fix: preserve invocation recovery across rewind and revalidate replay authority
This commit is contained in:
@@ -383,8 +383,11 @@ impl Store for FsStore {
|
||||
content.push_str(&serde_json::to_string(entry)?);
|
||||
content.push('\n');
|
||||
}
|
||||
fs::write(&path, content.as_bytes())?;
|
||||
Ok(())
|
||||
let _guard = self
|
||||
.append_lock
|
||||
.lock()
|
||||
.map_err(|_| std::io::Error::other("session store append lock was poisoned"))?;
|
||||
crate::worker_session_store::atomic_write_bytes(&path, content.as_bytes())
|
||||
}
|
||||
|
||||
fn exists(&self, session_id: SessionId, segment_id: SegmentId) -> Result<bool, StoreError> {
|
||||
|
||||
@@ -110,7 +110,14 @@ pub trait Store: Send + Sync {
|
||||
/// segment ID without its session ID (e.g. legacy `--session <UUID>`).
|
||||
fn lookup_session_of(&self, segment_id: SegmentId) -> Result<Option<SessionId>, StoreError>;
|
||||
|
||||
/// Create a new segment within `session_id`, with initial entries.
|
||||
/// Create or atomically replace a segment within `session_id` with all entries.
|
||||
///
|
||||
/// Readers and crash recovery must observe either the previous complete log
|
||||
/// or the complete replacement, never a truncated prefix or partial record.
|
||||
/// Failure before replacement leaves the old log intact; an error after the
|
||||
/// atomic commit (for example a durability sync failure) may leave the complete
|
||||
/// replacement. Filesystem implementations must sync replacement bytes before
|
||||
/// rename and the containing directory afterward before reporting success.
|
||||
fn create_segment(
|
||||
&self,
|
||||
session_id: SessionId,
|
||||
|
||||
@@ -1166,7 +1166,7 @@ fn atomic_write_json<T: Serialize>(path: &Path, value: &T) -> Result<(), StoreEr
|
||||
atomic_write_bytes(path, &bytes)
|
||||
}
|
||||
|
||||
fn atomic_write_bytes(path: &Path, bytes: &[u8]) -> Result<(), StoreError> {
|
||||
pub(crate) fn atomic_write_bytes(path: &Path, bytes: &[u8]) -> Result<(), StoreError> {
|
||||
let parent = path
|
||||
.parent()
|
||||
.ok_or_else(|| std::io::Error::other("Worker Session path has no parent"))?;
|
||||
|
||||
@@ -115,6 +115,100 @@ fn create_segment_writes_all_entries() {
|
||||
assert_eq!(state.session_id, Some(sid));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn create_segment_replacement_keeps_open_old_log_and_commits_complete_new_log() {
|
||||
use std::io::Read;
|
||||
let dir = tempfile::tempdir().unwrap();
|
||||
let store = FsStore::new(dir.path()).unwrap();
|
||||
let sid = new_session_id();
|
||||
let segid = new_segment_id();
|
||||
let old = vec![
|
||||
nil_session_start(1, sid),
|
||||
LogEntry::TurnEnd {
|
||||
ts: 2,
|
||||
turn_count: 1,
|
||||
},
|
||||
];
|
||||
store.create_segment(sid, segid, &old).unwrap();
|
||||
let path = dir
|
||||
.path()
|
||||
.join(sid.to_string())
|
||||
.join(format!("{segid}.jsonl"));
|
||||
let old_bytes = std::fs::read(&path).unwrap();
|
||||
let mut old_reader = std::fs::File::open(&path).unwrap();
|
||||
let new = vec![
|
||||
nil_session_start(1, sid),
|
||||
LogEntry::Extension {
|
||||
ts: 3,
|
||||
domain: "recovery-checkpoint".into(),
|
||||
payload: serde_json::json!({"receipt": "retained"}),
|
||||
},
|
||||
];
|
||||
store.create_segment(sid, segid, &new).unwrap();
|
||||
let mut retained_inode = Vec::new();
|
||||
old_reader.read_to_end(&mut retained_inode).unwrap();
|
||||
// An in-place fs::write would truncate the inode held by this reader.
|
||||
assert_eq!(retained_inode, old_bytes);
|
||||
let reopened = FsStore::new(dir.path()).unwrap();
|
||||
assert_eq!(
|
||||
serde_json::to_value(reopened.read_all(sid, segid).unwrap()).unwrap(),
|
||||
serde_json::to_value(&new).unwrap()
|
||||
);
|
||||
assert_eq!(reopened.read_entry_count(sid, segid).unwrap(), new.len());
|
||||
assert_eq!(
|
||||
std::fs::read_dir(path.parent().unwrap()).unwrap().count(),
|
||||
1,
|
||||
"successful replacement must not leave temporary files"
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn create_segment_atomic_replacement_readers_only_observe_complete_old_or_new() {
|
||||
let dir = tempfile::tempdir().unwrap();
|
||||
let store = FsStore::new(dir.path()).unwrap();
|
||||
let sid = new_session_id();
|
||||
let segid = new_segment_id();
|
||||
let old = vec![
|
||||
nil_session_start(1, sid),
|
||||
LogEntry::Extension {
|
||||
ts: 2,
|
||||
domain: "old-evidence".into(),
|
||||
payload: serde_json::json!({"result": "old"}),
|
||||
},
|
||||
];
|
||||
let new = vec![
|
||||
nil_session_start(1, sid),
|
||||
LogEntry::Extension {
|
||||
ts: 3,
|
||||
domain: "new-checkpoint".into(),
|
||||
payload: serde_json::json!({"result": "new"}),
|
||||
},
|
||||
];
|
||||
store.create_segment(sid, segid, &old).unwrap();
|
||||
let old_json = serde_json::to_value(&old).unwrap();
|
||||
let new_json = serde_json::to_value(&new).unwrap();
|
||||
let barrier = std::sync::Arc::new(std::sync::Barrier::new(2));
|
||||
let reader_store = store.clone();
|
||||
let reader_barrier = barrier.clone();
|
||||
let reader = std::thread::spawn(move || {
|
||||
reader_barrier.wait();
|
||||
for _ in 0..1000 {
|
||||
let observed =
|
||||
serde_json::to_value(reader_store.read_all(sid, segid).unwrap()).unwrap();
|
||||
assert!(
|
||||
observed == old_json || observed == new_json,
|
||||
"replacement must never expose a partial log: {observed}"
|
||||
);
|
||||
}
|
||||
});
|
||||
barrier.wait();
|
||||
for _ in 0..50 {
|
||||
store.create_segment(sid, segid, &new).unwrap();
|
||||
store.create_segment(sid, segid, &old).unwrap();
|
||||
}
|
||||
reader.join().unwrap();
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn list_sessions_and_segments() {
|
||||
let dir = tempfile::tempdir().unwrap();
|
||||
|
||||
Reference in New Issue
Block a user