Compare commits

...

1 Commits

Author SHA1 Message Date
overtrue 60d90f085a feat(ecstore): add bounded persisted config observations 2026-10-11 20:29:27 +08:00
3 changed files with 396 additions and 0 deletions
+9
View File
@@ -372,6 +372,15 @@ 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
@@ -449,6 +449,22 @@ 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,
@@ -5425,6 +5441,166 @@ 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
@@ -1095,6 +1095,217 @@ 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 {