Compare commits

..

2 Commits

Author SHA1 Message Date
overtrue 63cf8e2f2f feat(capture-authority): add native OpenRaft log reader 2026-10-11 13:01:43 +08:00
overtrue b7fe40f4b3 feat(capture): support complete native log range bounds 2026-10-11 12:28:23 +08:00
4 changed files with 334 additions and 11 deletions
Generated
+1
View File
@@ -9810,6 +9810,7 @@ dependencies = [
"serde_json",
"tempfile",
"thiserror 2.0.21",
"tokio",
"url",
]
+1
View File
@@ -12,6 +12,7 @@ openraft = { workspace = true }
redb = { workspace = true }
serde = { workspace = true, features = ["derive"] }
thiserror = { workspace = true }
tokio = { workspace = true }
url = { workspace = true }
[dev-dependencies]
+47 -11
View File
@@ -1,3 +1,40 @@
fn read_native_range_in<R: std::ops::RangeBounds<u64>>(
tx: &redb::ReadTransaction,
range: R,
) -> Result<Vec<crate::CaptureRaftEntry>, StoreError> {
use std::ops::Bound::{Excluded, Included, Unbounded};
let logs = tx.open_table(LOGS)?;
let start = match range.start_bound() {
Included(value) => *value,
Excluded(value) => match value.checked_add(1) {
Some(value) => value,
None => return Ok(Vec::new()),
},
Unbounded => 0,
};
let end = match range.end_bound() {
Included(value) => *value,
Excluded(value) => match value.checked_sub(1) {
Some(value) => value,
None => return Ok(Vec::new()),
},
Unbounded => u64::MAX,
};
if start > end {
return Ok(Vec::new());
}
let mut entries = Vec::new();
for row in logs.range(start..=end)? {
let (key, value) = row?;
let entry = decode_log(value.value())?;
if entry.id.index != key.value() {
return Err(StoreError::CorruptRecord);
}
entries.push(decode_native_entry(entry)?);
}
Ok(entries)
}
use crate::{
LOGS, LogId, META, Store, StoreError, committed::validate_committed, created::*, decode_log, encode_id, encode_log,
next_index, number,
@@ -368,6 +405,14 @@ impl Store {
})
.transpose()
}
/// Read typed native entries within the requested bounds.
pub fn read_native_range<R: std::ops::RangeBounds<u64>>(&self, range: R) -> Result<Vec<crate::CaptureRaftEntry>, StoreError> {
let tx = self.database.begin_read()?;
if number(required(tx.open_table(META)?.get("schema")?)?.value())? != 4 {
return Err(StoreError::UnsupportedSchema);
}
read_native_range_in(&tx, range)
}
pub fn read_native_entries(&self, start: u64, end: u64) -> Result<Vec<crate::CaptureRaftEntry>, StoreError> {
let tx = self.database.begin_read()?;
if number(required(tx.open_table(META)?.get("schema")?)?.value())? != 4 {
@@ -376,18 +421,9 @@ impl Store {
if start > end {
return Err(StoreError::InvalidLogSequence);
}
let logs = tx.open_table(LOGS)?;
let mut entries = Vec::new();
for row in logs.range(start..end)? {
let (key, value) = row?;
let entry = decode_log(value.value())?;
if entry.id.index != key.value() {
return Err(StoreError::CorruptRecord);
}
entries.push(decode_native_entry(entry)?);
}
Ok(entries)
read_native_range_in(&tx, start..end)
}
pub fn apply_next_native(&mut self, id: LogId) -> Result<crate::CaptureResponse, StoreError> {
let tx = self.write_transaction()?;
let response;
+285
View File
@@ -23,6 +23,61 @@ pub use raft_types::{
mod application_tests;
mod truncate;
/// OpenRaft log reader backed by exclusive local authority storage.
#[derive(Clone)]
pub struct NativeLogReader {
store: std::sync::Arc<std::sync::Mutex<Store>>,
}
impl NativeLogReader {
pub fn new(store: Store) -> Self {
Self {
store: std::sync::Arc::new(std::sync::Mutex::new(store)),
}
}
}
#[derive(Debug, thiserror::Error)]
enum NativeReaderError {
#[error(transparent)]
Store(#[from] StoreError),
#[error("native log reader lock poisoned")]
Poisoned,
#[error("native log reader worker failed")]
Worker,
}
fn native_read_error(error: NativeReaderError) -> openraft::StorageError<u64> {
openraft::StorageError::IO {
source: openraft::StorageIOError::read_logs(openraft::AnyError::new(&error)),
}
}
async fn native_read_worker<F>(
store: std::sync::Arc<std::sync::Mutex<Store>>,
read: F,
) -> Result<Vec<CaptureRaftEntry>, openraft::StorageError<u64>>
where
F: FnOnce(&Store) -> Result<Vec<CaptureRaftEntry>, StoreError> + Send + 'static,
{
tokio::task::spawn_blocking(move || {
let store = store.lock().map_err(|_| native_read_error(NativeReaderError::Poisoned))?;
read(&store).map_err(|error| native_read_error(NativeReaderError::Store(error)))
})
.await
.map_err(|_| native_read_error(NativeReaderError::Worker))?
}
impl openraft::RaftLogReader<CaptureTypeConfig> for NativeLogReader {
async fn try_get_log_entries<RB: std::ops::RangeBounds<u64> + Clone + std::fmt::Debug + openraft::OptionalSend>(
&mut self,
range: RB,
) -> Result<Vec<CaptureRaftEntry>, openraft::StorageError<u64>> {
let bounds = (range.start_bound().cloned(), range.end_bound().cloned());
native_read_worker(self.store.clone(), move |store| store.read_native_range(bounds)).await
}
}
const META: TableDefinition<&str, &[u8]> = TableDefinition::new("meta");
const VOTE: TableDefinition<u8, &[u8]> = TableDefinition::new("vote");
const LOGS: TableDefinition<u64, &[u8]> = TableDefinition::new("logs");
@@ -459,6 +514,153 @@ mod tests {
node_id: 0,
}
}
fn u19_runtime() -> tokio::runtime::Runtime {
tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.expect("reader runtime")
}
fn u19_assert_io(error: openraft::StorageError<u64>) {
match error {
openraft::StorageError::IO { source } => {
assert!(source.to_string().starts_with("when Read Logs:"), "U19_READER_ERROR: {source}");
}
other => panic!("U19_READER_ERROR: unexpected error {other:?}"),
}
}
#[test]
fn u19_reader_roundtrip() {
use openraft::RaftLogReader;
let (_dir, _backend, mut store) = u17a_store();
let entries = vec![u17a_blank(0), u17a_membership(1).0, u17a_created(2, u17a_command())];
store.append_native(&entries).expect("append reader fixture");
let mut reader = NativeLogReader::new(store);
let read = u19_runtime().block_on(reader.try_get_log_entries(..)).expect("trait read");
assert_eq!(read, entries, "U19_TRAIT_READ_SUCCESS");
}
#[test]
fn u19_reader_bounds_reopen() {
use openraft::RaftLogReader;
use std::ops::Bound::{Excluded, Included, Unbounded};
let (dir, backend, mut store) = u17a_store();
let entries = vec![u17a_blank(0), u17a_membership(1).0, u17a_created(2, u17a_command())];
store.append_native(&entries).expect("append range fixture");
drop(store);
drop(backend);
let store = Store::open(dir.path().join("db"), identity()).expect("reopen reader file");
let mut reader = NativeLogReader::new(store);
u19_runtime().block_on(async {
assert_eq!(
reader.try_get_log_entries(1..2).await.expect("bounded"),
entries[1..2],
"U19_READER_RANGE"
);
assert_eq!(reader.try_get_log_entries(..=1).await.expect("inclusive"), entries[..2]);
assert_eq!(
reader
.try_get_log_entries((Excluded(0), Included(2)))
.await
.expect("excluded"),
entries[1..]
);
assert_eq!(reader.try_get_log_entries(2..).await.expect("unbounded end"), entries[2..]);
assert_eq!(
reader
.try_get_log_entries((Unbounded, Included(u64::MAX)))
.await
.expect("max"),
entries
);
for range in [
(Included(2), Excluded(2)),
(Included(2), Included(1)),
(Excluded(u64::MAX), Unbounded),
] {
assert!(reader.try_get_log_entries(range).await.expect("empty bounds").is_empty());
}
let start = 1;
let end = 2;
assert_eq!(
reader
.try_get_log_entries((Included(&start), Excluded(&end)))
.await
.expect("borrowed bounds"),
entries[1..2]
);
});
}
#[test]
fn u19_reader_legacy_corrupt() {
use openraft::RaftLogReader;
let (_legacy_dir, _legacy_backend, legacy) = u17a_bound();
let mut legacy = NativeLogReader::new(legacy);
let runtime = u19_runtime();
u19_assert_io(
runtime
.block_on(legacy.try_get_log_entries(0..0))
.expect_err("U19_READER_ERROR: legacy empty range"),
);
let (_dir, _backend, mut store) = u17a_store();
store.append_native(&[u17a_blank(0)]).expect("append corruption fixture");
let tx = store.database.begin_write().expect("corruption transaction");
{
let mut logs = tx.open_table(LOGS).expect("logs");
logs.insert(0, b"corrupt".as_slice()).expect("corrupt row");
}
tx.commit().expect("persist corruption");
let mut reader = NativeLogReader::new(store);
u19_assert_io(
runtime
.block_on(reader.try_get_log_entries(..))
.expect_err("U19_READER_ERROR: corrupt row"),
);
}
#[test]
fn u19_reader_worker_thread() {
use openraft::RaftLogReader;
let (_dir, _backend, mut store) = u17a_store();
let entries = vec![u17a_blank(0)];
store.append_native(&entries).expect("append worker fixture");
let mut reader = NativeLogReader::new(store);
let runtime = u19_runtime();
let async_thread = std::thread::current().id();
let (sender, receiver) = mpsc::channel();
runtime.block_on(async {
let read = native_read_worker(reader.store.clone(), move |store| {
sender.send(std::thread::current().id()).expect("send worker thread");
store.read_native_range(..)
})
.await
.expect("worker read");
assert_eq!(read, entries);
assert_ne!(receiver.recv().expect("worker thread"), async_thread, "U19_BLOCKING_THREAD");
assert_eq!(reader.try_get_log_entries(..).await.expect("trait read"), entries);
});
}
#[test]
fn u19_reader_worker_errors() {
use openraft::RaftLogReader;
let (_dir, _backend, store) = u17a_store();
let mut reader = NativeLogReader::new(store);
let runtime = u19_runtime();
let error = runtime
.block_on(native_read_worker(reader.store.clone(), |_| panic!("test worker panic")))
.expect_err("worker panic must fail");
assert!(error.to_string().contains("native log reader worker failed"));
u19_assert_io(error);
let error = runtime
.block_on(reader.try_get_log_entries(..))
.expect_err("poison must fail");
assert!(error.to_string().contains("native log reader lock poisoned"));
u19_assert_io(error);
}
fn u17a_id(index: u64) -> LogId {
LogId {
term: 11,
@@ -535,6 +737,89 @@ mod tests {
)
}
#[test]
fn u18_native_read_roundtrip() {
let (_dir, _backend, mut store) = u17a_store();
let command = u17a_command();
let entries = vec![u17a_blank(0), u17a_membership(1).0, u17a_created(2, command.clone())];
store.append_native(&entries).expect("append native fixture");
let result = store.read_native_range(..);
assert!(result.is_ok(), "U18_NATIVE_READ_SUCCESS: {result:?}");
let read = result.expect("native read");
assert_eq!(read, entries, "full identities and payloads");
assert_eq!(
store.read_entries(2, 3).expect("canonical row")[0].payload[14..],
command.encode().expect("canonical command")
);
}
#[test]
fn u18_native_read_range_bounds() {
use std::ops::Bound::{Excluded, Included, Unbounded};
let (_dir, _backend, mut store) = u17a_store();
let entries = vec![u17a_blank(0), u17a_membership(1).0, u17a_created(2, u17a_command())];
store.append_native(&entries).expect("append range fixture");
assert_eq!(store.read_native_range(1..2).expect("bounded read"), entries[1..2], "U18_RANGE_BOUNDARY");
assert_eq!(store.read_native_range(..=1).expect("inclusive end"), entries[..2]);
assert_eq!(store.read_native_range((Excluded(0), Included(2))).expect("excluded start"), entries[1..]);
assert_eq!(store.read_native_range(2..).expect("unbounded end"), entries[2..]);
assert_eq!(store.read_native_range(..).expect("all"), entries);
for range in [
(Included(2), Excluded(2)),
(Included(2), Included(1)),
(Unbounded, Excluded(0)),
(Excluded(u64::MAX), Unbounded),
(Included(u64::MAX), Included(u64::MAX)),
] {
assert!(store.read_native_range(range).expect("empty mathematical range").is_empty());
}
}
#[test]
fn u18_native_read_reopen() {
let (dir, backend, mut store) = u17a_store();
let entries = vec![u17a_blank(0), u17a_membership(1).0, u17a_created(2, u17a_command())];
store.append_native(&entries).expect("append reopen fixture");
drop(store);
drop(backend);
let store = Store::open(dir.path().join("db"), identity()).expect("reopen native file");
assert_eq!(store.read_native_range(..).expect("reopened native read"), entries);
}
#[test]
fn u18_native_read_legacy_rejected() {
let (_dir, _backend, store) = u17a_bound();
assert!(
matches!(store.read_native_range(0..0), Err(StoreError::UnsupportedSchema)),
"U18_SCHEMA_GUARD"
);
assert!(matches!(store.read_native_range(..), Err(StoreError::UnsupportedSchema)));
}
#[test]
fn u18_native_read_corrupt_row() {
let (_dir, _backend, mut store) = u17a_store();
store.append_native(&[u17a_blank(0)]).expect("append corruption fixture");
let original = store.read_entry(0).expect("read stored entry").expect("present entry");
let tx = store.database.begin_write().expect("corruption transaction");
{
let mut logs = tx.open_table(LOGS).expect("logs");
let bytes = logs.get(0).expect("get row").expect("present row").value().to_vec();
logs.insert(1, bytes.as_slice()).expect("mismatched index row");
}
tx.commit().expect("persist mismatched row");
assert!(matches!(store.read_native_range(1..=1), Err(StoreError::CorruptRecord)));
let tx = store.database.begin_write().expect("malformed transaction");
{
let mut logs = tx.open_table(LOGS).expect("logs");
let mut entry = original;
entry.payload = b"CENR\x01".to_vec();
let bytes = encode_log(&entry).expect("encode malformed native envelope");
logs.insert(0, bytes.as_slice()).expect("malformed native row");
}
tx.commit().expect("persist malformed native row");
assert!(matches!(store.read_native_range(..1), Err(StoreError::CorruptRecord)));
let tx = store.database.begin_write().expect("missing-table transaction");
assert!(tx.delete_table(LOGS).expect("delete logs table"));
tx.commit().expect("persist missing logs table");
assert!(store.read_native_entries(0, 0).is_err(), "U18_EMPTY_RANGE_TABLE_ERROR");
}
#[test]
fn u17a_native_enable_empty_idempotent() {
let (_dir, backend, mut store) = u17a_bound();
assert_eq!(committed_schema(&store), 3);