an SDK CRC32 checksum (multipart-layout ETags, managed SSE, SSE-C passthrough — this one is an `aws-chunked` trailer, so a target that also rejects that framing cannot take such objects). Releases before this fix (`1.0.0-rc.5`) need `RUSTFS_REPLICATION_STREAMING_CHECKSUMS=true` as a workaround. | Outbound target matrix, `RequireChecksumWithObjectLock` mode |
| Stores `x-amz-checksum-*` from a PutObject and returns it on `HEAD ?ChecksumMode=ENABLED` (AWS S3, Wasabi, RustFS) | Satisfied for single-part objects: the replica answers with the source's checksum. Before this fix (`1.0.0-rc.5`) the checksum left the source as `x-amz-meta-<algorithm>` user metadata and no replica carried it (rustfs/backlog#2340). | Outbound target matrix, `Checksummed` shape |
| Mints its own version ids (AWS S3, Wasabi, Impossible Cloud) | Data lands; version-addressed convergence does not. See rustfs/backlog#2085 and `docs/operations/replication-check.md` (VersionFidelity). | `replication-check`, outbound target matrix, `MintOwnVersionIds` mode |
| Returns an ETag that is not the content MD5 without announcing SSE | Every single-part object fails ETag verification. Set `RUSTFS_REPLICATION_REPLICA_ETAG_VERIFY=false`. | Replication status FAILED with `replica etag mismatch` |
@@ -25,7 +24,7 @@
| Variable | Default | Meaning |
| --- | --- | --- |
| `RUSTFS_REPLICATION_STREAMING_CHECKSUMS` | unset (plain payloads) | `true` or `1` restores SDK trailer checksums (`RequestChecksumCalculation::WhenSupported`). Every streaming upload is then `aws-chunked` with an `x-amz-trailer`, except a single-part PUT that forwards the source's `x-amz-checksum-*` header, which is sent plain so the target does not receive a second algorithm; use only when every target decodes that framing. |
| `RUSTFS_REPLICATION_STREAMING_CHECKSUMS` | unset (plain payloads) | `true` or `1` restores SDK trailer checksums (`RequestChecksumCalculation::WhenSupported`). Every streaming upload is then `aws-chunked` with an `x-amz-trailer`; use only when every target decodes that framing. |
| `RUSTFS_REPLICATION_REPLICA_ETAG_VERIFY` | enabled | `false` or `0` disables the post-PUT ETag comparison for targets whose 32-hex ETags are legitimately not the content MD5. |
Both knobs are read by the RustFS process that owns the replication target, at client build time; restart the server after changing them.