fix: restore console run accounting from persisted logs
This commit is contained in:
@@ -1054,6 +1054,23 @@ pub enum SessionSnapshotEntryData {
|
||||
source: RunResumeSource,
|
||||
active_run_turn_count: usize,
|
||||
},
|
||||
/// Snapshot-only replay of persisted invocation/usage records. These are
|
||||
/// accounting metadata, never model context or conversation rows.
|
||||
Invoke {
|
||||
trigger: InvokeKind,
|
||||
},
|
||||
Usage {
|
||||
input_tokens: u64,
|
||||
cache_read_input_tokens: u64,
|
||||
output_tokens: u64,
|
||||
},
|
||||
/// Display-only accounting reconstructed from persisted run/usage records.
|
||||
RunStats {
|
||||
elapsed_ms: u64,
|
||||
requests: u64,
|
||||
upload_tokens: u64,
|
||||
output_tokens: u64,
|
||||
},
|
||||
/// Durable terminal for an intentional cancellation. Consumers must not
|
||||
/// present this lifecycle result as an execution failure.
|
||||
RunCancelled,
|
||||
@@ -1177,6 +1194,9 @@ pub enum Event {
|
||||
/// following `UserMessage` / `SystemItem` event.
|
||||
InvokeStart {
|
||||
kind: InvokeKind,
|
||||
/// Timestamp of the persisted Invoke, not the client's attach time.
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
timestamp_ms: Option<u64>,
|
||||
},
|
||||
/// One AgentTurn boundary opened. An AgentTurn is a maximal run of
|
||||
/// LLM generation calls whose input messages are identical (i.e.
|
||||
@@ -1293,6 +1313,9 @@ pub enum Event {
|
||||
/// actually paid full price to send on this request, which is what
|
||||
/// the TUI status line accumulates per turn.
|
||||
Usage {
|
||||
/// Present for usage delivered from the committed session-log lane.
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
timestamp_ms: Option<u64>,
|
||||
input_tokens: Option<u64>,
|
||||
output_tokens: Option<u64>,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
@@ -2335,19 +2358,29 @@ mod tests {
|
||||
InvokeKind::SystemReminder,
|
||||
InvokeKind::Wakeup,
|
||||
] {
|
||||
let event = Event::InvokeStart { kind };
|
||||
let event = Event::InvokeStart {
|
||||
kind,
|
||||
timestamp_ms: Some(1234),
|
||||
};
|
||||
let json = serde_json::to_string(&event).unwrap();
|
||||
let parsed: serde_json::Value = serde_json::from_str(&json).unwrap();
|
||||
assert_eq!(parsed["event"], "invoke_start");
|
||||
let decoded: Event = serde_json::from_str(&json).unwrap();
|
||||
match decoded {
|
||||
Event::InvokeStart { kind: k } => assert_eq!(k, kind),
|
||||
Event::InvokeStart {
|
||||
kind: k,
|
||||
timestamp_ms,
|
||||
} => {
|
||||
assert_eq!(k, kind);
|
||||
assert_eq!(timestamp_ms, Some(1234));
|
||||
}
|
||||
other => panic!("expected InvokeStart, got {other:?}"),
|
||||
}
|
||||
}
|
||||
let parsed: serde_json::Value = serde_json::from_str(
|
||||
&serde_json::to_string(&Event::InvokeStart {
|
||||
kind: InvokeKind::UserSend,
|
||||
timestamp_ms: None,
|
||||
})
|
||||
.unwrap(),
|
||||
)
|
||||
|
||||
@@ -1,3 +1,5 @@
|
||||
mod run_stats;
|
||||
|
||||
use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD};
|
||||
use protocol::{
|
||||
Segment, SessionContentPart, SessionConversationTurn, SessionEntryProvenance,
|
||||
@@ -850,6 +852,17 @@ fn read_history_turns_backward(
|
||||
let mut pending_entries = Vec::new();
|
||||
let mut seen_entries = HashSet::new();
|
||||
let mut corrections = HashMap::<String, SessionSnapshotEntryData>::new();
|
||||
// Accounting needs the whole Invoke range, including the marker immediately
|
||||
// before the oldest user entry on this page. The existing scan budgets also
|
||||
// bound this suffix; no unbounded second pass over the Session is needed.
|
||||
let mut accounting_records = Vec::new();
|
||||
let finish = |mut turns: Vec<PositionedHistoryTurn>, records: &[LogEntry]| {
|
||||
let stats = run_stats::project(records.iter().rev());
|
||||
for turn in &mut turns {
|
||||
run_stats::append_to(&mut turn.turn.entries, &stats);
|
||||
}
|
||||
turns
|
||||
};
|
||||
|
||||
// Carry only bounded log positions, never correction text or image bodies,
|
||||
// across pages. Re-read and validate the referenced records from adopted
|
||||
@@ -994,11 +1007,58 @@ fn read_history_turns_backward(
|
||||
&mut pending_entries,
|
||||
&mut turns,
|
||||
) {
|
||||
return Ok(turns);
|
||||
return Ok(finish(turns, &accounting_records));
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
// TurnEnd is the lineage boundary, but Worker persists final usage and
|
||||
// the run outcome immediately after it. For a compacted ancestor, read
|
||||
// only this accounting tail, never adopt later conversation records.
|
||||
if before_offset == adopted_end
|
||||
&& segment_index > 0
|
||||
&& matches!(
|
||||
lineage[segment_index - 1].origin.as_ref().map(|o| o.kind),
|
||||
Some(LineageOriginKind::Compact)
|
||||
)
|
||||
{
|
||||
let mut offset = adopted_end;
|
||||
let mut tail = Vec::new();
|
||||
while offset < segment.file_len {
|
||||
let (record, bytes) = store
|
||||
.read_next_log_record_read_only_bounded(
|
||||
session_id,
|
||||
segment.segment_id,
|
||||
offset,
|
||||
limits.max_scan_bytes.saturating_sub(*scanned_bytes),
|
||||
)
|
||||
.map_err(map_history_store_error)?;
|
||||
*scanned_bytes = scanned_bytes
|
||||
.checked_add(bytes)
|
||||
.ok_or(RetainedHistoryReadError::ResourceLimit)?;
|
||||
let Some(record) = record else { break };
|
||||
*scanned_entries = scanned_entries
|
||||
.checked_add(persisted_entry_units(std::slice::from_ref(&record.entry)))
|
||||
.ok_or(RetainedHistoryReadError::ResourceLimit)?;
|
||||
if *scanned_entries > limits.max_entries {
|
||||
return Err(RetainedHistoryReadError::ResourceLimit);
|
||||
}
|
||||
offset = record.end_offset;
|
||||
match record.entry {
|
||||
entry @ LogEntry::LlmUsage { .. } => tail.push(entry),
|
||||
LogEntry::Extension { .. } => {}
|
||||
entry @ (LogEntry::RunCompleted { .. }
|
||||
| LogEntry::RunYielded { .. }
|
||||
| LogEntry::RunCancelled { .. }
|
||||
| LogEntry::RunErrored { .. }) => {
|
||||
tail.push(entry);
|
||||
break;
|
||||
}
|
||||
_ => break,
|
||||
}
|
||||
}
|
||||
accounting_records.extend(tail.into_iter().rev());
|
||||
}
|
||||
let mut reader = store
|
||||
.open_retained_segment_reader(session_id, segment.segment_id, before_offset)
|
||||
.map_err(map_history_store_error)?;
|
||||
@@ -1025,6 +1085,16 @@ fn read_history_turns_backward(
|
||||
}
|
||||
continue;
|
||||
}
|
||||
let at_invoke = matches!(record.entry, LogEntry::Invoke { .. });
|
||||
accounting_records.push(record.entry.clone());
|
||||
if turns.len() >= turn_limit {
|
||||
if at_invoke {
|
||||
return Ok(finish(turns, &accounting_records));
|
||||
}
|
||||
// Older records here supply accounting only, never additional
|
||||
// conversation turns beyond the requested page boundary.
|
||||
continue;
|
||||
}
|
||||
if capture_tool_correction(&record.entry, &mut corrections) {
|
||||
let LogEntry::ToolResultCorrected { entry, .. } = &record.entry else {
|
||||
unreachable!()
|
||||
@@ -1044,7 +1114,7 @@ fn read_history_turns_backward(
|
||||
&project_history_record(session_id, segment.segment_id, &record.entry),
|
||||
&corrections,
|
||||
);
|
||||
if collect_history_entries_backward(
|
||||
collect_history_entries_backward(
|
||||
&entries,
|
||||
segment.segment_id,
|
||||
record.start_offset,
|
||||
@@ -1054,12 +1124,11 @@ fn read_history_turns_backward(
|
||||
&mut seen_entries,
|
||||
&mut pending_entries,
|
||||
&mut turns,
|
||||
) {
|
||||
return Ok(turns);
|
||||
}
|
||||
);
|
||||
}
|
||||
|
||||
if segment_index + 1 == lineage.len()
|
||||
if turns.len() < turn_limit
|
||||
&& segment_index + 1 == lineage.len()
|
||||
&& collect_history_entries_backward(
|
||||
&corrected_history_entries(&segment.seed_entries, &corrections),
|
||||
segment.segment_id,
|
||||
@@ -1072,10 +1141,10 @@ fn read_history_turns_backward(
|
||||
&mut turns,
|
||||
)
|
||||
{
|
||||
return Ok(turns);
|
||||
return Ok(finish(turns, &accounting_records));
|
||||
}
|
||||
}
|
||||
Ok(turns)
|
||||
Ok(finish(turns, &accounting_records))
|
||||
}
|
||||
|
||||
fn capture_tool_correction(
|
||||
@@ -1322,6 +1391,148 @@ pub fn project_current_session_snapshot(log: &[LogEntry]) -> SessionSnapshot {
|
||||
project_session_snapshot(session_id.unwrap_or_else(SessionId::nil), log)
|
||||
}
|
||||
|
||||
fn run_accounting_entries(session_id: &SessionId, log: &[LogEntry]) -> Vec<SessionSnapshotEntry> {
|
||||
let start = log
|
||||
.iter()
|
||||
.rposition(|entry| matches!(entry, LogEntry::Invoke { .. }))
|
||||
.unwrap_or(0);
|
||||
log.iter()
|
||||
.enumerate()
|
||||
.skip(start)
|
||||
.filter_map(|(index, record)| {
|
||||
let (ts, data) = match record {
|
||||
LogEntry::Invoke { ts, trigger } => {
|
||||
(*ts, SessionSnapshotEntryData::Invoke { trigger: *trigger })
|
||||
}
|
||||
LogEntry::LlmUsage {
|
||||
ts,
|
||||
input_total_tokens,
|
||||
cache_read_tokens,
|
||||
output_tokens,
|
||||
..
|
||||
} => (
|
||||
*ts,
|
||||
SessionSnapshotEntryData::Usage {
|
||||
input_tokens: *input_total_tokens,
|
||||
cache_read_input_tokens: *cache_read_tokens,
|
||||
output_tokens: *output_tokens,
|
||||
},
|
||||
),
|
||||
_ => return None,
|
||||
};
|
||||
Some(legacy_entry(session_id, None, index, 0, ts, data))
|
||||
})
|
||||
.collect()
|
||||
}
|
||||
|
||||
/// Reconstruct snapshot accounting from existing compacted ancestors, without a
|
||||
/// second accumulator or checkpoint. The live suffix is the caller's frozen
|
||||
/// mirror prefix; never reread the active segment after subscribing.
|
||||
pub fn project_current_session_snapshot_with_accounting(
|
||||
store: &dyn crate::Store,
|
||||
log: &[LogEntry],
|
||||
) -> SessionSnapshot {
|
||||
let mut snapshot = project_current_session_snapshot(log);
|
||||
let session_id = log
|
||||
.iter()
|
||||
.find_map(|entry| match entry {
|
||||
LogEntry::AnnotatedSegmentStart { session_id, .. } => Some(*session_id),
|
||||
_ => None,
|
||||
})
|
||||
.unwrap_or_else(SessionId::nil);
|
||||
snapshot
|
||||
.entries
|
||||
.extend(run_accounting_entries(&session_id, log));
|
||||
if log
|
||||
.iter()
|
||||
.any(|entry| matches!(entry, LogEntry::Invoke { .. }))
|
||||
{
|
||||
return snapshot;
|
||||
}
|
||||
let mut source = log.to_vec();
|
||||
let mut ancestors = Vec::new();
|
||||
let mut visited = HashSet::new();
|
||||
let mut count = log.len();
|
||||
for _ in 0..DEFAULT_RETAINED_HISTORY_MAX_SEGMENTS {
|
||||
let Some((session_id, origin)) = source.iter().find_map(|entry| match entry {
|
||||
LogEntry::AnnotatedSegmentStart {
|
||||
session_id,
|
||||
compacted_from: Some(origin),
|
||||
..
|
||||
} => Some((*session_id, origin.clone())),
|
||||
_ => None,
|
||||
}) else {
|
||||
break;
|
||||
};
|
||||
if !visited.insert(origin.segment_id) {
|
||||
break;
|
||||
}
|
||||
let Ok(mut entries) = store.read_all(session_id, origin.segment_id) else {
|
||||
break;
|
||||
};
|
||||
count = count.saturating_add(entries.len());
|
||||
if count > DEFAULT_RETAINED_HISTORY_MAX_ENTRIES {
|
||||
break;
|
||||
}
|
||||
// Compaction adopts through TurnEnd plus its trailing usage/outcome,
|
||||
// never through a subsequent Invoke or unrelated conversation.
|
||||
if let Some(end) = entries.iter().position(|entry| {
|
||||
matches!(entry, LogEntry::TurnEnd { turn_count, .. } if *turn_count == origin.at_turn_index)
|
||||
}) {
|
||||
let mut cutoff = end + 1;
|
||||
while cutoff < entries.len()
|
||||
&& matches!(entries[cutoff],
|
||||
LogEntry::LlmUsage { .. } | LogEntry::Extension { .. }
|
||||
| LogEntry::RunCompleted { .. } | LogEntry::RunYielded { .. }
|
||||
| LogEntry::RunErrored { .. } | LogEntry::RunCancelled { .. })
|
||||
{
|
||||
cutoff += 1;
|
||||
}
|
||||
entries.truncate(cutoff);
|
||||
} else if matches!(entries.first(), Some(LogEntry::AnnotatedSegmentStart {
|
||||
compacted_from: Some(inherited), ..
|
||||
}) if origin.at_turn_index <= inherited.at_turn_index) {
|
||||
// Consecutive compactions can adopt only the inherited seed.
|
||||
entries.truncate(1);
|
||||
} else {
|
||||
break;
|
||||
}
|
||||
let complete = entries
|
||||
.iter()
|
||||
.any(|entry| matches!(entry, LogEntry::Invoke { .. }));
|
||||
source = entries.clone();
|
||||
ancestors.push(entries);
|
||||
if complete {
|
||||
break;
|
||||
}
|
||||
}
|
||||
if !ancestors.is_empty() {
|
||||
let combined: Vec<LogEntry> = ancestors
|
||||
.into_iter()
|
||||
.rev()
|
||||
.flatten()
|
||||
.chain(log.iter().cloned())
|
||||
.collect();
|
||||
snapshot.entries.retain(|entry| {
|
||||
!matches!(
|
||||
entry.data,
|
||||
SessionSnapshotEntryData::Invoke { .. } | SessionSnapshotEntryData::Usage { .. }
|
||||
)
|
||||
});
|
||||
let session_id = log
|
||||
.iter()
|
||||
.find_map(|entry| match entry {
|
||||
LogEntry::AnnotatedSegmentStart { session_id, .. } => Some(*session_id),
|
||||
_ => None,
|
||||
})
|
||||
.unwrap_or_else(SessionId::nil);
|
||||
snapshot
|
||||
.entries
|
||||
.extend(run_accounting_entries(&session_id, &combined));
|
||||
}
|
||||
snapshot
|
||||
}
|
||||
|
||||
/// Project the current durable segment into the only public session-history
|
||||
/// representation. Append-log records remain an internal persistence format.
|
||||
pub fn project_session_snapshot(session_id: SessionId, log: &[LogEntry]) -> SessionSnapshot {
|
||||
@@ -1523,6 +1734,7 @@ pub(crate) fn project_session_snapshot_for_segment(
|
||||
}
|
||||
}
|
||||
|
||||
run_stats::append_to(&mut entries, &run_stats::project(log));
|
||||
SessionSnapshot {
|
||||
pending_submissions: protocol::PendingSubmissionsSnapshot::default(),
|
||||
entries,
|
||||
@@ -1800,6 +2012,258 @@ mod tests {
|
||||
LoggedWorkerSubject, Store, WorkerMetadataStore,
|
||||
};
|
||||
|
||||
fn usage(ts: u64) -> LogEntry {
|
||||
LogEntry::LlmUsage {
|
||||
ts,
|
||||
history_len: 2,
|
||||
input_total_tokens: 2500,
|
||||
cache_read_tokens: 1000,
|
||||
cache_write_tokens: 0,
|
||||
output_tokens: 200,
|
||||
}
|
||||
}
|
||||
|
||||
fn complete(ts: u64) -> LogEntry {
|
||||
LogEntry::RunCompleted {
|
||||
ts,
|
||||
interrupted: false,
|
||||
result: agen::EngineResult::Finished,
|
||||
active_run_turn_count: None,
|
||||
}
|
||||
}
|
||||
|
||||
fn stats_entries(entries: &[SessionSnapshotEntry]) -> Vec<SessionSnapshotEntry> {
|
||||
entries
|
||||
.iter()
|
||||
.filter(|entry| matches!(entry.data, SessionSnapshotEntryData::RunStats { .. }))
|
||||
.cloned()
|
||||
.collect()
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn run_stats_restore_identically_from_snapshot_and_history_pages() {
|
||||
let session_id = SessionId::now_v7();
|
||||
let segment_id = SegmentId::now_v7();
|
||||
let mut log = vec![segment_start(session_id, vec![], None, None)];
|
||||
for turn in 1..=6 {
|
||||
append_turn(&mut log, turn);
|
||||
log.push(usage(turn as u64 * 10 + 4));
|
||||
log.push(complete(turn as u64 * 10 + 8));
|
||||
}
|
||||
let expected = stats_entries(&project_current_session_snapshot(&log).entries);
|
||||
assert_eq!(expected.len(), 6);
|
||||
assert_eq!(
|
||||
expected[0].data,
|
||||
SessionSnapshotEntryData::RunStats {
|
||||
elapsed_ms: 8,
|
||||
requests: 1,
|
||||
upload_tokens: 1500,
|
||||
output_tokens: 200,
|
||||
}
|
||||
);
|
||||
let (_root, path) =
|
||||
persist_history_fixture("stats", session_id, segment_id, vec![(segment_id, log)]);
|
||||
let mut cursor = None;
|
||||
let mut actual = Vec::new();
|
||||
loop {
|
||||
let page = read_retained_session_history_page(
|
||||
&path,
|
||||
"stats",
|
||||
cursor.as_deref(),
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(page.turns.len(), 1);
|
||||
actual.splice(0..0, stats_entries(&page.turns[0].entries));
|
||||
cursor = page.next_cursor;
|
||||
if !page.has_more {
|
||||
break;
|
||||
}
|
||||
}
|
||||
assert_eq!(actual, expected);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn run_stats_span_compaction_without_counting_seed_history_twice() {
|
||||
let session_id = SessionId::now_v7();
|
||||
let source_id = SegmentId::now_v7();
|
||||
let active_id = SegmentId::now_v7();
|
||||
let mut source = vec![segment_start(session_id, vec![], None, None)];
|
||||
append_turn(&mut source, 1);
|
||||
source.push(usage(15));
|
||||
source.push(LogEntry::RunYielded {
|
||||
ts: 16,
|
||||
entry_id: None,
|
||||
reason: protocol::RunYieldReason::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
});
|
||||
let active = vec![
|
||||
segment_start(
|
||||
session_id,
|
||||
vec![user_message(1), assistant_message(1, "final")],
|
||||
None,
|
||||
Some(SegmentOrigin {
|
||||
segment_id: source_id,
|
||||
at_turn_index: 1,
|
||||
}),
|
||||
),
|
||||
LogEntry::RunResumed {
|
||||
ts: 20,
|
||||
entry_id: None,
|
||||
source: protocol::RunResumeSource::Compaction,
|
||||
active_run_turn_count: 1,
|
||||
},
|
||||
usage(25),
|
||||
LogEntry::AnnotatedAssistantItem {
|
||||
ts: 28,
|
||||
entry: assistant_message(1, "after-compact"),
|
||||
},
|
||||
complete(30),
|
||||
];
|
||||
assert!(
|
||||
stats_entries(&project_current_session_snapshot(&active).entries).is_empty(),
|
||||
"a partial segment must not invent the missing start or traffic"
|
||||
);
|
||||
let (_root, path) = persist_history_fixture(
|
||||
"stats",
|
||||
session_id,
|
||||
active_id,
|
||||
vec![(source_id, source), (active_id, active)],
|
||||
);
|
||||
let page = read_retained_session_history_page(
|
||||
&path,
|
||||
"stats",
|
||||
None,
|
||||
Some(1),
|
||||
RetainedHistoryReadLimits::default(),
|
||||
)
|
||||
.unwrap();
|
||||
let stats = stats_entries(&page.turns[0].entries);
|
||||
assert_eq!(stats.len(), 1);
|
||||
assert_eq!(stats[0].entry_id, "run-stats:assistant-1-after-compact");
|
||||
assert_eq!(
|
||||
stats[0].data,
|
||||
SessionSnapshotEntryData::RunStats {
|
||||
elapsed_ms: 20,
|
||||
requests: 2,
|
||||
upload_tokens: 3000,
|
||||
output_tokens: 400,
|
||||
}
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn reconnect_accounting_replays_existing_usage_across_compaction() {
|
||||
let session_id = SessionId::now_v7();
|
||||
let source_id = SegmentId::now_v7();
|
||||
let active_id = SegmentId::now_v7();
|
||||
let mut source = vec![segment_start(session_id, vec![], None, None)];
|
||||
append_turn(&mut source, 1);
|
||||
source.push(usage(15));
|
||||
let active = vec![
|
||||
segment_start(
|
||||
session_id,
|
||||
vec![],
|
||||
None,
|
||||
Some(SegmentOrigin {
|
||||
segment_id: source_id,
|
||||
at_turn_index: 1,
|
||||
}),
|
||||
),
|
||||
usage(25),
|
||||
];
|
||||
let (_root, path) = persist_history_fixture(
|
||||
"accounting",
|
||||
session_id,
|
||||
active_id,
|
||||
vec![(source_id, source), (active_id, active.clone())],
|
||||
);
|
||||
let store = WorkerSessionStore::new(path.join("session")).unwrap();
|
||||
let snapshot = project_current_session_snapshot_with_accounting(&store, &active);
|
||||
let records: Vec<_> = snapshot
|
||||
.entries
|
||||
.iter()
|
||||
.filter(|entry| {
|
||||
matches!(
|
||||
entry.data,
|
||||
SessionSnapshotEntryData::Invoke { .. }
|
||||
| SessionSnapshotEntryData::Usage { .. }
|
||||
)
|
||||
})
|
||||
.collect();
|
||||
assert_eq!(records.len(), 3);
|
||||
assert_eq!(
|
||||
records
|
||||
.iter()
|
||||
.map(|entry| entry.timestamp)
|
||||
.collect::<Vec<_>>(),
|
||||
[10, 15, 25]
|
||||
);
|
||||
assert!(matches!(
|
||||
records[0].data,
|
||||
SessionSnapshotEntryData::Invoke {
|
||||
trigger: InvokeKind::UserSend
|
||||
}
|
||||
));
|
||||
assert!(matches!(
|
||||
records[1].data,
|
||||
SessionSnapshotEntryData::Usage {
|
||||
input_tokens: 2500,
|
||||
cache_read_input_tokens: 1000,
|
||||
output_tokens: 200
|
||||
}
|
||||
));
|
||||
assert!(matches!(
|
||||
records[2].data,
|
||||
SessionSnapshotEntryData::Usage {
|
||||
input_tokens: 2500,
|
||||
cache_read_input_tokens: 1000,
|
||||
output_tokens: 200
|
||||
}
|
||||
));
|
||||
|
||||
// A second compaction can adopt only the first one's inherited seed.
|
||||
// Its unadopted usage must not leak into the restored totals.
|
||||
let next = vec![
|
||||
segment_start(
|
||||
session_id,
|
||||
vec![],
|
||||
None,
|
||||
Some(SegmentOrigin {
|
||||
segment_id: active_id,
|
||||
at_turn_index: 1,
|
||||
}),
|
||||
),
|
||||
usage(35),
|
||||
];
|
||||
let snapshot = project_current_session_snapshot_with_accounting(&store, &next);
|
||||
let timestamps: Vec<_> = snapshot
|
||||
.entries
|
||||
.iter()
|
||||
.filter_map(|entry| {
|
||||
matches!(
|
||||
entry.data,
|
||||
SessionSnapshotEntryData::Invoke { .. }
|
||||
| SessionSnapshotEntryData::Usage { .. }
|
||||
)
|
||||
.then_some(entry.timestamp)
|
||||
})
|
||||
.collect();
|
||||
assert_eq!(timestamps, [10, 15, 35]);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn run_stats_do_not_complete_an_active_or_unmeasured_run() {
|
||||
let mut log = Vec::new();
|
||||
append_turn(&mut log, 1);
|
||||
log.push(usage(15));
|
||||
assert!(stats_entries(&project_current_session_snapshot(&log).entries).is_empty());
|
||||
log.pop();
|
||||
log.push(complete(20));
|
||||
assert!(stats_entries(&project_current_session_snapshot(&log).entries).is_empty());
|
||||
}
|
||||
|
||||
fn history_message(
|
||||
entry_id: impl Into<String>,
|
||||
role: LoggedRole,
|
||||
|
||||
@@ -0,0 +1,133 @@
|
||||
//! Display-only accounting; never inserted into model history or written back
|
||||
//! to the journal. Both snapshot and bounded history paging use this projection.
|
||||
use std::collections::HashMap;
|
||||
|
||||
use protocol::{SessionEntryProvenance, SessionSnapshotEntry, SessionSnapshotEntryData};
|
||||
|
||||
use crate::LogEntry;
|
||||
|
||||
#[derive(Default)]
|
||||
struct Activity {
|
||||
started_at: u64,
|
||||
requests: u64,
|
||||
upload: u64,
|
||||
output: u64,
|
||||
anchor: Option<String>,
|
||||
}
|
||||
|
||||
pub(super) fn project<'a>(
|
||||
records: impl IntoIterator<Item = &'a LogEntry>,
|
||||
) -> HashMap<String, SessionSnapshotEntry> {
|
||||
let mut active: Option<Activity> = None;
|
||||
let mut stats = HashMap::new();
|
||||
for record in records {
|
||||
match record {
|
||||
LogEntry::Invoke { ts, .. } => {
|
||||
active = Some(Activity {
|
||||
started_at: *ts,
|
||||
..Activity::default()
|
||||
})
|
||||
}
|
||||
LogEntry::LlmUsage {
|
||||
input_total_tokens,
|
||||
cache_read_tokens,
|
||||
output_tokens,
|
||||
..
|
||||
} => {
|
||||
if let Some(run) = &mut active {
|
||||
run.requests += 1;
|
||||
run.upload = run
|
||||
.upload
|
||||
.saturating_add(input_total_tokens.saturating_sub(*cache_read_tokens));
|
||||
run.output = run.output.saturating_add(*output_tokens);
|
||||
}
|
||||
}
|
||||
LogEntry::AnnotatedUserInput { history, .. } => {
|
||||
if let Some(run) = &mut active {
|
||||
for entry in history {
|
||||
if super::project_item(&entry.metadata.entry_id.0, &entry.item).is_some() {
|
||||
run.anchor = Some(entry.metadata.entry_id.0.clone());
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
LogEntry::AnnotatedAssistantItem { entry, .. }
|
||||
| LogEntry::AnnotatedToolResult { entry, .. } => {
|
||||
if let Some(run) = &mut active
|
||||
&& super::project_item(&entry.metadata.entry_id.0, &entry.item).is_some()
|
||||
{
|
||||
run.anchor = Some(entry.metadata.entry_id.0.clone());
|
||||
}
|
||||
}
|
||||
LogEntry::AnnotatedSystemItem { entry, .. } => {
|
||||
if let Some(run) = &mut active {
|
||||
run.anchor = Some(entry.metadata.entry_id.0.clone());
|
||||
}
|
||||
}
|
||||
LogEntry::RunCompleted { ts, result, .. }
|
||||
if !matches!(result, agen::EngineResult::Yielded) =>
|
||||
{
|
||||
finish(&active, *ts, &mut stats);
|
||||
// Pause/resume retains the Invoke's cumulative accounting, just
|
||||
// like the live TUI. Terminal outcomes cannot leak traffic into
|
||||
// a later resumed segment whose Invoke is unavailable.
|
||||
if !matches!(result, agen::EngineResult::Paused) {
|
||||
active = None;
|
||||
}
|
||||
}
|
||||
LogEntry::RunCancelled { ts, .. } | LogEntry::RunErrored { ts, .. } => {
|
||||
finish(&active, *ts, &mut stats);
|
||||
active = None;
|
||||
}
|
||||
// Compaction seeds repeat retained entries, not fresh traffic.
|
||||
// A page traverses adopted lineage in journal order and retains the
|
||||
// open Invoke across these checkpoints. A standalone new segment
|
||||
// without that Invoke cannot claim complete accounting.
|
||||
_ => {}
|
||||
}
|
||||
}
|
||||
stats
|
||||
}
|
||||
|
||||
fn finish(
|
||||
active: &Option<Activity>,
|
||||
ended_at: u64,
|
||||
stats: &mut HashMap<String, SessionSnapshotEntry>,
|
||||
) {
|
||||
let Some(run) = active else { return };
|
||||
let Some(anchor) = &run.anchor else { return };
|
||||
// No usage records means unavailable measurement, not known zero traffic.
|
||||
if run.requests == 0 {
|
||||
return;
|
||||
}
|
||||
stats.insert(
|
||||
anchor.clone(),
|
||||
SessionSnapshotEntry {
|
||||
entry_id: format!("run-stats:{anchor}"),
|
||||
timestamp: ended_at,
|
||||
provenance: SessionEntryProvenance::LegacyUnknown,
|
||||
derived_from: vec![anchor.clone()],
|
||||
data: SessionSnapshotEntryData::RunStats {
|
||||
elapsed_ms: ended_at.saturating_sub(run.started_at),
|
||||
requests: run.requests,
|
||||
upload_tokens: run.upload,
|
||||
output_tokens: run.output,
|
||||
},
|
||||
},
|
||||
);
|
||||
}
|
||||
|
||||
pub(super) fn append_to(
|
||||
entries: &mut Vec<SessionSnapshotEntry>,
|
||||
stats: &HashMap<String, SessionSnapshotEntry>,
|
||||
) {
|
||||
let mut with_stats = Vec::with_capacity(entries.len());
|
||||
for entry in std::mem::take(entries) {
|
||||
let summary = stats.get(&entry.entry_id);
|
||||
with_stats.push(entry);
|
||||
if let Some(summary) = summary {
|
||||
with_stats.push(summary.clone());
|
||||
}
|
||||
}
|
||||
*entries = with_stats;
|
||||
}
|
||||
@@ -1741,6 +1741,7 @@ impl App {
|
||||
input_tokens,
|
||||
output_tokens,
|
||||
cache_read_input_tokens,
|
||||
..
|
||||
} => {
|
||||
if let Some(input_tokens) = input_tokens {
|
||||
self.session_context_tokens = input_tokens;
|
||||
@@ -3000,7 +3001,10 @@ impl App {
|
||||
self.apply_system_item(data);
|
||||
}
|
||||
}
|
||||
SessionSnapshotEntryData::RunYielded { .. }
|
||||
SessionSnapshotEntryData::Invoke { .. }
|
||||
| SessionSnapshotEntryData::Usage { .. }
|
||||
| SessionSnapshotEntryData::RunStats { .. }
|
||||
| SessionSnapshotEntryData::RunYielded { .. }
|
||||
| SessionSnapshotEntryData::RunResumed { .. }
|
||||
| SessionSnapshotEntryData::RunCancelled => {
|
||||
// Durable logical-Run transitions are intentionally not
|
||||
@@ -3981,6 +3985,7 @@ mod completion_flow_tests {
|
||||
});
|
||||
app.handle_worker_event(Event::TurnStart { turn: 1 });
|
||||
app.handle_worker_event(Event::Usage {
|
||||
timestamp_ms: None,
|
||||
input_tokens: Some(100),
|
||||
output_tokens: Some(0),
|
||||
cache_read_input_tokens: Some(40),
|
||||
@@ -4210,6 +4215,7 @@ mod completion_flow_tests {
|
||||
let mut app = App::new("test".into());
|
||||
app.handle_worker_event(Event::TurnStart { turn: 1 });
|
||||
app.handle_worker_event(Event::InvokeStart {
|
||||
timestamp_ms: None,
|
||||
kind: protocol::InvokeKind::UserSend,
|
||||
});
|
||||
app.handle_worker_event(Event::RunEnd {
|
||||
@@ -5156,6 +5162,7 @@ mod completion_flow_tests {
|
||||
let mut app = App::new("test".into());
|
||||
|
||||
app.handle_worker_event(Event::Usage {
|
||||
timestamp_ms: None,
|
||||
input_tokens: Some(42_000),
|
||||
output_tokens: Some(9),
|
||||
cache_read_input_tokens: Some(40_000),
|
||||
|
||||
@@ -83,6 +83,7 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
|
||||
input_tokens,
|
||||
output_tokens,
|
||||
cache_read_input_tokens,
|
||||
..
|
||||
} => {
|
||||
println!(
|
||||
"[usage] in={} (cache_read={}) out={}",
|
||||
|
||||
@@ -178,7 +178,10 @@ impl WorkerHandle {
|
||||
(entries, entry_rx, in_flight)
|
||||
};
|
||||
let mut session =
|
||||
session_store::public_snapshot::project_current_session_snapshot(&entries);
|
||||
session_store::public_snapshot::project_current_session_snapshot_with_accounting(
|
||||
self.artifact_store.as_ref(),
|
||||
&entries,
|
||||
);
|
||||
session.pending_submissions = self
|
||||
.pending_activations
|
||||
.lock()
|
||||
@@ -1286,14 +1289,8 @@ pub(crate) fn wire_event_bridges_on_engine<C, St>(
|
||||
});
|
||||
});
|
||||
|
||||
let tx = working_event_tx.clone();
|
||||
worker.on_usage(move |event| {
|
||||
let _ = tx.send(Event::Usage {
|
||||
input_tokens: event.input_tokens,
|
||||
output_tokens: event.output_tokens,
|
||||
cache_read_input_tokens: event.cache_read_input_tokens,
|
||||
});
|
||||
});
|
||||
// Usage is published from committed LlmUsage records through the session-log
|
||||
// lane. Snapshot and live usage therefore share one exact, gap-free boundary.
|
||||
|
||||
let tx = working_event_tx.clone();
|
||||
worker.on_error(move |event| {
|
||||
|
||||
@@ -494,7 +494,11 @@ impl InternalWorkerSessionHandle {
|
||||
(entries, snapshot_from_guard(&guard))
|
||||
};
|
||||
InternalWorkerSessionSnapshot {
|
||||
session: session_store::public_snapshot::project_current_session_snapshot(&entries),
|
||||
session:
|
||||
session_store::public_snapshot::project_current_session_snapshot_with_accounting(
|
||||
&self.store,
|
||||
&entries,
|
||||
),
|
||||
greeting: self.greeting.as_ref().map(|greeting| {
|
||||
greeting
|
||||
.read()
|
||||
|
||||
@@ -96,7 +96,22 @@ pub fn live_log_entry_event(entry: LogEntry) -> Option<Event> {
|
||||
item: value,
|
||||
})
|
||||
}
|
||||
LogEntry::Invoke { trigger, .. } => Some(Event::InvokeStart { kind: trigger }),
|
||||
LogEntry::Invoke { ts, trigger } => Some(Event::InvokeStart {
|
||||
kind: trigger,
|
||||
timestamp_ms: Some(ts),
|
||||
}),
|
||||
LogEntry::LlmUsage {
|
||||
ts,
|
||||
input_total_tokens,
|
||||
cache_read_tokens,
|
||||
output_tokens,
|
||||
..
|
||||
} => Some(Event::Usage {
|
||||
timestamp_ms: Some(ts),
|
||||
input_tokens: Some(input_total_tokens),
|
||||
cache_read_input_tokens: Some(cache_read_tokens),
|
||||
output_tokens: Some(output_tokens),
|
||||
}),
|
||||
other => {
|
||||
// `SegmentLogSink::is_live_relevant` keeps non-live-relevant
|
||||
// variants off the broadcast lane; reaching here means the two are
|
||||
|
||||
@@ -124,6 +124,7 @@ impl SegmentLogSink {
|
||||
| LogEntry::ToolResultCorrected { .. }
|
||||
| LogEntry::AnnotatedSystemItem { .. }
|
||||
| LogEntry::Invoke { .. }
|
||||
| LogEntry::LlmUsage { .. }
|
||||
| LogEntry::RunYielded { .. }
|
||||
| LogEntry::RunResumed { .. }
|
||||
| LogEntry::RunCancelled { .. }
|
||||
@@ -357,6 +358,54 @@ mod tests {
|
||||
assert!(rx.try_recv().is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn persisted_usage_is_split_once_between_snapshot_and_live_even_with_equal_timestamps() {
|
||||
let sink = SegmentLogSink::new();
|
||||
sink.publish(LogEntry::Invoke {
|
||||
ts: 100,
|
||||
trigger: protocol::InvokeKind::UserSend,
|
||||
});
|
||||
let usage = LogEntry::LlmUsage {
|
||||
ts: 200,
|
||||
history_len: 2,
|
||||
input_total_tokens: 2000,
|
||||
cache_read_tokens: 1000,
|
||||
cache_write_tokens: 0,
|
||||
output_tokens: 100,
|
||||
};
|
||||
sink.publish(usage.clone());
|
||||
let (snapshot, mut rx) = sink.subscribe_with_snapshot();
|
||||
assert_eq!(
|
||||
snapshot
|
||||
.iter()
|
||||
.filter(|e| matches!(e, LogEntry::LlmUsage { .. }))
|
||||
.count(),
|
||||
1
|
||||
);
|
||||
assert!(rx.try_recv().is_err());
|
||||
sink.publish(usage);
|
||||
let event =
|
||||
crate::ipc::protocol_session::live_log_entry_event(rx.try_recv().unwrap()).unwrap();
|
||||
assert!(matches!(
|
||||
event,
|
||||
protocol::Event::Usage {
|
||||
timestamp_ms: Some(200),
|
||||
input_tokens: Some(2000),
|
||||
..
|
||||
}
|
||||
));
|
||||
assert!(rx.try_recv().is_err());
|
||||
let (reconnected, mut next_rx) = sink.subscribe_with_snapshot();
|
||||
assert_eq!(
|
||||
reconnected
|
||||
.iter()
|
||||
.filter(|e| matches!(e, LogEntry::LlmUsage { .. }))
|
||||
.count(),
|
||||
2
|
||||
);
|
||||
assert!(next_rx.try_recv().is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn reset_with_initial_clears_and_broadcasts() {
|
||||
let sink = SegmentLogSink::new();
|
||||
|
||||
@@ -294,6 +294,9 @@ impl SessionCapture {
|
||||
// Observation deliberately excludes system items and
|
||||
// controller errors from model-visible session evidence.
|
||||
SessionSnapshotEntryData::SystemItem { .. }
|
||||
| SessionSnapshotEntryData::Invoke { .. }
|
||||
| SessionSnapshotEntryData::Usage { .. }
|
||||
| SessionSnapshotEntryData::RunStats { .. }
|
||||
| SessionSnapshotEntryData::RunYielded { .. }
|
||||
| SessionSnapshotEntryData::RunResumed { .. }
|
||||
| SessionSnapshotEntryData::RunCancelled
|
||||
|
||||
@@ -1556,6 +1556,70 @@ async fn snapshot_includes_user_input_for_in_flight_turn() {
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn reconnect_snapshot_restores_live_run_accounting_from_engine_events() {
|
||||
let client = MockClient::new(vec![
|
||||
LlmEvent::tool_use_start(0, "hold-call", "Hold"),
|
||||
LlmEvent::tool_input_delta(0, "{}"),
|
||||
LlmEvent::tool_use_stop(0),
|
||||
LlmEvent::Usage(UsageEvent {
|
||||
input_tokens: Some(25_000),
|
||||
output_tokens: Some(300),
|
||||
cache_read_input_tokens: Some(20_000),
|
||||
..Default::default()
|
||||
}),
|
||||
LlmEvent::Status(StatusEvent {
|
||||
status: ResponseStatus::Completed,
|
||||
}),
|
||||
]);
|
||||
let mut worker = make_worker(client).await;
|
||||
worker
|
||||
.engine_mut()
|
||||
.register_tool(hanging_tool_definition("Hold"));
|
||||
let handle = spawn_controller(worker).await;
|
||||
let mut events = handle.sink.subscribe_with_snapshot().1;
|
||||
handle
|
||||
.send(Method::submit_text(
|
||||
protocol::new_submission_request_id(),
|
||||
"keep running",
|
||||
))
|
||||
.await
|
||||
.unwrap();
|
||||
let observed = tokio::time::timeout(std::time::Duration::from_secs(2), async {
|
||||
loop {
|
||||
if let LogEntry::LlmUsage {
|
||||
ts,
|
||||
input_total_tokens,
|
||||
cache_read_tokens,
|
||||
output_tokens,
|
||||
..
|
||||
} = events.recv().await.unwrap()
|
||||
{
|
||||
break (ts, input_total_tokens, cache_read_tokens, output_tokens);
|
||||
}
|
||||
}
|
||||
})
|
||||
.await
|
||||
.expect("live accounting");
|
||||
assert_eq!((observed.1, observed.2, observed.3), (25_000, 20_000, 300));
|
||||
for _ in 0..2 {
|
||||
let Event::Snapshot { session, .. } = handle.snapshot_event() else {
|
||||
panic!("expected snapshot")
|
||||
};
|
||||
let usage: Vec<_> = session
|
||||
.entries
|
||||
.iter()
|
||||
.filter(|e| matches!(e.data, protocol::SessionSnapshotEntryData::Usage { .. }))
|
||||
.collect();
|
||||
assert_eq!(usage.len(), 1);
|
||||
assert_eq!(usage[0].timestamp, observed.0);
|
||||
assert!(session.entries.iter().any(|e| matches!(
|
||||
e.data,
|
||||
protocol::SessionSnapshotEntryData::Invoke { .. }
|
||||
) && e.timestamp > 0));
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn attach_snapshot_includes_current_status() {
|
||||
let client = MockClient::sequential(vec![MockResponse::Hang(simple_text_events())]);
|
||||
|
||||
Reference in New Issue
Block a user