fix: stabilize persisted run transition identities
This commit is contained in:
@@ -211,7 +211,11 @@ impl FsStore {
|
||||
}
|
||||
}
|
||||
|
||||
fn parse_jsonl<T: serde::de::DeserializeOwned>(content: &[u8]) -> Result<Vec<T>, StoreError> {
|
||||
fn parse_jsonl(
|
||||
content: &[u8],
|
||||
session_id: SessionId,
|
||||
segment_id: SegmentId,
|
||||
) -> Result<Vec<LogEntry>, StoreError> {
|
||||
let complete = Self::complete_jsonl_prefix(content);
|
||||
let content = std::str::from_utf8(complete).map_err(|error| StoreError::Corrupt {
|
||||
line: complete[..error.valid_up_to()]
|
||||
@@ -222,15 +226,19 @@ impl FsStore {
|
||||
message: error.to_string(),
|
||||
})?;
|
||||
let mut entries = Vec::new();
|
||||
for (i, line) in content.lines().enumerate() {
|
||||
if line.trim().is_empty() {
|
||||
continue;
|
||||
let mut record_start = 0_u64;
|
||||
for (index, record) in content.split_inclusive('\n').enumerate() {
|
||||
let line = record.strip_suffix('\n').unwrap_or(record);
|
||||
if !line.trim().is_empty() {
|
||||
let mut entry: LogEntry =
|
||||
serde_json::from_str(line).map_err(|error| StoreError::Corrupt {
|
||||
line: index + 1,
|
||||
message: error.to_string(),
|
||||
})?;
|
||||
entry.ensure_legacy_session_entry_id(session_id, segment_id, record_start);
|
||||
entries.push(entry);
|
||||
}
|
||||
let entry: T = serde_json::from_str(line).map_err(|e| StoreError::Corrupt {
|
||||
line: i + 1,
|
||||
message: e.to_string(),
|
||||
})?;
|
||||
entries.push(entry);
|
||||
record_start = record_start.saturating_add(record.len() as u64);
|
||||
}
|
||||
Ok(entries)
|
||||
}
|
||||
@@ -294,7 +302,7 @@ impl Store for FsStore {
|
||||
return Err(StoreError::NotFound(segment_id));
|
||||
}
|
||||
let content = fs::read(&path)?;
|
||||
Self::parse_jsonl(&content)
|
||||
Self::parse_jsonl(&content, session_id, segment_id)
|
||||
}
|
||||
|
||||
fn list_sessions(&self) -> Result<Vec<SessionId>, StoreError> {
|
||||
|
||||
@@ -4,7 +4,10 @@ use serde::{Deserialize, Serialize};
|
||||
|
||||
use crate::LoggedItem;
|
||||
|
||||
/// Stable logical identity of one model-visible history entry.
|
||||
/// Stable logical identity of one durable public Session entry.
|
||||
///
|
||||
/// Model-visible history entries and typed Run transitions share this identity
|
||||
/// space so every projection can reconcile the same committed record.
|
||||
///
|
||||
/// This value is generated at the trusted Worker session boundary and copied
|
||||
/// unchanged across fork, rewind, compaction retention, restore, and reboot.
|
||||
|
||||
@@ -1041,15 +1041,16 @@ fn project_session_snapshot_for_segment(
|
||||
)),
|
||||
LogEntry::RunYielded {
|
||||
ts,
|
||||
entry_id,
|
||||
reason,
|
||||
active_run_turn_count: count,
|
||||
} => {
|
||||
active_run_turn_count = Some(*count);
|
||||
entries.push(legacy_entry(
|
||||
entries.push(run_transition_entry(
|
||||
entry_id.as_ref(),
|
||||
&session_key,
|
||||
segment_id.as_ref(),
|
||||
log_index,
|
||||
0,
|
||||
*ts,
|
||||
SessionSnapshotEntryData::RunYielded {
|
||||
reason: *reason,
|
||||
@@ -1059,15 +1060,16 @@ fn project_session_snapshot_for_segment(
|
||||
}
|
||||
LogEntry::RunResumed {
|
||||
ts,
|
||||
entry_id,
|
||||
source,
|
||||
active_run_turn_count: count,
|
||||
} => {
|
||||
active_run_turn_count = Some(*count);
|
||||
entries.push(legacy_entry(
|
||||
entries.push(run_transition_entry(
|
||||
entry_id.as_ref(),
|
||||
&session_key,
|
||||
segment_id.as_ref(),
|
||||
log_index,
|
||||
0,
|
||||
*ts,
|
||||
SessionSnapshotEntryData::RunResumed {
|
||||
source: *source,
|
||||
@@ -1075,29 +1077,30 @@ fn project_session_snapshot_for_segment(
|
||||
},
|
||||
));
|
||||
}
|
||||
LogEntry::RunCancelled { ts } => {
|
||||
LogEntry::RunCancelled { ts, entry_id } => {
|
||||
active_run_turn_count = None;
|
||||
entries.push(legacy_entry(
|
||||
entries.push(run_transition_entry(
|
||||
entry_id.as_ref(),
|
||||
&session_key,
|
||||
segment_id.as_ref(),
|
||||
log_index,
|
||||
0,
|
||||
*ts,
|
||||
SessionSnapshotEntryData::RunCancelled,
|
||||
));
|
||||
}
|
||||
LogEntry::RunErrored {
|
||||
ts,
|
||||
entry_id,
|
||||
message,
|
||||
failure,
|
||||
..
|
||||
} => {
|
||||
active_run_turn_count = None;
|
||||
entries.push(legacy_entry(
|
||||
entries.push(run_transition_entry(
|
||||
entry_id.as_ref(),
|
||||
&session_key,
|
||||
segment_id.as_ref(),
|
||||
log_index,
|
||||
0,
|
||||
*ts,
|
||||
SessionSnapshotEntryData::RunError {
|
||||
message: message.clone(),
|
||||
@@ -1234,6 +1237,25 @@ fn derivation_ids(entry: &crate::LoggedSystemHistoryEntry) -> Vec<String> {
|
||||
.unwrap_or_default()
|
||||
}
|
||||
|
||||
fn run_transition_entry(
|
||||
persisted_entry_id: Option<&crate::LoggedSessionHistoryEntryId>,
|
||||
session_key: &SessionId,
|
||||
segment_id: Option<&SegmentId>,
|
||||
log_index: usize,
|
||||
timestamp: u64,
|
||||
data: SessionSnapshotEntryData,
|
||||
) -> SessionSnapshotEntry {
|
||||
SessionSnapshotEntry {
|
||||
entry_id: persisted_entry_id
|
||||
.map(|entry_id| entry_id.0.clone())
|
||||
.unwrap_or_else(|| legacy_entry_id(session_key, segment_id, log_index, 0)),
|
||||
timestamp,
|
||||
provenance: SessionEntryProvenance::LegacyUnknown,
|
||||
derived_from: Vec::new(),
|
||||
data,
|
||||
}
|
||||
}
|
||||
|
||||
fn legacy_entry(
|
||||
session_key: &SessionId,
|
||||
segment_id: Option<&SegmentId>,
|
||||
@@ -2812,22 +2834,344 @@ mod tests {
|
||||
));
|
||||
}
|
||||
|
||||
fn run_entry_ids(entries: &[SessionSnapshotEntry]) -> Vec<String> {
|
||||
entries
|
||||
.iter()
|
||||
.filter(|entry| {
|
||||
matches!(
|
||||
entry.data,
|
||||
SessionSnapshotEntryData::RunYielded { .. }
|
||||
| SessionSnapshotEntryData::RunResumed { .. }
|
||||
| SessionSnapshotEntryData::RunCancelled
|
||||
| SessionSnapshotEntryData::RunError { .. }
|
||||
)
|
||||
})
|
||||
.map(|entry| entry.entry_id.clone())
|
||||
.collect()
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn saved_run_transition_ids_match_snapshot_retained_and_paged_history() {
|
||||
let worker_name = "worker-run-identities";
|
||||
let session_id = crate::new_session_id();
|
||||
let segment_id = crate::new_segment_id();
|
||||
let initial = vec![
|
||||
segment_start(session_id, Vec::new(), None, None),
|
||||
LogEntry::AnnotatedUserInput {
|
||||
ts: 2,
|
||||
segments: vec![Segment::text("compact")],
|
||||
history: vec![user_message(1)],
|
||||
extensions: Vec::new(),
|
||||
},
|
||||
];
|
||||
let (_root, aggregate_root) = persist_history_fixture(
|
||||
worker_name,
|
||||
session_id,
|
||||
segment_id,
|
||||
vec![(segment_id, initial)],
|
||||
);
|
||||
let store = WorkerSessionStore::new(aggregate_root.join("session")).unwrap();
|
||||
let transitions = [
|
||||
LogEntry::RunYielded {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunResumed {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "mid-run compaction failed: unavailable".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "mid-run compaction failed: unavailable".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
},
|
||||
LogEntry::RunCancelled {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
},
|
||||
];
|
||||
for entry in transitions {
|
||||
crate::segment::append_entry(&store, session_id, segment_id, entry).unwrap();
|
||||
}
|
||||
|
||||
let stored = store.read_all(session_id, segment_id).unwrap();
|
||||
let stored_ids = run_entry_ids(
|
||||
&project_session_snapshot_for_segment(session_id, Some(segment_id), &stored).entries,
|
||||
);
|
||||
assert_eq!(stored_ids.len(), 5);
|
||||
assert_eq!(stored_ids.iter().collect::<HashSet<_>>().len(), 5);
|
||||
assert!(
|
||||
stored_ids
|
||||
.iter()
|
||||
.all(|entry_id| !entry_id.starts_with("l-run-"))
|
||||
);
|
||||
|
||||
let retained = read_retained_session_snapshot(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
DEFAULT_RETAINED_SNAPSHOT_MAX_BYTES,
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&retained.snapshot.entries), stored_ids);
|
||||
|
||||
let page = read_retained_session_history_page(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
None,
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(page.turns.len(), 1);
|
||||
assert_eq!(run_entry_ids(&page.turns[0].entries), stored_ids);
|
||||
|
||||
let reconnect = read_retained_session_snapshot(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
DEFAULT_RETAINED_SNAPSHOT_MAX_BYTES,
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&reconnect.snapshot.entries), stored_ids);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn run_transition_ids_survive_segment_lineage_and_history_page_boundaries() {
|
||||
let worker_name = "worker-run-lineage-identities";
|
||||
let session_id = crate::new_session_id();
|
||||
let source_segment = crate::new_segment_id();
|
||||
let active_segment = crate::new_segment_id();
|
||||
let source = vec![
|
||||
segment_start(session_id, Vec::new(), None, None),
|
||||
LogEntry::AnnotatedUserInput {
|
||||
ts: 2,
|
||||
segments: vec![Segment::text("source turn")],
|
||||
history: vec![user_message(1)],
|
||||
extensions: Vec::new(),
|
||||
},
|
||||
];
|
||||
let (_root, aggregate_root) = persist_history_fixture(
|
||||
worker_name,
|
||||
session_id,
|
||||
source_segment,
|
||||
vec![(source_segment, source)],
|
||||
);
|
||||
let store = WorkerSessionStore::new(aggregate_root.join("session")).unwrap();
|
||||
crate::segment::append_entry(
|
||||
&store,
|
||||
session_id,
|
||||
source_segment,
|
||||
LogEntry::RunYielded {
|
||||
ts: 3,
|
||||
entry_id: None,
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
},
|
||||
)
|
||||
.unwrap();
|
||||
store
|
||||
.append(
|
||||
session_id,
|
||||
source_segment,
|
||||
&LogEntry::TurnEnd {
|
||||
ts: 4,
|
||||
turn_count: 1,
|
||||
},
|
||||
)
|
||||
.unwrap();
|
||||
store
|
||||
.create_segment(
|
||||
session_id,
|
||||
active_segment,
|
||||
&[
|
||||
segment_start(
|
||||
session_id,
|
||||
vec![user_message(1)],
|
||||
None,
|
||||
Some(SegmentOrigin {
|
||||
segment_id: source_segment,
|
||||
at_turn_index: 1,
|
||||
}),
|
||||
),
|
||||
LogEntry::AnnotatedUserInput {
|
||||
ts: 5,
|
||||
segments: vec![Segment::text("active turn")],
|
||||
history: vec![user_message(2)],
|
||||
extensions: Vec::new(),
|
||||
},
|
||||
],
|
||||
)
|
||||
.unwrap();
|
||||
for entry in [
|
||||
LogEntry::RunResumed {
|
||||
ts: 6,
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 7,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "mid-run compaction failed".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
},
|
||||
] {
|
||||
crate::segment::append_entry(&store, session_id, active_segment, entry).unwrap();
|
||||
}
|
||||
WorkerAggregateStore::new(&aggregate_root, worker_name)
|
||||
.unwrap()
|
||||
.write(&crate::WorkerMetadata::new(
|
||||
worker_name,
|
||||
Some(crate::WorkerActiveSegmentRef::active_segment(
|
||||
session_id,
|
||||
active_segment,
|
||||
)),
|
||||
))
|
||||
.unwrap();
|
||||
|
||||
let active_ids = run_entry_ids(
|
||||
&project_session_snapshot_for_segment(
|
||||
session_id,
|
||||
Some(active_segment),
|
||||
&store.read_all(session_id, active_segment).unwrap(),
|
||||
)
|
||||
.entries,
|
||||
);
|
||||
let source_ids = run_entry_ids(
|
||||
&project_session_snapshot_for_segment(
|
||||
session_id,
|
||||
Some(source_segment),
|
||||
&store.read_all(session_id, source_segment).unwrap(),
|
||||
)
|
||||
.entries,
|
||||
);
|
||||
assert_eq!(active_ids.len(), 2);
|
||||
assert_eq!(source_ids.len(), 1);
|
||||
|
||||
let newest = read_retained_session_history_page(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
None,
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&newest.turns[0].entries), active_ids);
|
||||
assert!(newest.has_more);
|
||||
let older = read_retained_session_history_page(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
newest.next_cursor.as_deref(),
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&older.turns[0].entries), source_ids);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn legacy_run_transition_ids_use_stable_distinct_record_positions() {
|
||||
let worker_name = "worker-legacy-run-identities";
|
||||
let session_id = crate::new_session_id();
|
||||
let segment_id = crate::new_segment_id();
|
||||
let log = vec![
|
||||
segment_start(session_id, Vec::new(), None, None),
|
||||
LogEntry::AnnotatedUserInput {
|
||||
ts: 2,
|
||||
segments: vec![Segment::text("legacy")],
|
||||
history: vec![user_message(1)],
|
||||
extensions: Vec::new(),
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "same legacy failure".into(),
|
||||
failure: None,
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "same legacy failure".into(),
|
||||
failure: None,
|
||||
},
|
||||
];
|
||||
let (_root, aggregate_root) =
|
||||
persist_history_fixture(worker_name, session_id, segment_id, vec![(segment_id, log)]);
|
||||
let log_path = aggregate_root
|
||||
.join("session")
|
||||
.join("segments")
|
||||
.join(format!("{segment_id}.jsonl"));
|
||||
let persisted_before = std::fs::read(&log_path).unwrap();
|
||||
assert!(!String::from_utf8_lossy(&persisted_before).contains("entry_id\":\"l-run-"));
|
||||
|
||||
let retained = read_retained_session_snapshot(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
DEFAULT_RETAINED_SNAPSHOT_MAX_BYTES,
|
||||
)
|
||||
.unwrap();
|
||||
let snapshot_ids = run_entry_ids(&retained.snapshot.entries);
|
||||
assert_eq!(snapshot_ids.len(), 2);
|
||||
assert_ne!(snapshot_ids[0], snapshot_ids[1]);
|
||||
assert!(
|
||||
snapshot_ids
|
||||
.iter()
|
||||
.all(|entry_id| entry_id.starts_with("l-run-"))
|
||||
);
|
||||
|
||||
let page = read_retained_session_history_page(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
None,
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&page.turns[0].entries), snapshot_ids);
|
||||
let reread = read_retained_session_snapshot(
|
||||
&aggregate_root,
|
||||
worker_name,
|
||||
DEFAULT_RETAINED_SNAPSHOT_MAX_BYTES,
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(run_entry_ids(&reread.snapshot.entries), snapshot_ids);
|
||||
assert_eq!(std::fs::read(log_path).unwrap(), persisted_before);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn run_transitions_project_with_timestamps_and_typed_causes() {
|
||||
let session_id = crate::new_session_id();
|
||||
let log = vec![
|
||||
LogEntry::RunYielded {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunResumed {
|
||||
ts: 20,
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 30,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "mid-run compaction failed: unavailable".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
@@ -2869,7 +3213,10 @@ mod tests {
|
||||
ts: 10,
|
||||
trigger: protocol::InvokeKind::UserSend,
|
||||
},
|
||||
LogEntry::RunCancelled { ts: 20 },
|
||||
LogEntry::RunCancelled {
|
||||
ts: 20,
|
||||
entry_id: None,
|
||||
},
|
||||
],
|
||||
);
|
||||
|
||||
|
||||
@@ -347,6 +347,7 @@ pub fn save_run_yielded(
|
||||
segment_id,
|
||||
LogEntry::RunYielded {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
reason,
|
||||
active_run_turn_count,
|
||||
},
|
||||
@@ -367,6 +368,7 @@ pub fn save_run_resumed(
|
||||
segment_id,
|
||||
LogEntry::RunResumed {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
source,
|
||||
active_run_turn_count,
|
||||
},
|
||||
@@ -390,6 +392,7 @@ pub fn save_run_errored(
|
||||
segment_id,
|
||||
LogEntry::RunErrored {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
interrupted,
|
||||
message,
|
||||
failure: Some(RunFailureKind::Engine),
|
||||
@@ -568,7 +571,8 @@ pub fn append_entry(
|
||||
store: &impl Store,
|
||||
session_id: SessionId,
|
||||
segment_id: SegmentId,
|
||||
entry: LogEntry,
|
||||
mut entry: LogEntry,
|
||||
) -> Result<(), StoreError> {
|
||||
entry.ensure_session_entry_id();
|
||||
store.append(session_id, segment_id, &entry)
|
||||
}
|
||||
|
||||
@@ -140,6 +140,12 @@ pub enum LogEntry {
|
||||
/// preceding Invoke and a later RunResumed or terminal record.
|
||||
RunYielded {
|
||||
ts: u64,
|
||||
/// Stable public identity allocated before the record is committed.
|
||||
/// `None` is accepted only when reading logs written before this field
|
||||
/// existed; read paths derive the same legacy identity from the durable
|
||||
/// record position without rewriting the log.
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
entry_id: Option<crate::LoggedSessionHistoryEntryId>,
|
||||
reason: RunYieldReason,
|
||||
active_run_turn_count: usize,
|
||||
},
|
||||
@@ -149,19 +155,27 @@ pub enum LogEntry {
|
||||
/// It deliberately does not assert that a subsequent model call occurred.
|
||||
RunResumed {
|
||||
ts: u64,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
entry_id: Option<crate::LoggedSessionHistoryEntryId>,
|
||||
source: RunResumeSource,
|
||||
active_run_turn_count: usize,
|
||||
},
|
||||
|
||||
/// The logical Run ended because the user intentionally cancelled it.
|
||||
/// This is terminal, but is not an execution failure.
|
||||
RunCancelled { ts: u64 },
|
||||
RunCancelled {
|
||||
ts: u64,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
entry_id: Option<crate::LoggedSessionHistoryEntryId>,
|
||||
},
|
||||
|
||||
/// `run()` / `resume()` が `EngineError` で終了した。
|
||||
/// `EngineError` は `Serialize` 不可なので `message` のみ lossy 保持する。
|
||||
/// Audit-only metadata: replay は `interrupted` のみ反映する。
|
||||
RunErrored {
|
||||
ts: u64,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
entry_id: Option<crate::LoggedSessionHistoryEntryId>,
|
||||
interrupted: bool,
|
||||
message: String,
|
||||
/// Older records omit this field. New records distinguish failures in
|
||||
@@ -236,6 +250,48 @@ pub struct SegmentOrigin {
|
||||
pub at_turn_index: usize,
|
||||
}
|
||||
|
||||
impl LogEntry {
|
||||
/// Allocate the durable public identity for a newly committed Run
|
||||
/// transition. Calling this repeatedly is idempotent so the exact value
|
||||
/// written to storage is also the value published to live subscribers.
|
||||
pub fn ensure_session_entry_id(&mut self) {
|
||||
let entry_id = match self {
|
||||
Self::RunYielded { entry_id, .. }
|
||||
| Self::RunResumed { entry_id, .. }
|
||||
| Self::RunCancelled { entry_id, .. }
|
||||
| Self::RunErrored { entry_id, .. } => entry_id,
|
||||
_ => return,
|
||||
};
|
||||
if entry_id.is_none() {
|
||||
*entry_id = Some(crate::LoggedSessionHistoryEntryId::new());
|
||||
}
|
||||
}
|
||||
|
||||
/// Supply an in-memory compatibility identity for a legacy Run transition
|
||||
/// that predates persisted entry IDs. The physical record position is
|
||||
/// stable within the append-only Segment and distinguishes equal payloads.
|
||||
/// This helper never writes the synthesized value back to the log.
|
||||
pub(crate) fn ensure_legacy_session_entry_id(
|
||||
&mut self,
|
||||
session_id: crate::SessionId,
|
||||
segment_id: crate::SegmentId,
|
||||
record_start: u64,
|
||||
) {
|
||||
let entry_id = match self {
|
||||
Self::RunYielded { entry_id, .. }
|
||||
| Self::RunResumed { entry_id, .. }
|
||||
| Self::RunCancelled { entry_id, .. }
|
||||
| Self::RunErrored { entry_id, .. } => entry_id,
|
||||
_ => return,
|
||||
};
|
||||
if entry_id.is_none() {
|
||||
*entry_id = Some(crate::LoggedSessionHistoryEntryId(format!(
|
||||
"l-run-{session_id}-{segment_id}-{record_start}"
|
||||
)));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// State collected from log entries.
|
||||
#[derive(Debug, Clone)]
|
||||
pub struct RestoredState {
|
||||
@@ -878,6 +934,7 @@ mod tests {
|
||||
},
|
||||
LogEntry::RunYielded {
|
||||
ts: 300,
|
||||
entry_id: None,
|
||||
reason: RunYieldReason::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
@@ -888,6 +945,7 @@ mod tests {
|
||||
},
|
||||
LogEntry::RunResumed {
|
||||
ts: 500,
|
||||
entry_id: None,
|
||||
source: RunResumeSource::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
@@ -909,7 +967,10 @@ mod tests {
|
||||
ts: 200,
|
||||
turn_count: 1,
|
||||
},
|
||||
LogEntry::RunCancelled { ts: 300 },
|
||||
LogEntry::RunCancelled {
|
||||
ts: 300,
|
||||
entry_id: None,
|
||||
},
|
||||
]);
|
||||
|
||||
assert!(!state.last_run_interrupted);
|
||||
|
||||
@@ -61,6 +61,8 @@ pub(crate) struct RetainedLogRecord {
|
||||
|
||||
pub(crate) struct RetainedSegmentReader {
|
||||
file: File,
|
||||
session_id: SessionId,
|
||||
segment_id: SegmentId,
|
||||
read_end: u64,
|
||||
buffer_start: u64,
|
||||
buffer: Vec<u8>,
|
||||
@@ -103,7 +105,12 @@ impl RetainedSegmentReader {
|
||||
message: "empty retained Segment log record".to_string(),
|
||||
});
|
||||
}
|
||||
let entry = serde_json::from_slice(&record)?;
|
||||
let mut entry: LogEntry = serde_json::from_slice(&record)?;
|
||||
entry.ensure_legacy_session_entry_id(
|
||||
self.session_id,
|
||||
self.segment_id,
|
||||
record_start,
|
||||
);
|
||||
return Ok((
|
||||
Some(RetainedLogRecord {
|
||||
start_offset: record_start,
|
||||
@@ -255,7 +262,10 @@ impl WorkerSessionStore {
|
||||
if byte_len > max_bytes {
|
||||
return Err(StoreError::ReadLimitExceeded);
|
||||
}
|
||||
Ok((parse_jsonl(&bytes)?, byte_len))
|
||||
Ok((
|
||||
parse_session_log_jsonl(&bytes, session_id, segment_id)?,
|
||||
byte_len,
|
||||
))
|
||||
}
|
||||
|
||||
pub(crate) fn read_first_log_record_read_only_bounded(
|
||||
@@ -264,6 +274,7 @@ impl WorkerSessionStore {
|
||||
segment_id: SegmentId,
|
||||
max_bytes: u64,
|
||||
) -> Result<(Option<RetainedLogRecord>, u64, u64), StoreError> {
|
||||
let record_start = 0;
|
||||
self.validate_retained_session(session_id)?;
|
||||
let file = open_retained_file(&self.log_path(segment_id)).map_err(|error| {
|
||||
if error.kind() == std::io::ErrorKind::NotFound {
|
||||
@@ -289,7 +300,8 @@ impl WorkerSessionStore {
|
||||
message: "empty retained Segment log record".to_string(),
|
||||
});
|
||||
}
|
||||
let entry = serde_json::from_slice(&record)?;
|
||||
let mut entry: LogEntry = serde_json::from_slice(&record)?;
|
||||
entry.ensure_legacy_session_entry_id(session_id, segment_id, record_start);
|
||||
Ok((
|
||||
Some(RetainedLogRecord {
|
||||
start_offset: 0,
|
||||
@@ -308,6 +320,7 @@ impl WorkerSessionStore {
|
||||
start_offset: u64,
|
||||
max_bytes: u64,
|
||||
) -> Result<(Option<RetainedLogRecord>, u64), StoreError> {
|
||||
let record_start = start_offset;
|
||||
self.validate_retained_session(session_id)?;
|
||||
let mut file = open_retained_file(&self.log_path(segment_id)).map_err(|error| {
|
||||
if error.kind() == std::io::ErrorKind::NotFound {
|
||||
@@ -334,7 +347,8 @@ impl WorkerSessionStore {
|
||||
message: "empty retained Segment log record".to_string(),
|
||||
});
|
||||
}
|
||||
let entry = serde_json::from_slice(&record)?;
|
||||
let mut entry: LogEntry = serde_json::from_slice(&record)?;
|
||||
entry.ensure_legacy_session_entry_id(session_id, segment_id, record_start);
|
||||
Ok((
|
||||
Some(RetainedLogRecord {
|
||||
start_offset,
|
||||
@@ -353,6 +367,7 @@ impl WorkerSessionStore {
|
||||
end_offset: u64,
|
||||
max_bytes: u64,
|
||||
) -> Result<RetainedLogRecord, StoreError> {
|
||||
let record_start = start_offset;
|
||||
self.validate_retained_session(session_id)?;
|
||||
let record_len =
|
||||
end_offset
|
||||
@@ -387,7 +402,8 @@ impl WorkerSessionStore {
|
||||
});
|
||||
}
|
||||
record.pop();
|
||||
let entry = serde_json::from_slice(&record)?;
|
||||
let mut entry: LogEntry = serde_json::from_slice(&record)?;
|
||||
entry.ensure_legacy_session_entry_id(session_id, segment_id, record_start);
|
||||
Ok(RetainedLogRecord {
|
||||
start_offset,
|
||||
end_offset,
|
||||
@@ -429,6 +445,8 @@ impl WorkerSessionStore {
|
||||
}
|
||||
Ok(RetainedSegmentReader {
|
||||
file,
|
||||
session_id,
|
||||
segment_id,
|
||||
read_end: before_offset,
|
||||
buffer_start: before_offset,
|
||||
buffer: Vec::new(),
|
||||
@@ -645,7 +663,7 @@ impl Store for WorkerSessionStore {
|
||||
if !path.exists() {
|
||||
return Err(StoreError::NotFound(segment_id));
|
||||
}
|
||||
parse_jsonl(&fs::read(path)?)
|
||||
parse_session_log_jsonl(&fs::read(path)?, session_id, segment_id)
|
||||
}
|
||||
|
||||
fn list_sessions(&self) -> Result<Vec<SessionId>, StoreError> {
|
||||
@@ -1154,6 +1172,38 @@ fn complete_jsonl_prefix(content: &[u8]) -> &[u8] {
|
||||
.unwrap_or(&[])
|
||||
}
|
||||
|
||||
fn parse_session_log_jsonl(
|
||||
content: &[u8],
|
||||
session_id: SessionId,
|
||||
segment_id: SegmentId,
|
||||
) -> Result<Vec<LogEntry>, StoreError> {
|
||||
let complete = complete_jsonl_prefix(content);
|
||||
let content = std::str::from_utf8(complete).map_err(|error| StoreError::Corrupt {
|
||||
line: complete[..error.valid_up_to()]
|
||||
.iter()
|
||||
.filter(|byte| **byte == b'\n')
|
||||
.count()
|
||||
+ 1,
|
||||
message: error.to_string(),
|
||||
})?;
|
||||
let mut entries = Vec::new();
|
||||
let mut record_start = 0_u64;
|
||||
for (index, record) in content.split_inclusive('\n').enumerate() {
|
||||
let line = record.strip_suffix('\n').unwrap_or(record);
|
||||
if !line.trim().is_empty() {
|
||||
let mut entry: LogEntry =
|
||||
serde_json::from_str(line).map_err(|error| StoreError::Corrupt {
|
||||
line: index + 1,
|
||||
message: error.to_string(),
|
||||
})?;
|
||||
entry.ensure_legacy_session_entry_id(session_id, segment_id, record_start);
|
||||
entries.push(entry);
|
||||
}
|
||||
record_start = record_start.saturating_add(record.len() as u64);
|
||||
}
|
||||
Ok(entries)
|
||||
}
|
||||
|
||||
fn parse_jsonl<T: serde::de::DeserializeOwned>(content: &[u8]) -> Result<Vec<T>, StoreError> {
|
||||
let complete = complete_jsonl_prefix(content);
|
||||
let content = std::str::from_utf8(complete).map_err(|error| StoreError::Corrupt {
|
||||
|
||||
@@ -3800,6 +3800,7 @@ mod completion_flow_tests {
|
||||
|
||||
let run_errored = session_store::LogEntry::RunErrored {
|
||||
ts: 3,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "provider unavailable".into(),
|
||||
failure: None,
|
||||
|
||||
@@ -117,11 +117,128 @@ pub async fn dispatch_worker_protocol_method(
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
use session_store::Store;
|
||||
|
||||
use crate::segment_log_sink::SegmentLogSink;
|
||||
use crate::worker::{LogWriterHandle, SegmentState};
|
||||
|
||||
fn committed_entry(event: Event) -> protocol::SessionSnapshotEntry {
|
||||
match event {
|
||||
Event::SessionEntryCommitted { entry } => entry,
|
||||
other => panic!("expected SessionEntryCommitted, got {other:?}"),
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn persisted_run_transitions_keep_identity_across_live_and_snapshot_projection() {
|
||||
let temp = tempfile::tempdir().unwrap();
|
||||
let store = session_store::FsStore::new(temp.path()).unwrap();
|
||||
let session_id = session_store::new_session_id();
|
||||
let segment_id = session_store::new_segment_id();
|
||||
let start = LogEntry::AnnotatedSegmentStart {
|
||||
ts: 1,
|
||||
session_id,
|
||||
system_prompt: None,
|
||||
config: agen::llm_client::RequestConfig::default(),
|
||||
history: Vec::new(),
|
||||
forked_from: None,
|
||||
compacted_from: None,
|
||||
};
|
||||
store
|
||||
.create_segment(session_id, segment_id, std::slice::from_ref(&start))
|
||||
.unwrap();
|
||||
let sink = SegmentLogSink::with_initial(vec![start]);
|
||||
let writer = LogWriterHandle {
|
||||
store: store.clone(),
|
||||
state: SegmentState::new(session_id, segment_id, 1),
|
||||
sink: sink.clone(),
|
||||
in_flight: None,
|
||||
};
|
||||
let (_, mut receiver) = sink.subscribe_with_snapshot();
|
||||
let transitions = [
|
||||
LogEntry::RunYielded {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunResumed {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 2,
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "same compaction failure".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
},
|
||||
LogEntry::RunErrored {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: "same compaction failure".into(),
|
||||
failure: Some(protocol::RunFailureKind::Compaction),
|
||||
},
|
||||
LogEntry::RunCancelled {
|
||||
ts: 10,
|
||||
entry_id: None,
|
||||
},
|
||||
];
|
||||
|
||||
let mut live_ids = Vec::new();
|
||||
for transition in transitions {
|
||||
writer.append_entry(transition).unwrap();
|
||||
let committed = receiver.try_recv().expect("committed transition broadcast");
|
||||
live_ids.push(
|
||||
committed_entry(live_log_entry_event(committed).expect("live projection")).entry_id,
|
||||
);
|
||||
}
|
||||
assert_eq!(
|
||||
live_ids
|
||||
.iter()
|
||||
.collect::<std::collections::HashSet<_>>()
|
||||
.len(),
|
||||
5
|
||||
);
|
||||
|
||||
let (mirror, _) = sink.subscribe_with_snapshot();
|
||||
let snapshot = session_store::public_snapshot::project_current_session_snapshot(&mirror);
|
||||
let snapshot_ids: Vec<_> = snapshot
|
||||
.entries
|
||||
.iter()
|
||||
.filter(|entry| {
|
||||
matches!(
|
||||
entry.data,
|
||||
protocol::SessionSnapshotEntryData::RunYielded { .. }
|
||||
| protocol::SessionSnapshotEntryData::RunResumed { .. }
|
||||
| protocol::SessionSnapshotEntryData::RunError { .. }
|
||||
| protocol::SessionSnapshotEntryData::RunCancelled
|
||||
)
|
||||
})
|
||||
.map(|entry| entry.entry_id.clone())
|
||||
.collect();
|
||||
assert_eq!(snapshot_ids, live_ids);
|
||||
|
||||
let persisted = store.read_all(session_id, segment_id).unwrap();
|
||||
let persisted_snapshot =
|
||||
session_store::public_snapshot::project_current_session_snapshot(&persisted);
|
||||
let persisted_ids: Vec<_> = persisted_snapshot
|
||||
.entries
|
||||
.iter()
|
||||
.filter(|entry| snapshot_ids.contains(&entry.entry_id))
|
||||
.map(|entry| entry.entry_id.clone())
|
||||
.collect();
|
||||
assert_eq!(persisted_ids, live_ids);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn durable_run_transition_maps_to_session_entry_committed() {
|
||||
let event = live_log_entry_event(LogEntry::RunYielded {
|
||||
ts: 42,
|
||||
entry_id: Some(session_store::LoggedSessionHistoryEntryId::new()),
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 3,
|
||||
})
|
||||
|
||||
@@ -330,6 +330,7 @@ mod tests {
|
||||
// Durable logical-Run transitions share the committed-entry lane.
|
||||
sink.publish(LogEntry::RunResumed {
|
||||
ts: now_millis(),
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
});
|
||||
|
||||
@@ -1426,7 +1426,8 @@ where
|
||||
/// Append `entry` to the log: disk write → counter bump → in-memory
|
||||
/// mirror push → broadcast. The Store owns physical write ordering and
|
||||
/// partial-write recovery; publication happens only after it returns Ok.
|
||||
pub fn append_entry(&self, entry: LogEntry) -> Result<(), StoreError> {
|
||||
pub fn append_entry(&self, mut entry: LogEntry) -> Result<(), StoreError> {
|
||||
entry.ensure_session_entry_id();
|
||||
let _append_guard = self
|
||||
.state
|
||||
.append_lock
|
||||
@@ -3055,7 +3056,8 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
/// Append `entry` to the session log AND publish it through the
|
||||
/// broadcast sink. The Store is the commit boundary: a failed write is
|
||||
/// never counted or published.
|
||||
pub(crate) fn commit_entry(&self, entry: LogEntry) -> Result<(), StoreError> {
|
||||
pub(crate) fn commit_entry(&self, mut entry: LogEntry) -> Result<(), StoreError> {
|
||||
entry.ensure_session_entry_id();
|
||||
let loc = self.segment_state.location();
|
||||
self.store.append(loc.session_id, loc.segment_id, &entry)?;
|
||||
self.segment_state.increment_entries();
|
||||
@@ -4707,6 +4709,7 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
.expect("pause/yield resume must retain the logical-run turn budget");
|
||||
self.commit_entry(LogEntry::RunResumed {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
source,
|
||||
active_run_turn_count,
|
||||
})?;
|
||||
@@ -5024,6 +5027,7 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
// marked as already delivered to clients through the session-log sink.
|
||||
self.commit_entry(LogEntry::RunErrored {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message,
|
||||
failure: Some(RunFailureKind::Compaction),
|
||||
@@ -5435,6 +5439,7 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
{
|
||||
self.commit_entry(LogEntry::RunErrored {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
interrupted: false,
|
||||
message: message.to_string(),
|
||||
failure: Some(RunFailureKind::Compaction),
|
||||
@@ -5458,6 +5463,7 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
EngineRunExit::Yielded => {
|
||||
self.commit_entry(LogEntry::RunYielded {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
reason: RunYieldReason::Compaction,
|
||||
active_run_turn_count: active_run_turn_count
|
||||
.expect("yielded logical run must retain its turn budget"),
|
||||
@@ -5474,11 +5480,13 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
|
||||
EngineRunExit::Interrupted(RunInterruptionReason::Cancelled) => {
|
||||
self.commit_entry(LogEntry::RunCancelled {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
})?;
|
||||
}
|
||||
EngineRunExit::Interrupted(reason) => {
|
||||
self.commit_entry(LogEntry::RunErrored {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
interrupted,
|
||||
message: run_interruption_reason_message(reason),
|
||||
failure: Some(RunFailureKind::Engine),
|
||||
@@ -11442,6 +11450,7 @@ mod build_summary_prompt_tests {
|
||||
worker
|
||||
.commit_entry(LogEntry::RunErrored {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
interrupted: true,
|
||||
message: "provider failed".to_string(),
|
||||
failure: Some(RunFailureKind::Engine),
|
||||
@@ -11472,6 +11481,7 @@ mod build_summary_prompt_tests {
|
||||
worker
|
||||
.commit_entry(LogEntry::RunCancelled {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
})
|
||||
.unwrap();
|
||||
assert_eq!(
|
||||
@@ -11482,6 +11492,7 @@ mod build_summary_prompt_tests {
|
||||
worker
|
||||
.commit_entry(LogEntry::RunResumed {
|
||||
ts: segment_log::now_millis(),
|
||||
entry_id: None,
|
||||
source: RunResumeSource::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user