fix(storage): restore multipart disk compression and make the legacy decompressor resumable

Multipart uploads have bypassed disk compression since #5169 removed the session marker as a stopgap for mid-stream GET failures. The actual root cause was never the multipart layout: the legacy DecompressReader reset its payload consumption state on every poll re-entry, so a Poll::Pending in the middle of a block payload (routine under the erasure duplex) desynchronized the block framing and surfaced as LZ4 frameType errors. This rewrites the decoder as a resumable state machine, restores the multipart session compression marker, reports logical part sizes in ListParts, and makes the rebalance migration read raw stored bytes so compressed and encrypted objects survive migration verbatim.

Fixes #5957. Internal tracking: backlog#1848, backlog#1850.
This commit is contained in:
overtrue
2026-08-13 06:25:19 +08:00
parent e2be34cade
commit 7c873bfe8b
9 changed files with 1579 additions and 69 deletions
+417
View File
@@ -1727,6 +1727,423 @@ mod tests {
assert_eq!(actual, b"fghijkl");
}
/// Compresses one multipart part exactly like the write path does
/// (`WritePlan::with_compression` wraps each part in its own
/// `compression_reader`), returning the on-disk bytes and the storage-format
/// compression index.
async fn compressed_part_fixture(data: &[u8]) -> (Vec<u8>, Option<Bytes>) {
use crate::io_support::rio::TryGetIndex as _;
let mut compressor =
crate::io_support::rio::compression_reader(Cursor::new(data.to_vec()), CompressionAlgorithm::default(), false);
let mut compressed = Vec::new();
compressor.read_to_end(&mut compressed).await.expect("compress part stream");
let index = compressor
.try_get_index()
.map(crate::io_support::rio::compression_index_storage_bytes);
(compressed, index)
}
struct CompressedMultipartFixture {
object_info: ObjectInfo,
stored: Vec<u8>,
plaintext: Vec<u8>,
}
/// Builds the on-disk representation of a compressed multipart object: each
/// part is an independent compressed stream and the storage layer serves
/// their concatenation.
async fn compressed_multipart_fixture(part_sizes: &[usize]) -> CompressedMultipartFixture {
let pattern = b"compressed multipart read path fixture data ";
let mut plaintext = Vec::new();
let mut stored = Vec::new();
let mut parts = Vec::with_capacity(part_sizes.len());
for (i, part_size) in part_sizes.iter().enumerate() {
let mut part_plaintext = Vec::with_capacity(*part_size);
while part_plaintext.len() < *part_size {
part_plaintext.extend_from_slice(pattern);
part_plaintext.push(i as u8);
}
part_plaintext.truncate(*part_size);
let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
parts.push(ObjectPartInfo {
number: i + 1,
size: compressed.len(),
actual_size: *part_size as i64,
index,
..Default::default()
});
stored.extend_from_slice(&compressed);
plaintext.extend_from_slice(&part_plaintext);
}
let mut user_defined = HashMap::new();
rustfs_utils::http::insert_str(
&mut user_defined,
rustfs_utils::http::SUFFIX_COMPRESSION,
crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
);
rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
let object_info = ObjectInfo {
bucket: "test-bucket".to_string(),
name: "compressed-multipart".to_string(),
size: stored.len() as i64,
etag: Some(format!("6bcf86bed8807b8e78f0fc6e0a53079d-{}", part_sizes.len())),
parts: Arc::new(parts),
user_defined: Arc::new(user_defined),
..Default::default()
};
CompressedMultipartFixture {
object_info,
stored,
plaintext,
}
}
/// Plans the read once to learn the storage window, then serves exactly that
/// window — mirroring how `set_disk` feeds the erasure read into the
/// returned reader.
async fn read_compressed_multipart(
fixture: &CompressedMultipartFixture,
rs: Option<HTTPRangeSpec>,
opts: &ObjectOptions,
) -> Vec<u8> {
let headers = HeaderMap::new();
let (_, offset, length) =
GetObjectReader::new(Box::new(Cursor::new(Vec::new())), rs.clone(), &fixture.object_info, opts, &headers)
.await
.expect("plan compressed multipart read");
let end = offset + usize::try_from(length).expect("storage window length must be non-negative");
assert!(
end <= fixture.stored.len(),
"planned storage window {offset}..{end} exceeds stored stream of {} bytes",
fixture.stored.len()
);
let window = fixture.stored[offset..end].to_vec();
let (mut reader, replay_offset, replay_length) =
GetObjectReader::new(Box::new(Cursor::new(window)), rs, &fixture.object_info, opts, &headers)
.await
.expect("build compressed multipart reader");
assert_eq!((replay_offset, replay_length), (offset, length), "read plan must be deterministic");
reader.read_all().await.expect("read compressed multipart stream")
}
/// Byte pattern with a 2 KiB period: it compresses extremely well while
/// looking nothing like ASCII fixtures. Mirrors the e2e generator that
/// exposed a truncated full GET on high-ratio multipart payloads.
fn high_ratio_binary_payload(size: usize, seed: u8) -> Vec<u8> {
(0..size)
.map(|i| ((i as u64).wrapping_mul(2_654_435_761).wrapping_add(seed as u64) >> 3) as u8)
.collect()
}
#[tokio::test]
async fn compressed_multipart_full_get_handles_high_ratio_binary_payload() {
let part_sizes = [5 * 1024 * 1024_usize, 1024 * 1024];
let mut plaintext = Vec::new();
let mut stored = Vec::new();
let mut parts = Vec::with_capacity(part_sizes.len());
for (i, part_size) in part_sizes.iter().enumerate() {
let part_plaintext = high_ratio_binary_payload(*part_size, if i == 0 { 7 } else { 61 });
let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
parts.push(ObjectPartInfo {
number: i + 1,
size: compressed.len(),
actual_size: *part_size as i64,
index,
..Default::default()
});
stored.extend_from_slice(&compressed);
plaintext.extend_from_slice(&part_plaintext);
}
let mut user_defined = HashMap::new();
rustfs_utils::http::insert_str(
&mut user_defined,
rustfs_utils::http::SUFFIX_COMPRESSION,
crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
);
rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
let fixture = CompressedMultipartFixture {
object_info: ObjectInfo {
bucket: "test-bucket".to_string(),
name: "high-ratio-multipart".to_string(),
size: stored.len() as i64,
etag: Some("6bcf86bed8807b8e78f0fc6e0a53079d-2".to_string()),
parts: Arc::new(parts),
user_defined: Arc::new(user_defined),
..Default::default()
},
stored,
plaintext,
};
let read = read_compressed_multipart(&fixture, None, &ObjectOptions::default()).await;
assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
assert_eq!(read, fixture.plaintext, "high-ratio multipart payload must survive the roundtrip");
}
/// Full GET over a compressed multipart object must decode across part
/// boundaries: every part is an independent compressed stream (this is also
/// the on-disk shape written by builds before rustfs/rustfs#5169 disabled
/// multipart compression, so this pins legacy-object readability).
#[tokio::test]
async fn compressed_multipart_full_get_decodes_across_part_boundaries() {
let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 2 * 1024 * 1024, 512 * 1024]).await;
let read = read_compressed_multipart(&fixture, None, &ObjectOptions::default()).await;
assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
assert_eq!(read, fixture.plaintext, "full GET must reassemble all parts");
}
#[tokio::test]
async fn compressed_multipart_range_get_crosses_part_boundary() {
let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 2 * 1024 * 1024]).await;
let boundary = 3 * 1024 * 1024_i64;
let rs = HTTPRangeSpec {
is_suffix_length: false,
start: boundary - 100_000,
end: boundary + 100_000 - 1,
};
let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
let expected = &fixture.plaintext[(boundary - 100_000) as usize..(boundary + 100_000) as usize];
assert_eq!(read, expected, "boundary-crossing range must splice both parts");
}
#[tokio::test]
async fn compressed_multipart_range_get_seeks_into_later_part() {
let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 4 * 1024 * 1024]).await;
// Deep inside part 2 so the plan skips part 1 entirely and (when the
// part carries an index) seeks within part 2.
let start = 3 * 1024 * 1024_i64 + 2 * 1024 * 1024_i64 + 137;
let rs = HTTPRangeSpec {
is_suffix_length: false,
start,
end: start + 64 * 1024 - 1,
};
let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
let expected = &fixture.plaintext[start as usize..(start + 64 * 1024) as usize];
assert_eq!(read, expected, "range inside a later part must decode from that part");
}
/// Parts written without a compression index (small parts skip the index in
/// the rio-v2 backend) must still be rangeable: the plan starts at the part
/// boundary and skips decompressed bytes.
#[tokio::test]
async fn compressed_multipart_range_get_works_without_part_indexes() {
let mut fixture = compressed_multipart_fixture(&[1024 * 1024, 1024 * 1024]).await;
let parts = fixture
.object_info
.parts
.iter()
.map(|part| ObjectPartInfo {
index: None,
..part.clone()
})
.collect::<Vec<_>>();
fixture.object_info.parts = Arc::new(parts);
let start = 1024 * 1024_i64 + 4096;
let rs = HTTPRangeSpec {
is_suffix_length: false,
start,
end: start + 32 * 1024 - 1,
};
let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
let expected = &fixture.plaintext[start as usize..(start + 32 * 1024) as usize];
assert_eq!(read, expected, "index-less parts must fall back to part-boundary skip");
}
#[tokio::test]
async fn compressed_multipart_part_number_get_returns_single_part() {
let part_sizes = [3 * 1024 * 1024, 2 * 1024 * 1024, 512 * 1024];
let fixture = compressed_multipart_fixture(&part_sizes).await;
let mut logical_offset = 0_usize;
for (i, part_size) in part_sizes.iter().enumerate() {
let opts = ObjectOptions {
part_number: Some(i + 1),
..Default::default()
};
let read = read_compressed_multipart(&fixture, None, &opts).await;
let expected = &fixture.plaintext[logical_offset..logical_offset + part_size];
assert_eq!(read.len(), *part_size, "partNumber={} GET must return the part's logical size", i + 1);
assert_eq!(read, expected, "partNumber={} GET must return the original part bytes", i + 1);
logical_offset += part_size;
}
}
#[tokio::test]
async fn compressed_multipart_suffix_range_reads_tail() {
let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 1024 * 1024]).await;
let suffix_len = 128 * 1024_i64;
let rs = HTTPRangeSpec {
is_suffix_length: true,
start: suffix_len,
end: -1,
};
let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
let expected = &fixture.plaintext[fixture.plaintext.len() - suffix_len as usize..];
assert_eq!(read, expected, "suffix range must return the tail of the last part");
}
/// Builds an SSE-C + disk-compression multipart object exactly like the
/// write path: each part is compressed into its own stream and then
/// encrypted with the per-part key schedule. The fixture is
/// legacy-encryption-specific (`rustfs_rio::EncryptReader`), matching the
/// pre-existing `build_legacy_ssec_multipart_fixture` shape, while the
/// compression layer follows the active backend feature.
async fn compressed_encrypted_multipart_fixture(key_bytes: [u8; 32], part_sizes: &[usize]) -> CompressedMultipartFixture {
let pattern = b"compressed encrypted multipart fixture data ";
let mut plaintext = Vec::new();
let mut stored = Vec::new();
let mut parts = Vec::with_capacity(part_sizes.len());
for (i, part_size) in part_sizes.iter().enumerate() {
let part_number = i + 1;
let mut part_plaintext = Vec::with_capacity(*part_size);
while part_plaintext.len() < *part_size {
part_plaintext.extend_from_slice(pattern);
part_plaintext.push(part_number as u8);
}
part_plaintext.truncate(*part_size);
let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
let mut part_cipher = Vec::new();
rustfs_rio::EncryptReader::new_multipart(Cursor::new(compressed), key_bytes, LEGACY_FIXTURE_BASE_NONCE, part_number)
.read_to_end(&mut part_cipher)
.await
.expect("encrypt compressed fixture part");
parts.push(ObjectPartInfo {
number: part_number,
size: part_cipher.len(),
actual_size: *part_size as i64,
index,
..Default::default()
});
stored.extend_from_slice(&part_cipher);
plaintext.extend_from_slice(&part_plaintext);
}
let mut user_defined = legacy_ssec_multipart_metadata(key_bytes, plaintext.len());
rustfs_utils::http::insert_str(
&mut user_defined,
rustfs_utils::http::SUFFIX_COMPRESSION,
crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
);
rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
let object_info = ObjectInfo {
bucket: "test-bucket".to_string(),
name: "compressed-encrypted-multipart".to_string(),
size: stored.len() as i64,
etag: Some(format!("6bcf86bed8807b8e78f0fc6e0a53079d-{}", part_sizes.len())),
parts: Arc::new(parts),
user_defined: Arc::new(user_defined),
..Default::default()
};
CompressedMultipartFixture {
object_info,
stored,
plaintext,
}
}
async fn read_compressed_encrypted_multipart(
fixture: &CompressedMultipartFixture,
key_bytes: [u8; 32],
rs: Option<HTTPRangeSpec>,
opts: &ObjectOptions,
) -> Vec<u8> {
let headers = ssec_headers_from_key(key_bytes);
let (_, offset, length) =
GetObjectReader::new(Box::new(Cursor::new(Vec::new())), rs.clone(), &fixture.object_info, opts, &headers)
.await
.expect("plan compressed encrypted multipart read");
let end = offset + usize::try_from(length).expect("storage window length must be non-negative");
assert!(
end <= fixture.stored.len(),
"planned storage window {offset}..{end} exceeds stored stream of {} bytes",
fixture.stored.len()
);
let window = fixture.stored[offset..end].to_vec();
let (mut reader, replay_offset, replay_length) =
GetObjectReader::new(Box::new(Cursor::new(window)), rs, &fixture.object_info, opts, &headers)
.await
.expect("build compressed encrypted multipart reader");
assert_eq!((replay_offset, replay_length), (offset, length), "read plan must be deterministic");
reader.read_all().await.expect("read compressed encrypted multipart stream")
}
#[tokio::test]
async fn compressed_encrypted_multipart_full_get_roundtrip() {
let key_bytes = [0x6Eu8; 32];
let fixture = compressed_encrypted_multipart_fixture(key_bytes, &[3 * 1024 * 1024, 1024 * 1024]).await;
let read = read_compressed_encrypted_multipart(&fixture, key_bytes, None, &ObjectOptions::default()).await;
assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
assert_eq!(read, fixture.plaintext, "SSE-C + compression full GET must reassemble all parts");
}
#[tokio::test]
async fn compressed_encrypted_multipart_range_crosses_part_boundary() {
let key_bytes = [0x6Eu8; 32];
let fixture = compressed_encrypted_multipart_fixture(key_bytes, &[3 * 1024 * 1024, 1024 * 1024]).await;
let boundary = 3 * 1024 * 1024_i64;
let rs = HTTPRangeSpec {
is_suffix_length: false,
start: boundary - 65_536,
end: boundary + 65_536 - 1,
};
let read = read_compressed_encrypted_multipart(&fixture, key_bytes, Some(rs), &ObjectOptions::default()).await;
let expected = &fixture.plaintext[(boundary - 65_536) as usize..(boundary + 65_536) as usize];
assert_eq!(read, expected, "SSE-C + compression boundary-crossing range must splice both parts");
}
#[tokio::test]
async fn compressed_encrypted_multipart_part_number_get_returns_single_part() {
let key_bytes = [0x6Eu8; 32];
let part_sizes = [3 * 1024 * 1024, 1024 * 1024];
let fixture = compressed_encrypted_multipart_fixture(key_bytes, &part_sizes).await;
let opts = ObjectOptions {
part_number: Some(2),
..Default::default()
};
let read = read_compressed_encrypted_multipart(&fixture, key_bytes, None, &opts).await;
let expected = &fixture.plaintext[part_sizes[0]..];
assert_eq!(read.len(), part_sizes[1], "partNumber=2 GET must return the part's logical size");
assert_eq!(read, expected, "partNumber=2 GET must return the original part bytes");
}
#[tokio::test]
async fn test_get_object_reader_rejects_ssec_read_without_headers() {
let object_info = ObjectInfo {