Compare commits

..

1 Commits

Author SHA1 Message Date
overtrue 5983594da9 feat(capture): add explicit schema-bound checkpoint capabilities 2026-10-11 21:28:05 +08:00
7 changed files with 591 additions and 403 deletions
@@ -103,6 +103,10 @@ fn decode_provenance(b: &[u8]) -> Result<CreatedOperationProvenance, StoreError>
})
}
pub(crate) fn validate_provenance_encoding(bytes: &[u8]) -> Result<(), StoreError> {
decode_provenance(bytes).map(|_| ())
}
fn decode_stored_operation(b: &[u8], schema: u64) -> Result<CreatedOperation, StoreError> {
if matches!(schema, 5 | 6) {
Ok(decode_provenance(b)?.operation)
+402
View File
@@ -22,6 +22,7 @@ pub use node::{CaptureNode, CaptureNodeError};
pub use raft_types::{
CaptureCommand, CaptureMembership, CaptureRaftEntry, CaptureResponse, CaptureResponseKind, CaptureTypeConfig, MembershipError,
};
pub use snapshot::{CapabilityArtifact, CapabilityManifest};
#[cfg(test)]
mod application_tests;
mod truncate;
@@ -2231,6 +2232,407 @@ mod tests {
membership,
)
}
#[derive(Default)]
struct U27aSink(Vec<(snapshot::Section, Vec<u8>, Vec<u8>)>);
impl snapshot::StagingSink for U27aSink {
fn row(&mut self, section: snapshot::Section, key: &[u8], value: &[u8]) -> io::Result<()> {
self.0.push((section, key.to_vec(), value.to_vec()));
Ok(())
}
}
fn u27a_manifest(payload: &[u8]) -> snapshot::CapabilityManifest {
use sha2::{Digest, Sha256};
let mut bytes = [0; 56];
bytes[..4].copy_from_slice(b"CCAP");
bytes[4] = 1;
bytes[5] = 1;
bytes[6..14].copy_from_slice(&6u64.to_be_bytes());
bytes[14] = 1;
bytes[15] = 1;
bytes[16..24].copy_from_slice(&u64::try_from(payload.len()).expect("small test payload length").to_be_bytes());
bytes[24..].copy_from_slice(&Sha256::digest(payload));
snapshot::CapabilityManifest { bytes }
}
fn u27a_fixture() -> (tempfile::TempDir, CrashBackend, Store, Vec<DecideCreated>) {
let (dir, backend, mut store) = u17a_store();
provenance_promote(&mut store);
let commands = provenance_commands();
let mut entries = vec![u17a_blank(0), u17a_membership(1).0];
for (index, command) in commands.iter().enumerate() {
entries.push(u17a_created(u64::try_from(index).expect("fixture index") + 2, command.clone()));
}
u24_apply(&mut store, &entries);
assert_eq!(committed_schema(&store), 6, "schema6 fixture setup");
(dir, backend, store, commands)
}
fn u27a_rows(store: &Store) -> Vec<(Vec<u8>, Vec<u8>)> {
let tx = store.database.begin_read().expect("operation fixture transaction");
let table = tx.open_table(application::OPERATIONS).expect("operation fixture table");
let rows = table
.iter()
.expect("operation fixture cursor")
.map(|row| {
let (key, value) = row.expect("operation fixture row");
(key.value().to_vec(), value.value().to_vec())
})
.collect();
rows
}
fn u27a_artifact(writer: &NativeLogWriter) -> snapshot::CapabilityArtifact<Vec<u8>> {
let view = u19_runtime()
.block_on(writer.snapshot_view())
.expect("capability fixture view");
let result = view.stream_capability_artifact(Vec::new(), u24_limits());
assert_eq!(result.is_ok(), true, "U27A_CAPABILITY_ENABLED");
result.expect("asserted capability production")
}
fn u27a_decode(
artifact: snapshot::CapabilityArtifact<Vec<u8>>,
) -> Result<(snapshot::Header, U27aSink), snapshot::SnapshotError> {
let mut sink = U27aSink::default();
let header = snapshot::decode_capability_artifact(
snapshot::CapabilityArtifact {
manifest: artifact.manifest,
payload: io::Cursor::new(artifact.payload),
},
&mut sink,
u24_limits(),
)?;
Ok((header, sink))
}
#[test]
fn u27a_checkpoint_schema6_copr_stream() {
let (_dir, _backend, store, commands) = u27a_fixture();
let expected = u27a_rows(&store);
let writer = NativeLogWriter::new(store);
let artifact = u27a_artifact(&writer);
assert_eq!(
artifact.manifest.encode(),
u27a_manifest(&artifact.payload).encode(),
"U27A_PAYLOAD_BINDING"
);
let decoded = u27a_decode(artifact);
assert_eq!(decoded.is_ok(), true, "U27A_COPR_BYTES");
let (header, sink) = decoded.expect("asserted capability consumer");
assert_eq!(header.applied, Some(u17a_id(10)), "U27A_COPR_BYTES");
assert_eq!(header.membership.log_id(), &Some(u17a_id(1)), "U27A_COPR_BYTES");
let operations: Vec<_> = sink
.0
.into_iter()
.filter(|row| row.0 == snapshot::Section::Operations)
.map(|(_, key, value)| (key, value))
.collect();
assert_eq!(operations, expected, "U27A_COPR_BYTES");
assert_eq!(operations.len(), 4, "U27A_COPR_BYTES");
for (command_index, first_index) in [(0, 2), (3, 5), (4, 6), (5, 7)] {
let command = &commands[command_index];
let value = &operations
.iter()
.find(|(key, _)| key.as_slice() == command.operation_id.as_slice())
.expect("canonical operation")
.1;
assert_eq!(&value[..5], b"COPR\x01", "U27A_COPR_BYTES");
assert_eq!(&value[5..29], encode_id(u17a_id(first_index)), "U27A_FIRST_ID");
let stored = writer
.store
.lock()
.expect("fixture store")
.created_operation_provenance(command.operation_id)
.expect("provenance read")
.expect("operation");
assert_eq!(stored.first_decision_log_id, u17a_id(first_index), "U27A_FIRST_ID");
assert_eq!(
stored.operation.canonical_command,
command.encode().expect("original command"),
"U27A_COPR_BYTES"
);
}
let store = writer.store.lock().expect("conflict receipt store");
assert_eq!(
store
.native_apply_response(4)
.expect("receipt")
.expect("response")
.created_result(),
Some(&CreatedResult::OperationConflict),
"U27A_FIRST_ID"
);
}
#[test]
fn u27a_checkpoint_schema4_legacy_unchanged() {
let (_dir, _backend, mut store) = u17a_store();
u24_apply(&mut store, &[u17a_blank(0), u17a_membership(1).0, u17a_created(2, u17a_command())]);
let before = u17b_snapshot(&store);
// Fixed bytes exported by the accepted pre-capability binary for this schema4 fixture.
const LEGACY_PAYLOAD_HEX: &str = concat!(
"43534e5001010707070707070707070707070707070701000000000000000b00000000000000070000000000000002000000000000010e434d454d01",
"01000000000000000b000000000000000700000000000000010000000200000002000000000000000700000000000000080000000200000000000000",
"08000000000000000900000004000000000000000700000024434e4f44010000001b68747470733a2f2f706565722d372e6578616d706c653a373030",
"37000000000000000800000023434e4f44010000001a687474703a2f2f706565722d382e6578616d706c653a37303038000000000000000900000024",
"434e4f44010000001b68747470733a2f2f706565722d392e6578616d706c653a37303039000000000000000a00000025434e4f44010000001c687474",
"70733a2f2f6c6561726e65722e6578616d706c653a373031300000000000000009060606060606060606060606060606060707070707070707070707",
"070707070701000000000000000100000000000000210000000000000010020202020202020202020202020202020000000d6e61746976652d6f626a",
"656374030303030303030303030303030303030200000000000000010000000000000031000000000000006802020202020202020202020202020202",
"0000000d6e61746976652d6f626a65637403030303030303030303030303030303030303030303030303030303030303030404040404040404040404",
"040404040405050505050505050505050505050505050505050505050505050505050505050000000000001001000000000000000b00000000000000",
"0700000000000000020000000000000009030000000000000001000000000000001000000000000000a9010101010101010101010101010101010202",
"02020202020202020202020202020000000d6e61746976652d6f626a6563740303030303030303030303030303030304040404040404040404040404",
"04040405050505050505050505050505050505050505050505050505050505050505050000000000001001000000000000000b000000000000000700",
"000000000000020000000000000009060606060606060606060606060606060707070707070707070707070707070704000000000000000100000000",
"0000001000000000000000c8010101010101010101010101010101010000000000000087435254440101010101010101010101010101010101020202",
"020202020202020202020202020000000d6e61746976652d6f626a656374000303030303030303030303030303030304040404040404040404040404",
"040404050505050505050505050505050505050505050505050505050505050505050500000000000010010000000000000009000303030303030303",
"030303030303030301010101010101010101010101010101000000000000000b000000000000000700000000000000024f43d9e18a25d702afdacde7",
"e9b00f87177c40b11232952e61ddb2aa18411bce",
);
let original: Vec<u8> = LEGACY_PAYLOAD_HEX
.as_bytes()
.chunks_exact(2)
.map(|pair| {
let digits = std::str::from_utf8(pair).expect("fixed legacy hex");
u8::from_str_radix(digits, 16).expect("fixed legacy byte")
})
.collect();
let writer = NativeLogWriter::new(store);
let view = u19_runtime().block_on(writer.snapshot_view()).expect("legacy view");
let payload = view.stream(Vec::new(), u24_limits()).expect("original stream");
assert_eq!(payload, original, "U27A_V1_OPAQUE");
assert_eq!(&payload[..6], b"CSNP\x01\x01", "U27A_V1_OPAQUE");
let view = u19_runtime()
.block_on(writer.snapshot_view())
.expect("legacy capability view");
let result = view.stream_capability_artifact(Vec::new(), u24_limits());
assert_eq!(
matches!(result, Err(NativeSnapshotError::Store(StoreError::UnsupportedSchema))),
true,
"U27A_SCHEMA_GATE"
);
let store = writer.store.lock().expect("legacy unchanged store");
assert_eq!(committed_schema(&store), 4, "U27A_SCHEMA_GATE");
assert_eq!(u17b_snapshot(&store), before, "U27A_SCHEMA_GATE");
}
#[test]
fn u27a_checkpoint_codec_mismatch_rejected() {
let (_dir, _backend, mut store) = u17a_store();
u24_apply(&mut store, &[u17a_created(0, u17a_command())]);
let writer = NativeLogWriter::new(store);
let old = u19_runtime()
.block_on(writer.snapshot_view())
.expect("legacy codec view")
.stream(Vec::new(), u24_limits())
.expect("legacy codec payload");
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest: u27a_manifest(&old),
payload: old
})
.is_err(),
true,
"U27A_CODEC_DISPATCH"
);
let (_dir6, _backend6, store6, _) = u27a_fixture();
let writer6 = NativeLogWriter::new(store6);
let artifact = u27a_artifact(&writer6);
for (offset, wrong) in [(6, 5), (14, 0), (14, 2), (15, 0), (15, 2)] {
let mut manifest = artifact.manifest.clone();
if offset == 6 {
manifest.bytes[6..14].copy_from_slice(&u64::from(wrong).to_be_bytes());
} else {
manifest.bytes[offset] = wrong;
}
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest,
payload: artifact.payload.clone()
})
.is_err(),
true,
"U27A_CODEC_DISPATCH"
);
}
}
#[test]
fn u27a_checkpoint_manifest_binding_rejected() {
struct ShortWriter {
bytes: Vec<u8>,
interrupted: bool,
}
impl Write for ShortWriter {
fn write(&mut self, bytes: &[u8]) -> io::Result<usize> {
if !self.interrupted {
self.interrupted = true;
return Err(io::ErrorKind::Interrupted.into());
}
let length = bytes.len().min(3);
self.bytes.extend_from_slice(&bytes[..length]);
Ok(length)
}
fn flush(&mut self) -> io::Result<()> {
Ok(())
}
}
let (_dir, _backend, store, _) = u27a_fixture();
let writer = NativeLogWriter::new(store);
let view = u19_runtime().block_on(writer.snapshot_view()).expect("short output view");
let result = view.stream_capability_artifact(
ShortWriter {
bytes: Vec::new(),
interrupted: false,
},
u24_limits(),
);
assert_eq!(result.is_ok(), true, "U27A_PAYLOAD_BINDING");
let artifact = result.expect("asserted short write producer");
assert_eq!(
artifact.manifest.encode(),
u27a_manifest(&artifact.payload.bytes).encode(),
"U27A_PAYLOAD_BINDING"
);
let encoded = artifact.manifest.encode();
assert_eq!(snapshot::CapabilityManifest::decode(&encoded).is_ok(), true, "U27A_PAYLOAD_BINDING");
for size in [0, 55, 57] {
let mut raw = encoded.to_vec();
raw.resize(size, 0);
assert_eq!(snapshot::CapabilityManifest::decode(&raw).is_err(), true, "U27A_PAYLOAD_BINDING");
}
for offset in [0, 4, 5, 14, 15] {
let mut raw = encoded;
raw[offset] = 255;
assert_eq!(snapshot::CapabilityManifest::decode(&raw).is_err(), true, "U27A_PAYLOAD_BINDING");
}
for offset in [16, 23, 24, 55] {
let mut manifest = artifact.manifest.clone();
manifest.bytes[offset] ^= 1;
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest,
payload: artifact.payload.bytes.clone()
})
.is_err(),
true,
"U27A_PAYLOAD_BINDING"
);
}
let mut corrupt = artifact.payload.bytes.clone();
*corrupt.last_mut().expect("footer byte") ^= 1;
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest: artifact.manifest.clone(),
payload: corrupt
})
.is_err(),
true,
"U27A_PAYLOAD_BINDING"
);
let mut extra = artifact.payload.bytes.clone();
extra.push(0);
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest: artifact.manifest,
payload: extra
})
.is_err(),
true,
"U27A_PAYLOAD_BINDING"
);
}
#[test]
fn u27a_checkpoint_negative_bytes_first_id() {
let (_dir, _backend, mut store, commands) = u27a_fixture();
let before = u27a_rows(&store);
let mut later = commands[0].clone();
later.operation_id = [20; 16];
later.prepared.version_id = [21; 16];
later.expected_head = Some(commands[0].prepared.version_id);
u24_apply(
&mut store,
&[
u17a_created(11, later),
u17a_created(12, commands[4].clone()),
u17a_created(13, commands[2].clone()),
],
);
assert_eq!(
store.created_head(&commands[0].object).expect("changed head"),
Some([21; 16]),
"U27A_FIRST_ID"
);
let after = u27a_rows(&store);
for command_index in [3, 4, 5] {
let key = commands[command_index].operation_id.to_vec();
let old = before.iter().find(|row| row.0 == key).expect("first negative bytes");
assert_eq!(after.iter().find(|row| row.0 == key), Some(old), "U27A_FIRST_ID");
}
assert_eq!(
store
.native_apply_response(13)
.expect("conflict response")
.expect("receipt")
.created_result(),
Some(&CreatedResult::OperationConflict),
"U27A_FIRST_ID"
);
let writer = NativeLogWriter::new(store);
let (_, sink) = u27a_decode(u27a_artifact(&writer)).expect("capability preserves negatives");
let captured: Vec<_> = sink
.0
.into_iter()
.filter(|row| row.0 == snapshot::Section::Operations)
.map(|(_, key, value)| (key, value))
.collect();
assert_eq!(captured, after, "U27A_FIRST_ID");
}
#[test]
fn u27a_checkpoint_v1_framing_is_opaque() {
let (_dir, _backend, mut store) = u17a_store();
u24_apply(&mut store, &[u17a_blank(0)]);
let tx = store.database.begin_read().expect("opaque header transaction");
let (applied, membership) = application::native_application_state(&tx).expect("opaque application pair");
let header = snapshot::Header {
cluster_id: identity().cluster_id,
applied,
membership,
binding: u17a_binding(),
};
let mut payload = Vec::new();
snapshot::encode(
&mut payload,
&header,
[0, 0, 0, 1],
[snapshot::Row {
section: snapshot::Section::Operations,
key: b"o",
value: b"opaque-operation",
}],
u24_limits(),
)
.expect("original opaque encode");
let mut sink = U27aSink::default();
let original = snapshot::decode(payload.as_slice(), &mut sink, u24_limits());
assert_eq!(original.is_ok(), true, "U27A_V1_OPAQUE");
assert_eq!(sink.0[0].2, b"opaque-operation", "U27A_V1_OPAQUE");
assert_eq!(
u27a_decode(snapshot::CapabilityArtifact {
manifest: u27a_manifest(&payload),
payload
})
.is_err(),
true,
"U27A_V1_OPAQUE"
);
}
fn u24_limits() -> snapshot::Limits {
snapshot::Limits {
total_bytes: 1_000_000,
@@ -23,6 +23,7 @@ pub struct NativeSnapshotView {
transaction: ReadTransaction,
header: snapshot::Header,
counts: [u64; 4],
schema: u64,
_owner: Arc<Mutex<Store>>,
}
impl NativeSnapshotView {
@@ -35,18 +36,27 @@ impl NativeSnapshotView {
store.database.begin_read().map_err(StoreError::from)?
};
let (applied, membership) = application::native_application_state(&transaction)?;
let header = {
let (header, schema) = {
let meta = transaction.open_table(crate::META).map_err(StoreError::from)?;
let identity = meta
.get("identity")
.map_err(StoreError::from)?
.ok_or(StoreError::CorruptRecord)?;
snapshot::Header {
cluster_id: crate::decode_identity(identity.value())?.cluster_id,
applied,
membership,
binding: application::read_binding(&meta)?,
}
let schema = crate::number(
meta.get("schema")
.map_err(StoreError::from)?
.ok_or(StoreError::CorruptRecord)?
.value(),
)?;
(
snapshot::Header {
cluster_id: crate::decode_identity(identity.value())?.cluster_id,
applied,
membership,
binding: application::read_binding(&meta)?,
},
schema,
)
};
let mut counts = [0; 4];
for (index, definition) in [
@@ -68,10 +78,23 @@ impl NativeSnapshotView {
transaction,
header,
counts,
schema,
_owner: owner,
})
}
pub fn stream_capability_artifact<W: Write>(
self,
writer: W,
limits: snapshot::Limits,
) -> Result<snapshot::CapabilityArtifact<W>, NativeSnapshotError> {
if self.schema != 6 {
return Err(StoreError::UnsupportedSchema.into());
}
let output = snapshot::CapabilityOutput::new(writer, limits.total_bytes);
Ok(self.stream(output, limits)?.finish())
}
pub fn stream<W: Write>(self, writer: W, limits: snapshot::Limits) -> Result<W, NativeSnapshotError> {
let mut encoder = snapshot::Encoder::new(writer, &self.header, self.counts, limits)?;
for (definition, section) in [
+155
View File
@@ -38,6 +38,161 @@ pub enum SnapshotError {
#[error("snapshot IO failed")]
Io(#[from] std::io::Error),
}
/// Fixed capability declaration. Consumers validate these untrusted bytes before decoding.
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct CapabilityManifest {
pub bytes: [u8; 56],
}
impl CapabilityManifest {
pub fn encode(&self) -> [u8; 56] {
self.bytes
}
pub fn decode(bytes: &[u8]) -> Result<Self, SnapshotError> {
format(bytes.len() == 56, "capability length")?;
format(&bytes[..6] == b"CCAP\x01\x01", "capability version/framing")?;
format(bytes[6..14] == 6u64.to_be_bytes(), "capability schema")?;
format(bytes[14..16] == [1, 1], "operation codec")?;
let mut encoded = [0; 56];
encoded.copy_from_slice(bytes);
Ok(Self { bytes: encoded })
}
fn payload_length(&self) -> u64 {
let mut bytes = [0; 8];
bytes.copy_from_slice(&self.bytes[16..24]);
u64::from_be_bytes(bytes)
}
}
/// Manifest and payload form one explicit input. This does not grant installation authority.
pub struct CapabilityArtifact<T> {
pub manifest: CapabilityManifest,
pub payload: T,
}
pub(crate) struct CapabilityOutput<W> {
io: W,
hash: Sha256,
used: u64,
max: u64,
}
impl<W: Write> CapabilityOutput<W> {
pub(crate) fn new(io: W, max: u64) -> Self {
Self {
io,
hash: Sha256::new(),
used: 0,
max,
}
}
pub(crate) fn finish(self) -> CapabilityArtifact<W> {
let mut bytes = [0; 56];
bytes[..6].copy_from_slice(b"CCAP\x01\x01");
bytes[6..14].copy_from_slice(&6u64.to_be_bytes());
bytes[14..16].copy_from_slice(&[1, 1]);
bytes[16..24].copy_from_slice(&self.used.to_be_bytes());
bytes[24..].copy_from_slice(&self.hash.finalize());
CapabilityArtifact {
manifest: CapabilityManifest { bytes },
payload: self.io,
}
}
}
impl<W: Write> Write for CapabilityOutput<W> {
fn write(&mut self, bytes: &[u8]) -> std::io::Result<usize> {
let requested = u64::try_from(bytes.len()).map_err(std::io::Error::other)?;
if self.used.checked_add(requested).is_none_or(|n| n > self.max) {
return Err(std::io::Error::other("capability payload limit"));
}
let written = self.io.write(bytes)?;
let part = bytes
.get(..written)
.ok_or_else(|| std::io::Error::other("invalid writer count"))?;
let count = u64::try_from(written).map_err(std::io::Error::other)?;
self.used = self
.used
.checked_add(count)
.ok_or_else(|| std::io::Error::other("capability length overflow"))?;
self.hash.update(part);
Ok(written)
}
fn flush(&mut self) -> std::io::Result<()> {
self.io.flush()
}
}
struct CapabilityInput<R> {
io: R,
hash: Sha256,
used: u64,
max: u64,
}
impl<R: Read> Read for CapabilityInput<R> {
fn read(&mut self, bytes: &mut [u8]) -> std::io::Result<usize> {
// At the boundary, permit only the existing decoder's one-byte EOF probe.
let remaining = self
.max
.checked_sub(self.used)
.ok_or_else(|| std::io::Error::other("capability length overflow"))?;
let requested = u64::try_from(bytes.len()).map_err(std::io::Error::other)?;
let allowed = requested.min(if remaining == 0 { 1 } else { remaining });
let allowed = usize::try_from(allowed).map_err(std::io::Error::other)?;
let read = self.io.read(&mut bytes[..allowed])?;
let part = bytes
.get(..read)
.ok_or_else(|| std::io::Error::other("invalid reader count"))?;
let count = u64::try_from(read).map_err(std::io::Error::other)?;
let used = self
.used
.checked_add(count)
.ok_or_else(|| std::io::Error::other("capability length overflow"))?;
if used > self.max {
return Err(std::io::Error::other("capability payload limit"));
}
self.used = used;
self.hash.update(part);
Ok(read)
}
}
struct CapabilitySink<'a, S>(&'a mut S);
impl<S: StagingSink> StagingSink for CapabilitySink<'_, S> {
fn row(&mut self, section: Section, key: &[u8], value: &[u8]) -> std::io::Result<()> {
if section == Section::Operations {
crate::application::validate_provenance_encoding(value).map_err(std::io::Error::other)?;
}
self.0.row(section, key, value)
}
}
/// Validates framing, declared capability and operation encoding into isolated staging.
/// Failure may leave partial staging rows, which the caller must discard.
/// Success does not prove cross-table consistency, historical truth or source authority.
pub fn decode_capability_artifact(
artifact: CapabilityArtifact<impl Read>,
sink: &mut impl StagingSink,
limits: Limits,
) -> Result<Header, SnapshotError> {
let manifest = CapabilityManifest::decode(&artifact.manifest.bytes)?;
let length = manifest.payload_length();
limit(length <= limits.total_bytes, "capability payload")?;
let mut reader = CapabilityInput {
io: artifact.payload,
hash: Sha256::new(),
used: 0,
max: length,
};
let header = decode(&mut reader, &mut CapabilitySink(sink), limits)?;
format(reader.used == length, "capability payload length")?;
if reader.hash.finalize().as_slice() != &manifest.bytes[24..] {
return Err(SnapshotError::Digest);
}
Ok(header)
}
/// Row values remain opaque. Operation values will contain the canonical decision
/// LogId followed by the existing canonical operation bytes, validated by callers.
pub struct Row<'a> {
-9
View File
@@ -372,15 +372,6 @@ pub mod config {
with_server_config_write_lock,
};
/// Observe persisted config bytes and their reader-associated metadata within a byte limit.
pub async fn read_config_observation(
api: std::sync::Arc<crate::store::ECStore>,
file: &str,
max_bytes: usize,
) -> crate::error::Result<(Vec<u8>, crate::object_api::ObjectInfo)> {
crate::config::com::read_config_observation(api, file, max_bytes).await
}
/// Read config bytes and metadata without treating an empty payload as absent.
pub async fn read_config_preserve_empty_with_metadata_opts(
api: std::sync::Arc<crate::store::ECStore>,
-176
View File
@@ -449,22 +449,6 @@ where
Ok(data)
}
pub(crate) async fn read_config_observation<S>(api: Arc<S>, file: &str, max_bytes: usize) -> Result<(Vec<u8>, ObjectInfo)>
where
S: EcstoreObjectIO,
{
let opts = ObjectOptions {
suppress_read_repair: true,
no_lock: false,
skip_verify_bitrot: false,
..Default::default()
};
crate::object_api::without_get_object_body_cache_hook(read_config_limited_preserve_empty_with_metadata_opts(
api, file, &opts, max_bytes,
))
.await
}
pub(crate) async fn read_config_limited_preserve_empty<S>(api: Arc<S>, file: &str, max_bytes: usize) -> Result<Vec<u8>>
where
S: EcstoreObjectIO,
@@ -5441,166 +5425,6 @@ mod tests {
assert_eq!(store.write_calls.load(Ordering::SeqCst), 0);
}
struct ObservationStream {
inner: Cursor<Vec<u8>>,
consumed: Arc<AtomicUsize>,
fail: bool,
}
impl AsyncRead for ObservationStream {
fn poll_read(mut self: Pin<&mut Self>, cx: &mut Context<'_>, buf: &mut ReadBuf<'_>) -> Poll<std::io::Result<()>> {
if self.fail {
return Poll::Ready(Err(std::io::Error::other("observation stream failed")));
}
let before = buf.filled().len();
let result = Pin::new(&mut self.inner).poll_read(cx, buf);
self.consumed.fetch_add(buf.filled().len() - before, Ordering::SeqCst);
result
}
}
struct ObservationStore {
body: Vec<u8>,
declared_size: i64,
error: Option<Error>,
stream_error: bool,
consumed: Arc<AtomicUsize>,
options: std::sync::Mutex<Option<(bool, bool, bool)>>,
writes: AtomicUsize,
}
impl Debug for ObservationStore {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
f.debug_struct("ObservationStore").finish()
}
}
impl ObservationStore {
fn new(body: Vec<u8>, declared_size: i64) -> Self {
Self {
body,
declared_size,
error: None,
stream_error: false,
consumed: Arc::new(AtomicUsize::new(0)),
options: std::sync::Mutex::new(None),
writes: AtomicUsize::new(0),
}
}
}
#[async_trait::async_trait]
impl crate::storage_api_contracts::object::ObjectIO for ObservationStore {
type Error = Error;
type RangeSpec = HTTPRangeSpec;
type HeaderMap = HeaderMap;
type ObjectOptions = ObjectOptions;
type ObjectInfo = ObjectInfo;
type GetObjectReader = GetObjectReader;
type PutObjectReader = PutObjReader;
async fn get_object_reader(
&self,
bucket: &str,
object: &str,
range: Option<HTTPRangeSpec>,
_headers: HeaderMap,
opts: &ObjectOptions,
) -> Result<GetObjectReader> {
assert_eq!(bucket, RUSTFS_META_BUCKET);
assert_eq!(object, "config/observation.bin");
assert!(range.is_none());
*self.options.lock().expect("observation options") =
Some((opts.suppress_read_repair, opts.no_lock, opts.skip_verify_bitrot));
if let Some(error) = &self.error {
return Err(error.clone());
}
Ok(GetObjectReader {
stream: Box::new(ObservationStream {
inner: Cursor::new(self.body.clone()),
consumed: self.consumed.clone(),
fail: self.stream_error,
}),
object_info: ObjectInfo {
size: self.declared_size,
etag: Some("observation-etag".to_string()),
..Default::default()
},
buffered_body: None,
body_source: Default::default(),
})
}
async fn put_object(
&self,
_bucket: &str,
_object: &str,
_data: &mut PutObjReader,
_opts: &ObjectOptions,
) -> Result<ObjectInfo> {
self.writes.fetch_add(1, Ordering::SeqCst);
Err(Error::other("observation must not write"))
}
}
#[tokio::test]
async fn observation_preserves_options_and_errors() {
for body in [Vec::new(), b"raw config".to_vec()] {
let store = Arc::new(ObservationStore::new(body.clone(), i64::try_from(body.len()).expect("fixture size")));
let result = super::read_config_observation(store.clone(), "config/observation.bin", 32).await;
assert!(result.is_ok(), "observation must preserve empty and nonempty objects: {result:?}");
let (observed, info) = result.expect("successful observation");
assert_eq!(observed, body);
assert_eq!(info.etag.as_deref(), Some("observation-etag"));
assert_eq!(
*store.options.lock().expect("options"),
Some((true, false, false)),
"I05_REPAIR_SUPPRESSED"
);
assert_eq!(store.writes.load(Ordering::SeqCst), 0);
}
for (source, expected) in [
(Error::FileNotFound, Error::ConfigNotFound),
(
Error::ObjectNotFound(RUSTFS_META_BUCKET.to_string(), "config/observation.bin".to_string()),
Error::ConfigNotFound,
),
(Error::ErasureReadQuorum, Error::ErasureReadQuorum),
] {
let mut fixture = ObservationStore::new(Vec::new(), 0);
fixture.error = Some(source);
let store = Arc::new(fixture);
let result = super::read_config_observation(store.clone(), "config/observation.bin", 32).await;
assert_eq!(result.expect_err("read error must propagate"), expected);
assert_eq!(store.writes.load(Ordering::SeqCst), 0);
}
let mut fixture = ObservationStore::new(Vec::new(), 0);
fixture.stream_error = true;
let store = Arc::new(fixture);
let result = super::read_config_observation(store.clone(), "config/observation.bin", 32).await;
let error = result.expect_err("stream error must propagate");
assert!(error.to_string().contains("observation stream failed"));
assert_eq!(store.writes.load(Ordering::SeqCst), 0);
}
#[tokio::test]
async fn observation_enforces_byte_limit() {
let store = Arc::new(ObservationStore::new(vec![7; 64], 0));
let result = super::read_config_observation(store.clone(), "config/observation.bin", 8).await;
assert_eq!(result.is_err(), true, "I05_BOUND_ENFORCED");
assert_eq!(
result.expect_err("bounded observation error"),
Error::CorruptedFormat,
"I05_BOUND_ENFORCED"
);
assert_eq!(store.consumed.load(Ordering::SeqCst), 9, "I05_BOUND_ENFORCED: drain exactly limit + 1");
let store = Arc::new(ObservationStore::new(vec![7; 8], 8));
let result = super::read_config_observation(store.clone(), "config/observation.bin", 8).await;
assert!(result.is_ok(), "exact byte limit must succeed: {result:?}");
assert_eq!(result.expect("exact limit").0, vec![7; 8]);
assert_eq!(store.consumed.load(Ordering::SeqCst), 8);
}
#[tokio::test]
async fn read_config_preserve_empty_distinguishes_empty_object() {
let store = Arc::new(RecoveryMockStore::new(RecoveryReadState::Blob(Vec::new()), None));
-211
View File
@@ -1095,217 +1095,6 @@ mod prepared_get_object_metadata_tests {
use http::HeaderMap;
use tokio::io::AsyncReadExt;
const OBSERVATION_HOOK_BODY_LEN: usize = 1024 * 1024 + 1;
struct ObservationBodyHook {
object: &'static str,
calls: Arc<AtomicU64>,
body: bytes::Bytes,
}
#[async_trait::async_trait]
impl crate::object_api::GetObjectBodyCacheHook for ObservationBodyHook {
async fn lookup(&self, bucket: &str, object: &str, info: &ObjectInfo) -> Option<bytes::Bytes> {
if bucket == crate::disk::RUSTFS_META_BUCKET && object == self.object {
assert_eq!(info.size, i64::try_from(self.body.len()).expect("hook fixture size"));
self.calls.fetch_add(1, Ordering::SeqCst);
Some(self.body.clone())
} else {
None
}
}
}
struct ObservationHookGuard;
impl Drop for ObservationHookGuard {
fn drop(&mut self) {
crate::object_api::unregister_get_object_body_cache_hook();
}
}
async fn ordinary_observation_get(set: &SetDisks, object: &str) -> Vec<u8> {
let mut reader = set
.get_object_reader(crate::disk::RUSTFS_META_BUCKET, object, None, HeaderMap::new(), &ObjectOptions::default())
.await
.expect("ordinary GET positive control");
let mut bytes = Vec::new();
reader.stream.read_to_end(&mut bytes).await.expect("ordinary GET body");
bytes
}
fn observation_register_hook(object: &'static str, body: Vec<u8>) -> (ObservationHookGuard, Arc<AtomicU64>) {
let calls = Arc::new(AtomicU64::new(0));
crate::object_api::register_get_object_body_cache_hook(Arc::new(ObservationBodyHook {
object,
calls: calls.clone(),
body: bytes::Bytes::from(body),
}));
(ObservationHookGuard, calls)
}
#[tokio::test]
#[serial_test::serial(body_cache_hook)]
async fn observation_real_backend_boundaries() {
let object = "config/observation-real.bin";
let payload = b"persisted config";
let (_dirs, set) = make_local_set_disks(4, 2).await;
let mut reader = PutObjReader::from_vec(payload.to_vec());
let written = set
.put_object(crate::disk::RUSTFS_META_BUCKET, object, &mut reader, &ObjectOptions::default())
.await
.expect("persist real config fixture");
let result = crate::config::com::read_config_observation(set.clone(), object, payload.len()).await;
assert_eq!(result.is_ok(), true, "I05_OBSERVATION_SUCCESS: {result:?}");
let (bytes, info) = result.expect("real observation success");
assert_eq!(bytes, payload);
assert_eq!(info.size, written.size);
assert_eq!(info.etag, written.etag);
assert_eq!(info.data_dir, written.data_dir);
assert_eq!(info.name, object);
let over = crate::config::com::read_config_observation(set.clone(), object, payload.len() - 1).await;
assert_eq!(over.is_err(), true, "I05_BOUND_ENFORCED");
assert_eq!(over.expect_err("over limit must fail"), Error::CorruptedFormat, "I05_BOUND_ENFORCED");
let absent = crate::config::com::read_config_observation(set.clone(), "config/observation-absent.bin", 32).await;
assert_eq!(absent.expect_err("absent config"), Error::ConfigNotFound);
let empty_object = "config/observation-empty.bin";
let mut empty = PutObjReader::from_vec(Vec::new());
set.put_object(crate::disk::RUSTFS_META_BUCKET, empty_object, &mut empty, &ObjectOptions::default())
.await
.expect("persist empty config");
let empty = crate::config::com::read_config_observation(set, empty_object, 0).await;
assert!(empty.is_ok(), "existing empty config must succeed: {empty:?}");
let (bytes, info) = empty.expect("empty observation");
assert!(bytes.is_empty());
assert_eq!(info.size, 0);
}
#[tokio::test]
#[serial_test::serial(body_cache_hook)]
async fn observation_bypasses_registered_body_hook() {
let object = "config/observation-hook.bin";
let (_dirs, set) = make_local_set_disks(4, 2).await;
let mut reader = PutObjReader::from_vec(vec![b'p'; OBSERVATION_HOOK_BODY_LEN]);
set.put_object(crate::disk::RUSTFS_META_BUCKET, object, &mut reader, &ObjectOptions::default())
.await
.expect("persist real config fixture");
let (_hook, calls) = observation_register_hook(object, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(ordinary_observation_get(&set, object).await, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(calls.swap(0, Ordering::SeqCst), 1);
let result = crate::config::com::read_config_observation(set, object, OBSERVATION_HOOK_BODY_LEN).await;
assert_eq!(result.is_ok(), true, "I05_HOOK_BYPASS: {result:?}");
assert_eq!(
result.expect("observed persisted body").0,
vec![b'p'; OBSERVATION_HOOK_BODY_LEN],
"I05_HOOK_BYPASS"
);
assert_eq!(calls.load(Ordering::SeqCst), 0, "I05_HOOK_BYPASS");
}
#[tokio::test]
#[serial_test::serial(body_cache_hook)]
async fn observation_hook_scope_restores_after_error() {
let object = "config/observation-hook-error.bin";
let (_dirs, set) = make_local_set_disks(4, 2).await;
let mut reader = PutObjReader::from_vec(vec![b'p'; OBSERVATION_HOOK_BODY_LEN]);
set.put_object(crate::disk::RUSTFS_META_BUCKET, object, &mut reader, &ObjectOptions::default())
.await
.expect("persist real config fixture");
let (_hook, calls) = observation_register_hook(object, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
let result = crate::config::com::read_config_observation(set.clone(), object, OBSERVATION_HOOK_BODY_LEN - 1).await;
assert_eq!(result.expect_err("bounded observation error"), Error::CorruptedFormat);
assert_eq!(calls.load(Ordering::SeqCst), 0);
assert_eq!(ordinary_observation_get(&set, object).await, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(calls.load(Ordering::SeqCst), 1);
}
struct PendingObservationStore {
set: Arc<SetDisks>,
ready: std::sync::Mutex<Option<tokio::sync::oneshot::Sender<()>>>,
}
impl std::fmt::Debug for PendingObservationStore {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("PendingObservationStore").finish()
}
}
#[async_trait::async_trait]
impl crate::storage_api_contracts::object::ObjectIO for PendingObservationStore {
type Error = Error;
type RangeSpec = HTTPRangeSpec;
type HeaderMap = HeaderMap;
type ObjectOptions = ObjectOptions;
type ObjectInfo = ObjectInfo;
type GetObjectReader = GetObjectReader;
type PutObjectReader = PutObjReader;
async fn get_object_reader(
&self,
bucket: &str,
object: &str,
range: Option<HTTPRangeSpec>,
headers: HeaderMap,
opts: &ObjectOptions,
) -> Result<GetObjectReader> {
let mut reader = self.set.get_object_reader(bucket, object, range, headers, opts).await?;
let mut body = Vec::new();
reader.stream.read_to_end(&mut body).await.expect("pending producer real GET");
assert_eq!(
body,
vec![b'p'; OBSERVATION_HOOK_BODY_LEN],
"I05_HOOK_BYPASS: producer must enter its own scope"
);
self.ready
.lock()
.expect("pending signal")
.take()
.expect("one GET")
.send(())
.expect("producer entry signal");
std::future::pending::<Result<GetObjectReader>>().await
}
async fn put_object(
&self,
_bucket: &str,
_object: &str,
_data: &mut PutObjReader,
_opts: &ObjectOptions,
) -> Result<ObjectInfo> {
assert!(false, "observation must not PUT");
Err(Error::other("unexpected observation PUT"))
}
}
#[tokio::test]
#[serial_test::serial(body_cache_hook)]
async fn observation_hook_scope_restores_after_cancel() {
let object = "config/observation-hook-cancel.bin";
let (_dirs, set) = make_local_set_disks(4, 2).await;
let mut reader = PutObjReader::from_vec(vec![b'p'; OBSERVATION_HOOK_BODY_LEN]);
set.put_object(crate::disk::RUSTFS_META_BUCKET, object, &mut reader, &ObjectOptions::default())
.await
.expect("persist real config fixture");
let (_hook, calls) = observation_register_hook(object, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(ordinary_observation_get(&set, object).await, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(calls.swap(0, Ordering::SeqCst), 1);
let (ready_tx, ready_rx) = tokio::sync::oneshot::channel();
let adapter = Arc::new(PendingObservationStore {
set: set.clone(),
ready: std::sync::Mutex::new(Some(ready_tx)),
});
let mut scope = Box::pin(crate::config::com::read_config_observation(adapter, object, OBSERVATION_HOOK_BODY_LEN));
tokio::select! {
result = ready_rx => { result.expect("producer reached pending GET"); }
result = &mut scope => { assert!(false, "producer must remain pending until cancelled: {result:?}"); }
}
assert_eq!(calls.load(Ordering::SeqCst), 0);
drop(scope);
assert_eq!(ordinary_observation_get(&set, object).await, vec![b'x'; OBSERVATION_HOOK_BODY_LEN]);
assert_eq!(calls.load(Ordering::SeqCst), 1);
}
const READ_VERSION_BARRIER_GUARD: std::time::Duration = std::time::Duration::from_secs(10);
fn object_with_initial_data_shards(bucket: &str, prefix: &str) -> String {