Merge remote-tracking branch 'origin/pr/592'

2026-09-14 23:20:19 +08:00 · 2026-05-28 16:07:07 +08:00
parent 5c3a1aecbe b108dc5ea6
commit 3504875922
26 changed files with 1034 additions and 92 deletions
@@ -9,9 +9,9 @@ use aether_admin::observability::usage::{
    admin_usage_data_unavailable_response, admin_usage_has_fallback, admin_usage_is_failed,
    admin_usage_matches_search, admin_usage_matches_username, admin_usage_parse_ids,
    admin_usage_parse_limit, admin_usage_parse_offset, admin_usage_provider_key_name,
-    admin_usage_record_json, build_admin_usage_active_requests_response,
+    admin_usage_record_json, attach_usage_server_now_header,
-    build_admin_usage_records_response, build_admin_usage_summary_stats_response_from_summary,
+    build_admin_usage_active_requests_response, build_admin_usage_records_response,
-    ADMIN_USAGE_DATA_UNAVAILABLE_DETAIL,
+    build_admin_usage_summary_stats_response_from_summary, ADMIN_USAGE_DATA_UNAVAILABLE_DETAIL,
 };
 use aether_data::repository::users::StoredUserSummary;
 use aether_data_contracts::repository::{
@@ -314,13 +314,15 @@ fn build_admin_usage_records_response_with_attempt_flags(
        })
        .collect();
    attach_usage_server_now_header(
        Json(json!({
            "records": records,
            "total": total,
            "limit": limit,
            "offset": offset,
        }))
-    .into_response()
+        .into_response(),
    )
 }
 fn build_admin_usage_records_query(
@@ -4,6 +4,7 @@ use aether_ai_serving::UPSTREAM_IS_STREAM_KEY;
 use aether_billing::{
    normalize_input_tokens_for_billing, normalize_total_input_context_for_cache_hit_rate,
 };
 use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
 use aether_data_contracts::repository::usage::{
    StoredRequestUsageAudit, StoredUsageBreakdownSummaryRow, StoredUsageDailySummary,
    UsageAuditKeywordSearchQuery, UsageAuditListQuery, UsageBreakdownGroupBy,
@@ -29,6 +30,21 @@ use super::{
 const USERS_ME_USAGE_DATA_UNAVAILABLE_DETAIL: &str = "用户用量数据暂不可用";
 fn users_me_usage_server_now_unix_ms() -> u64 {
    u64::try_from(Utc::now().timestamp_millis()).unwrap_or_default()
 }
 fn attach_users_me_usage_server_now_header(mut response: Response<Body>) -> Response<Body> {
    if let Ok(value) = http::HeaderValue::from_str(&users_me_usage_server_now_unix_ms().to_string())
    {
        response.headers_mut().insert(
            http::HeaderName::from_static(USAGE_SERVER_NOW_UNIX_MS_HEADER),
            value,
        );
    }
    response
 }
 fn build_users_me_usage_reader_unavailable_response() -> Response<Body> {
    build_auth_error_response(
        http::StatusCode::SERVICE_UNAVAILABLE,
@@ -1094,7 +1110,7 @@ pub(super) async fn handle_users_me_usage_get(
            &summary_by_provider
        ));
    }
-    Json(payload).into_response()
+    attach_users_me_usage_server_now_header(Json(payload).into_response())
 }
 pub(super) async fn handle_users_me_usage_active_get(
@@ -1169,13 +1185,15 @@ pub(super) async fn handle_users_me_usage_active_get(
            .collect::<Vec<_>>()
    };
    attach_users_me_usage_server_now_header(
        Json(json!({
            "requests": items
                .iter()
                .map(build_users_me_usage_active_payload)
                .collect::<Vec<_>>(),
        }))
-    .into_response()
+        .into_response(),
    )
 }
 pub(super) async fn handle_users_me_usage_interval_timeline_get(
@@ -1364,12 +1382,20 @@ async fn build_usage_heatmap_summaries(
 mod tests {
    use std::collections::BTreeMap;
    use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
    use aether_data_contracts::repository::usage::StoredRequestUsageAudit;
    use axum::{
        body::Body,
        response::{IntoResponse, Response},
        Json,
    };
    use chrono::Utc;
    use serde_json::json;
    use super::{
-        build_users_me_usage_active_payload, build_users_me_usage_record_payload,
+        attach_users_me_usage_server_now_header, build_users_me_usage_active_payload,
-        users_me_usage_client_is_stream, users_me_usage_is_failed,
+        build_users_me_usage_record_payload, users_me_usage_client_is_stream,
        users_me_usage_is_failed, users_me_usage_server_now_unix_ms,
        users_me_usage_upstream_is_stream,
    };
@@ -1415,6 +1441,39 @@ mod tests {
        .expect("usage should build")
    }
    #[test]
    fn users_me_usage_server_now_unix_ms_uses_epoch_millis() {
        let before = u64::try_from(Utc::now().timestamp_millis()).unwrap_or_default();
        let value = users_me_usage_server_now_unix_ms();
        let after = u64::try_from(Utc::now().timestamp_millis()).unwrap_or_default();
        assert!(value >= before);
        assert!(value <= after);
        assert!(value > 1_000_000_000_000);
    }
    fn assert_users_me_usage_server_now_header(response: &Response<Body>) {
        let value = response
            .headers()
            .get(USAGE_SERVER_NOW_UNIX_MS_HEADER)
            .expect("user usage response should include server now header")
            .to_str()
            .expect("server now header should be valid ASCII")
            .parse::<u64>()
            .expect("server now header should be epoch millis");
        assert!(value > 1_000_000_000_000);
    }
    #[test]
    fn users_me_usage_server_now_header_is_added_to_response() {
        let response = attach_users_me_usage_server_now_header(
            Json(json!({ "requests": [] })).into_response(),
        );
        assert_users_me_usage_server_now_header(&response);
    }
    #[test]
    fn user_usage_record_payload_rehydrates_cache_creation_total_from_classified_fields() {
        let item = StoredRequestUsageAudit {
@@ -1,3 +1,4 @@
 use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
 use axum::body::Body;
 use axum::extract::{Request, State};
 use axum::http::{self, HeaderValue, Response};
@@ -6,6 +7,8 @@ use axum::middleware::Next;
 use crate::headers::header_value_str;
 use crate::state::{AppState, FrontdoorCorsConfig};
 const FRONTDOOR_CREDENTIALS_EXPOSE_HEADERS: &str = "*, x-aether-server-now-unix-ms";
 fn append_vary(headers: &mut http::HeaderMap, value: &'static str) {
    headers.append(http::header::VARY, HeaderValue::from_static(value));
 }
@@ -31,7 +34,11 @@ fn apply_frontdoor_cors_headers(
    );
    headers.insert(
        http::header::ACCESS_CONTROL_EXPOSE_HEADERS,
-        HeaderValue::from_static("*"),
+        HeaderValue::from_static(if cors.allow_credentials() {
            FRONTDOOR_CREDENTIALS_EXPOSE_HEADERS
        } else {
            "*"
        }),
    );
    if let Some(value) = requested_headers {
        if let Ok(value) = HeaderValue::from_str(value) {
@@ -109,3 +116,52 @@ pub(crate) async fn frontdoor_cors_middleware(
    );
    response
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    fn assert_exposes_header(value: &HeaderValue, expected: &str) {
        let exposed_headers = value
            .to_str()
            .expect("expose headers should be valid ASCII");
        assert!(
            exposed_headers
                .split(',')
                .map(str::trim)
                .any(|header| header.eq_ignore_ascii_case(expected)),
            "{exposed_headers} should include {expected}"
        );
    }
    #[test]
    fn frontdoor_cors_explicitly_exposes_usage_server_time_for_credentials() {
        let cors = FrontdoorCorsConfig::new(vec!["http://localhost:5173".to_string()], true)
            .expect("cors config should build");
        let mut headers = http::HeaderMap::new();
        apply_frontdoor_cors_headers(&mut headers, &cors, "http://localhost:5173", None);
        let expose_headers = headers
            .get(http::header::ACCESS_CONTROL_EXPOSE_HEADERS)
            .expect("expose headers should be set");
        assert_exposes_header(expose_headers, "*");
        assert_exposes_header(expose_headers, USAGE_SERVER_NOW_UNIX_MS_HEADER);
    }
    #[test]
    fn frontdoor_cors_keeps_wildcard_expose_headers_without_credentials() {
        let cors = FrontdoorCorsConfig::new(vec!["http://localhost:5173".to_string()], false)
            .expect("cors config should build");
        let mut headers = http::HeaderMap::new();
        apply_frontdoor_cors_headers(&mut headers, &cors, "http://localhost:5173", None);
        assert_eq!(
            headers
                .get(http::header::ACCESS_CONTROL_EXPOSE_HEADERS)
                .expect("expose headers should be set"),
            "*"
        );
    }
 }
@@ -33,6 +33,7 @@ use crate::constants::{
 };
 use crate::control::resolve_public_request_context;
 use crate::data::GatewayDataState;
 use crate::tests::{assert_usage_server_now_header_between, unix_epoch_millis_for_tests};
 const ADMIN_USAGE_DATA_UNAVAILABLE_DETAIL: &str = "Admin usage data unavailable";
 const DAY_1_UNIX_SECS: i64 = 1_711_000_000;
@@ -1014,6 +1015,7 @@ async fn gateway_handles_admin_usage_active_locally_with_trusted_admin_principal
    );
    let (gateway_url, gateway_handle) = start_server(gateway).await;
    let client_send_unix_ms = unix_epoch_millis_for_tests();
    let response =
        admin_request(reqwest::Client::new().get(format!(
            "{gateway_url}/api/admin/usage/active?start_date=2024-03-21&end_date=2024-03-22&tz_offset_minutes=0"
@@ -1021,9 +1023,17 @@ async fn gateway_handles_admin_usage_active_locally_with_trusted_admin_principal
        .send()
        .await
        .expect("request should succeed");
    let client_receive_unix_ms = unix_epoch_millis_for_tests();
    assert_eq!(response.status(), StatusCode::OK);
    let response_headers = response.headers().clone();
    assert_usage_server_now_header_between(
        &response_headers,
        client_send_unix_ms,
        client_receive_unix_ms,
    );
    let payload: serde_json::Value = response.json().await.expect("json body should parse");
    assert!(payload.get("server_now_unix_ms").is_none());
    assert_eq!(payload["requests"].as_array().expect("array").len(), 1);
    assert_eq!(payload["requests"][0]["id"], "usage-pending");
    assert_eq!(payload["requests"][0]["effective_input_tokens"], 5);
@@ -1233,15 +1243,24 @@ async fn gateway_handles_admin_usage_records_locally_with_trusted_admin_principa
    );
    let (gateway_url, gateway_handle) = start_server(gateway).await;
    let client_send_unix_ms = unix_epoch_millis_for_tests();
    let response = admin_request(reqwest::Client::new().get(format!(
        "{gateway_url}/api/admin/usage/records?start_date=2024-03-21&end_date=2024-03-22&tz_offset_minutes=0&status=failed&provider=Anthropic&limit=10&offset=0"
    )))
    .send()
    .await
    .expect("request should succeed");
    let client_receive_unix_ms = unix_epoch_millis_for_tests();
    assert_eq!(response.status(), StatusCode::OK);
    let response_headers = response.headers().clone();
    assert_usage_server_now_header_between(
        &response_headers,
        client_send_unix_ms,
        client_receive_unix_ms,
    );
    let payload: serde_json::Value = response.json().await.expect("json body should parse");
    assert!(payload.get("server_now_unix_ms").is_none());
    assert_eq!(payload["total"], 1);
    assert_eq!(payload["records"][0]["id"], "usage-b");
    assert_eq!(payload["records"][0]["username"], "bob");
@@ -7,6 +7,7 @@ use crate::tests::{
    build_state_with_execution_runtime_override, json, start_server, AppState, Arc, Body,
    FrontdoorCorsConfig, Mutex, Request, Router, StatusCode, FRONTDOOR_MANIFEST_PATH, READYZ_PATH,
 };
 use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
 use aether_crypto::DEVELOPMENT_ENCRYPTION_KEY;
 use aether_data::repository::auth::InMemoryAuthApiKeySnapshotRepository;
 use aether_data::repository::candidate_selection::InMemoryMinimalCandidateSelectionReadRepository;
@@ -440,12 +441,19 @@ async fn gateway_adds_cors_headers_to_proxied_responses() {
            .expect("allow origin header"),
        "http://localhost:3000"
    );
-    assert_eq!(
+    let expose_headers = response_headers
        response_headers
        .get("access-control-expose-headers")
-            .expect("expose headers header"),
+        .expect("expose headers header")
-        "*"
+        .to_str()
-    );
+        .expect("expose headers should be valid ASCII");
    assert!(expose_headers
        .split(',')
        .map(str::trim)
        .any(|header| header == "*"));
    assert!(expose_headers
        .split(',')
        .map(str::trim)
        .any(|header| header.eq_ignore_ascii_case(USAGE_SERVER_NOW_UNIX_MS_HEADER)));
    assert_eq!(
        *execution_runtime_hits.lock().expect("mutex should lock"),
        1
@@ -11,10 +11,10 @@ use super::{
 };
 use crate::data::GatewayDataState;
 use crate::tests::{
-    any, build_router, build_router_with_state, json, start_server, to_bytes, AppState, Arc, Body,
+    any, assert_usage_server_now_header_between, build_router, build_router_with_state, json,
-    Json, Mutex, Request, Router, StatusCode, CONTROL_ROUTE_FAMILY_HEADER,
+    start_server, to_bytes, unix_epoch_millis_for_tests, AppState, Arc, Body, Json, Mutex, Request,
-    CONTROL_ROUTE_KIND_HEADER, TRUSTED_ADMIN_SESSION_ID_HEADER, TRUSTED_ADMIN_USER_ID_HEADER,
+    Router, StatusCode, CONTROL_ROUTE_FAMILY_HEADER, CONTROL_ROUTE_KIND_HEADER,
-    TRUSTED_ADMIN_USER_ROLE_HEADER,
+    TRUSTED_ADMIN_SESSION_ID_HEADER, TRUSTED_ADMIN_USER_ID_HEADER, TRUSTED_ADMIN_USER_ROLE_HEADER,
 };
 use aether_crypto::{encrypt_python_fernet_plaintext, DEVELOPMENT_ENCRYPTION_KEY};
 use aether_data::repository::announcements::{AnnouncementListQuery, AnnouncementReadRepository};
@@ -5345,6 +5345,7 @@ async fn gateway_handles_users_me_usage_locally_without_proxying_upstream() {
        })
        .await;
    let client_send_unix_ms = unix_epoch_millis_for_tests();
    let response = reqwest::Client::new()
        .get(format!(
            "{gateway_url}/api/users/me/usage?limit=10&offset=0&search=renamed-key"
@@ -5355,9 +5356,17 @@ async fn gateway_handles_users_me_usage_locally_without_proxying_upstream() {
        .send()
        .await
        .expect("request should succeed");
    let client_receive_unix_ms = unix_epoch_millis_for_tests();
    assert_eq!(response.status(), StatusCode::OK);
    let response_headers = response.headers().clone();
    assert_usage_server_now_header_between(
        &response_headers,
        client_send_unix_ms,
        client_receive_unix_ms,
    );
    let payload: serde_json::Value = response.json().await.expect("json body should parse");
    assert!(payload.get("server_now_unix_ms").is_none());
    assert_eq!(payload["total_requests"], 2);
    assert_eq!(payload["total_input_tokens"], 240);
    assert_eq!(payload["pagination"]["total"], 3);
@@ -5667,6 +5676,7 @@ async fn gateway_handles_users_me_usage_active_locally_without_proxying_upstream
        )
        .await;
    let client_send_unix_ms = unix_epoch_millis_for_tests();
    let response = reqwest::Client::new()
        .get(format!("{gateway_url}/api/users/me/usage/active"))
        .header("authorization", format!("Bearer {access_token}"))
@@ -5675,9 +5685,17 @@ async fn gateway_handles_users_me_usage_active_locally_without_proxying_upstream
        .send()
        .await
        .expect("request should succeed");
    let client_receive_unix_ms = unix_epoch_millis_for_tests();
    assert_eq!(response.status(), StatusCode::OK);
    let response_headers = response.headers().clone();
    assert_usage_server_now_header_between(
        &response_headers,
        client_send_unix_ms,
        client_receive_unix_ms,
    );
    let payload: serde_json::Value = response.json().await.expect("json body should parse");
    assert!(payload.get("server_now_unix_ms").is_none());
    let requests = payload["requests"].as_array().expect("requests array");
    assert_eq!(requests.len(), 2);
    assert_eq!(requests[0]["status"], "streaming");
@@ -1,6 +1,9 @@
 use std::time::{SystemTime, UNIX_EPOCH};
 pub(super) use std::convert::Infallible;
 pub(super) use std::sync::{Arc, Mutex};
 use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
 pub(super) use axum::body::{to_bytes, Body, Bytes};
 pub(super) use axum::response::Response;
 pub(super) use axum::routing::any;
@@ -29,6 +32,38 @@ pub(super) use super::router::{attach_static_frontend, build_router, build_route
 pub(super) use super::state::{AppState, FrontdoorCorsConfig};
 pub(super) use super::usage::UsageRuntimeConfig;
 const SERVER_NOW_HEADER_TEST_TOLERANCE_MS: u64 = 1_000;
 pub(super) fn unix_epoch_millis_for_tests() -> u64 {
    SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .expect("current time should be after epoch")
        .as_millis()
        .try_into()
        .expect("current epoch millis should fit in u64")
 }
 pub(super) fn assert_usage_server_now_header_between(
    headers: &reqwest::header::HeaderMap,
    lower_bound_unix_ms: u64,
    upper_bound_unix_ms: u64,
 ) {
    let server_now_unix_ms = headers
        .get(USAGE_SERVER_NOW_UNIX_MS_HEADER)
        .expect("usage response should include server timing header")
        .to_str()
        .expect("server timing header should be valid ASCII")
        .parse::<u64>()
        .expect("server timing header should be epoch millis");
    assert!(
        server_now_unix_ms >= lower_bound_unix_ms.saturating_sub(SERVER_NOW_HEADER_TEST_TOLERANCE_MS)
            && server_now_unix_ms
                <= upper_bound_unix_ms.saturating_add(SERVER_NOW_HEADER_TEST_TOLERANCE_MS),
        "server timing header {server_now_unix_ms} should be near request window {lower_bound_unix_ms}..={upper_bound_unix_ms}"
    );
 }
 pub(super) async fn start_server(app: Router) -> (String, tokio::task::JoinHandle<()>) {
    let listener = crate::test_support::bind_loopback_listener()
        .await
@@ -19,8 +19,24 @@ use serde_json::{json, Value};
 use std::collections::{BTreeMap, BTreeSet};
 use url::form_urlencoded;
 pub use aether_contracts::USAGE_SERVER_NOW_UNIX_MS_HEADER;
 pub const ADMIN_USAGE_DATA_UNAVAILABLE_DETAIL: &str = "Admin usage data unavailable";
 pub fn usage_server_now_unix_ms() -> u64 {
    u64::try_from(chrono::Utc::now().timestamp_millis()).unwrap_or_default()
 }
 pub fn attach_usage_server_now_header(mut response: Response<Body>) -> Response<Body> {
    if let Ok(value) = http::HeaderValue::from_str(&usage_server_now_unix_ms().to_string()) {
        response.headers_mut().insert(
            http::HeaderName::from_static(USAGE_SERVER_NOW_UNIX_MS_HEADER),
            value,
        );
    }
    response
 }
 pub fn admin_usage_data_unavailable_response(detail: &'static str) -> Response<Body> {
    (
        http::StatusCode::SERVICE_UNAVAILABLE,
@@ -2275,7 +2291,7 @@ pub fn build_admin_usage_active_requests_response(
        })
        .collect();
-    Json(json!({ "requests": payload })).into_response()
+    attach_usage_server_now_header(Json(json!({ "requests": payload })).into_response())
 }
 #[allow(clippy::too_many_arguments)]
@@ -2305,13 +2321,15 @@ pub fn build_admin_usage_records_response(
        })
        .collect();
    attach_usage_server_now_header(
        Json(json!({
            "records": records,
            "total": total,
            "limit": limit,
            "offset": offset,
        }))
-    .into_response()
+        .into_response(),
    )
 }
 pub fn build_admin_usage_curl_response(
@@ -2503,6 +2521,7 @@ pub fn build_admin_usage_replay_plan_response(
 mod tests {
    use std::collections::BTreeMap;
    use axum::{body::Body, response::Response};
    use serde_json::json;
    use super::{
@@ -2510,7 +2529,10 @@ mod tests {
        admin_usage_has_fallback, admin_usage_is_failed, admin_usage_is_success,
        admin_usage_matches_search, admin_usage_matches_status, admin_usage_matches_username,
        admin_usage_record_json, admin_usage_resolve_request_capture_body,
-        admin_usage_total_tokens, admin_usage_upstream_is_stream, build_admin_usage_detail_payload,
+        admin_usage_total_tokens, admin_usage_upstream_is_stream,
        build_admin_usage_active_requests_response, build_admin_usage_detail_payload,
        build_admin_usage_records_response, usage_server_now_unix_ms,
        USAGE_SERVER_NOW_UNIX_MS_HEADER,
    };
    use aether_data_contracts::repository::usage::{StoredRequestUsageAudit, UsageBodyField};
@@ -2560,6 +2582,62 @@ mod tests {
        .expect("usage should build")
    }
    #[test]
    fn usage_server_now_unix_ms_uses_epoch_millis() {
        let before = u64::try_from(chrono::Utc::now().timestamp_millis()).unwrap_or_default();
        let value = usage_server_now_unix_ms();
        let after = u64::try_from(chrono::Utc::now().timestamp_millis()).unwrap_or_default();
        assert!(value >= before);
        assert!(value <= after);
        assert!(value > 1_000_000_000_000);
    }
    fn assert_usage_server_now_header(response: &Response<Body>) {
        let value = response
            .headers()
            .get(USAGE_SERVER_NOW_UNIX_MS_HEADER)
            .expect("usage response should include server now header")
            .to_str()
            .expect("server now header should be valid ASCII")
            .parse::<u64>()
            .expect("server now header should be epoch millis");
        assert!(value > 1_000_000_000_000);
    }
    #[test]
    fn admin_usage_records_response_sets_server_now_header() {
        let item = sample_usage("completed", Some(200), None);
        let response = build_admin_usage_records_response(
            &[item],
            &BTreeMap::new(),
            &BTreeMap::new(),
            true,
            true,
            &BTreeMap::new(),
            1,
            20,
            0,
        );
        assert_usage_server_now_header(&response);
    }
    #[test]
    fn admin_usage_active_response_sets_server_now_header() {
        let item = sample_usage("streaming", Some(200), None);
        let response = build_admin_usage_active_requests_response(
            &[item],
            &BTreeMap::new(),
            true,
            &BTreeMap::new(),
            &BTreeMap::new(),
        );
        assert_usage_server_now_header(&response);
    }
    #[test]
    fn explicit_completed_status_wins_over_legacy_failure_fields() {
        let item = sample_usage(
@@ -16,4 +16,6 @@ pub use plan::{
    TRANSPORT_HTTP_MODE_HTTP1_ONLY, TRANSPORT_POOL_SCOPE_KEY,
 };
 pub use result::{ExecutionResult, ExecutionTelemetry, ResponseBody};
-pub use usage::{ExecutionStreamTerminalSummary, StandardizedUsage};
+pub use usage::{
    ExecutionStreamTerminalSummary, StandardizedUsage, USAGE_SERVER_NOW_UNIX_MS_HEADER,
 };
@@ -2,6 +2,8 @@ use std::collections::BTreeMap;
 use serde::{Deserialize, Serialize};
 pub const USAGE_SERVER_NOW_UNIX_MS_HEADER: &str = "x-aether-server-now-unix-ms";
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Default)]
 pub struct StandardizedUsage {
    pub input_tokens: i64,
@@ -0,0 +1,62 @@
 import { describe, expect, it } from 'vitest'
 import {
  SERVER_NOW_UNIX_MS_HEADER,
  buildServerTimingMetadata,
  readServerNowUnixMsFromHeaders,
  withServerTiming,
 } from '../serverTiming'
 describe('serverTiming', () => {
  it('reads server time from response headers', () => {
    expect(readServerNowUnixMsFromHeaders({
      [SERVER_NOW_UNIX_MS_HEADER]: '1779999000123',
    })).toBe(1_779_999_000_123)
    expect(readServerNowUnixMsFromHeaders({
      'X-Aether-Server-Now-Unix-Ms': '1779999000456',
    })).toBe(1_779_999_000_456)
  })
  it('does not fall back to body fields', () => {
    const timing = buildServerTimingMetadata({
      headers: {},
      data: {
        server_now_unix_ms: 1_779_999_000_123,
      },
    }, 1_000, 1_100)
    expect(timing).toBeUndefined()
  })
  it('builds metadata with round trip duration', () => {
    const timing = buildServerTimingMetadata({
      headers: {
        [SERVER_NOW_UNIX_MS_HEADER]: '1050',
      },
    }, 1_000, 1_125)
    expect(timing).toEqual({
      server_now_unix_ms: 1_050,
      client_send_unix_ms: 1_000,
      client_receive_unix_ms: 1_125,
      round_trip_ms: 125,
    })
  })
  it('returns the original payload when the header is missing or invalid', () => {
    const payload = { records: [] }
    expect(withServerTiming({ data: payload, headers: {} }, 1_000)).toBe(payload)
    expect(withServerTiming({
      data: payload,
      headers: { [SERVER_NOW_UNIX_MS_HEADER]: 'not-a-number' },
    }, 1_000)).toBe(payload)
    expect(withServerTiming({
      data: payload,
      headers: { [SERVER_NOW_UNIX_MS_HEADER]: '0' },
    }, 1_000)).toBe(payload)
    expect(withServerTiming({
      data: payload,
      headers: { [SERVER_NOW_UNIX_MS_HEADER]: '1050.5' },
    }, 1_000)).toBe(payload)
  })
 })
@@ -1,4 +1,4 @@
-import { beforeEach, describe, expect, it, vi } from 'vitest'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 const { getMock, cachedRequestMock, dedupedRequestMock, buildCacheKeyMock } = vi.hoisted(() => ({
  getMock: vi.fn(),
@@ -29,6 +29,10 @@ describe('usageApi contract alignment', () => {
    buildCacheKeyMock.mockClear()
  })
  afterEach(() => {
    vi.restoreAllMocks()
  })
  it('loads current-user usage records from the Rust usage endpoint and normalizes pagination', async () => {
    getMock.mockResolvedValueOnce({
      data: {
@@ -115,6 +119,56 @@ describe('usageApi contract alignment', () => {
    })
  })
  it('captures admin user usage server timing when the records request resolves', async () => {
    let now = 1_000
    vi.spyOn(Date, 'now').mockImplementation(() => now)
    let resolveStats: ((value: unknown) => void) | null = null
    getMock.mockImplementation((url: string) => {
      if (url === '/api/admin/usage/stats') {
        return new Promise(resolve => {
          resolveStats = resolve
        })
      }
      if (url === '/api/admin/usage/records') {
        return Promise.resolve({
          headers: {
            'x-aether-server-now-unix-ms': '10050',
          },
          data: {
            records: [{ id: 'record-3' }],
            total: 1,
            limit: 25,
            offset: 0,
          },
        })
      }
      return Promise.reject(new Error(`unexpected url: ${url}`))
    })
    const resultPromise = usageApi.getUserUsage('user-123', { page: 1, page_size: 25 })
    now = 1_100
    await Promise.resolve()
    now = 20_000
    resolveStats?.({
      data: {
        total_requests: 1,
        total_tokens: 10,
        total_cost: 0.1,
        avg_response_time: 500,
      },
    })
    const result = await resultPromise
    expect(result.server_timing).toEqual({
      server_now_unix_ms: 10_050,
      client_send_unix_ms: 1_000,
      client_receive_unix_ms: 1_100,
      round_trip_ms: 100,
    })
  })
  it('uses an extended timeout and cache bypass option for admin analytics', async () => {
    getMock
      .mockResolvedValueOnce({
@@ -5,6 +5,11 @@ import { cachedRequest, buildCacheKey } from '@/utils/cache'
 import type { BillingSummary } from './auth'
 import type { UserSession } from '@/types/session'
 import type { FeatureSettingsMap } from '@/utils/featureSettings'
 import {
  beginServerTimingSample,
  withServerTiming,
  type ServerTimedPayload,
 } from './serverTiming'
 const ACTIVITY_HEATMAP_CACHE_TTL_MS = 30 * 60 * 1000
@@ -142,7 +147,7 @@ export interface ApiFormatSummary {
 }
 // 使用统计响应接口
-export interface UsageResponse {
+export interface UsageResponse extends ServerTimedPayload {
  total_requests: number
  total_input_tokens: number
  total_output_tokens: number
@@ -321,12 +326,14 @@ export const meApi = {
    limit?: number
    offset?: number
  }): Promise<UsageResponse> {
    const clientSendUnixMs = beginServerTimingSample()
    const response = await apiClient.get<UsageResponse>('/api/users/me/usage', { params })
-    return response.data
+    return withServerTiming(response, clientSendUnixMs)
  },
  // 获取活跃请求状态（用于轮询更新）
  async getActiveRequests(ids?: string): Promise<{
    server_timing?: ServerTimedPayload['server_timing']
    requests: Array<{
      id: string
      status: 'pending' | 'streaming' | 'completed' | 'failed' | 'cancelled'
@@ -358,8 +365,9 @@ export const meApi = {
    }>
  }> {
    const params = ids ? { ids } : {}
    const clientSendUnixMs = beginServerTimingSample()
    const response = await apiClient.get('/api/users/me/usage/active', { params })
-    return response.data
+    return withServerTiming(response, clientSendUnixMs)
  },
  // 获取可用的提供商
@@ -0,0 +1,78 @@
 import type { AxiosResponse } from 'axios'
 export const SERVER_NOW_UNIX_MS_HEADER = 'x-aether-server-now-unix-ms'
 export interface ServerTimingMetadata {
  server_now_unix_ms: number
  client_send_unix_ms: number
  client_receive_unix_ms: number
  round_trip_ms: number
 }
 export interface ServerTimedPayload {
  server_timing?: ServerTimingMetadata
 }
 export function beginServerTimingSample(): number {
  return Date.now()
 }
 function readHeaderValue(headers: unknown, name: string): unknown {
  if (!headers || typeof headers !== 'object') return undefined
  const get = (headers as { get?: unknown }).get
  if (typeof get === 'function') {
    return get.call(headers, name)
  }
  const lowerName = name.toLowerCase()
  for (const [key, value] of Object.entries(headers as Record<string, unknown>)) {
    if (key.toLowerCase() === lowerName) return value
  }
  return undefined
 }
 export function readServerNowUnixMsFromHeaders(headers: unknown): number | null {
  const value = readHeaderValue(headers, SERVER_NOW_UNIX_MS_HEADER)
  const raw = Array.isArray(value) ? value[0] : value
  const parsed = typeof raw === 'number'
    ? raw
    : typeof raw === 'string'
      ? Number(raw.trim())
      : Number.NaN
  return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : null
 }
 export function buildServerTimingMetadata(
  response: Pick<AxiosResponse, 'headers'> | { headers?: unknown } | null | undefined,
  clientSendUnixMs: number,
  clientReceiveUnixMs = Date.now()
 ): ServerTimingMetadata | undefined {
  const serverNowUnixMs = readServerNowUnixMsFromHeaders(response?.headers)
  if (serverNowUnixMs == null) return undefined
  if (!Number.isFinite(clientSendUnixMs) || !Number.isFinite(clientReceiveUnixMs)) return undefined
  if (clientReceiveUnixMs < clientSendUnixMs) return undefined
  const roundTripMs = clientReceiveUnixMs - clientSendUnixMs
  return {
    server_now_unix_ms: serverNowUnixMs,
    client_send_unix_ms: clientSendUnixMs,
    client_receive_unix_ms: clientReceiveUnixMs,
    round_trip_ms: roundTripMs,
  }
 }
 export function withServerTiming<T extends object>(
  response: Pick<AxiosResponse<T>, 'data' | 'headers'>,
  clientSendUnixMs: number
 ): T & ServerTimedPayload {
  const serverTiming = buildServerTimingMetadata(response, clientSendUnixMs)
  if (!serverTiming) return response.data
  return {
    ...response.data,
    server_timing: serverTiming,
  }
 }
@@ -2,6 +2,11 @@ import apiClient from './client'
 import { cachedRequest, dedupedRequest, buildCacheKey } from '@/utils/cache'
 import type { ActivityHeatmap } from '@/types/activity'
 import type { ImageProgress } from './requestTrace'
 import {
  beginServerTimingSample,
  withServerTiming,
  type ServerTimedPayload,
 } from './serverTiming'
 const ACTIVITY_HEATMAP_CACHE_TTL_MS = 30 * 60 * 1000
 const USAGE_ANALYTICS_CACHE_TTL_MS = 30 * 1000
@@ -127,7 +132,7 @@ export interface UsageRequestOptions {
  skipCache?: boolean
 }
-type UsageListResponse = {
+type UsageListResponse = ServerTimedPayload & {
  records?: unknown
  pagination?: {
    total?: unknown
@@ -199,6 +204,7 @@ function normalizeUsageRecordPage(
  total: number
  page: number
  page_size: number
  server_timing?: ServerTimedPayload['server_timing']
 } {
  const records = assertUsageRecords(payload.records)
  const pagination = payload.pagination
@@ -220,6 +226,7 @@ function normalizeUsageRecordPage(
    total,
    page: resolvedPage,
    page_size: limit,
    ...(payload.server_timing ? { server_timing: payload.server_timing } : {}),
  }
 }
@@ -366,10 +373,12 @@ export const usageApi = {
    total: number
    page: number
    page_size: number
    server_timing?: ServerTimedPayload['server_timing']
  }> {
    const { params, pagination } = buildCurrentUserUsageParams(filters)
    const clientSendUnixMs = beginServerTimingSample()
    const response = await apiClient.get<UsageListResponse>('/api/users/me/usage', { params })
-    return normalizeUsageRecordPage(response.data, pagination)
+    return normalizeUsageRecordPage(withServerTiming(response, clientSendUnixMs), pagination)
  },
  async getUsageStats(filters?: UsageFilters, options?: UsageRequestOptions): Promise<UsageStats> {
@@ -444,17 +453,25 @@ export const usageApi = {
  async getUserUsage(userId: string, filters?: UsageFilters): Promise<{
    records: UsageRecord[]
    stats: UsageStats
    server_timing?: ServerTimedPayload['server_timing']
  }> {
    const statsParams = buildAdminUsageStatsParams(userId, filters)
    const { params: recordParams } = buildAdminUsageRecordParams(userId, filters)
    const statsRequest = apiClient.get<UsageStats>('/api/admin/usage/stats', { params: statsParams })
    const recordsClientSendUnixMs = beginServerTimingSample()
    const recordsRequest = apiClient
      .get<UsageListResponse>('/api/admin/usage/records', { params: recordParams })
      .then(response => withServerTiming(response, recordsClientSendUnixMs))
    const [statsResponse, recordsResponse] = await Promise.all([
-      apiClient.get<UsageStats>('/api/admin/usage/stats', { params: statsParams }),
+      statsRequest,
-      apiClient.get<UsageListResponse>('/api/admin/usage/records', { params: recordParams }),
+      recordsRequest,
    ])
    return {
-      records: assertUsageRecords(recordsResponse.data.records),
+      records: assertUsageRecords(recordsResponse.records),
      stats: statsResponse.data,
      ...(recordsResponse.server_timing ? { server_timing: recordsResponse.server_timing } : {}),
    }
  },
@@ -479,11 +496,13 @@ export const usageApi = {
    total: number
    limit: number
    offset: number
    server_timing?: ServerTimedPayload['server_timing']
  }> {
    const key = buildCacheKey('usage:records', params as Record<string, unknown> | undefined)
    return dedupedRequest(key, async () => {
      const clientSendUnixMs = beginServerTimingSample()
      const response = await apiClient.get('/api/admin/usage/records', { params })
-      return response.data
+      return withServerTiming(response, clientSendUnixMs)
    })
  },
@@ -495,6 +514,7 @@ export const usageApi = {
    ids?: string[],
    timeRange?: Pick<UsageFilters, 'start_date' | 'end_date' | 'preset' | 'timezone' | 'tz_offset_minutes'>
  ): Promise<{
    server_timing?: ServerTimedPayload['server_timing']
    requests: Array<{
      id: string
      status: 'pending' | 'streaming' | 'completed' | 'failed' | 'cancelled'
@@ -548,8 +568,9 @@ export const usageApi = {
    if (typeof timeRange?.tz_offset_minutes === 'number') {
      params.tz_offset_minutes = timeRange.tz_offset_minutes
    }
    const clientSendUnixMs = beginServerTimingSample()
    const response = await apiClient.get('/api/admin/usage/active', { params })
-    return response.data
+    return withServerTiming(response, clientSendUnixMs)
  },
  /**
@@ -3,26 +3,25 @@
 </template>
 <script setup lang="ts">
-import { computed, onUnmounted, ref, watch } from 'vue'
+import { computed } from 'vue'
 const props = withDefaults(defineProps<{
  createdAt?: string | null
  status?: string | null
  responseTimeMs?: number | null
  displayNowMs?: number | null
  precision?: number
 }>(), {
  createdAt: null,
  status: null,
  responseTimeMs: null,
  displayNowMs: null,
  precision: 2,
 })
 const now = ref(Date.now())
 const precision = computed(() => Math.max(0, props.precision))
 const isActive = computed(() => props.status === 'pending' || props.status === 'streaming')
 let rafId: number | null = null
 function parseCreatedAtMs(value: string | null | undefined): number {
  if (!value) return Number.NaN
  // 后端有时返回无时区时间，按 UTC 解析，和列表时间显示逻辑保持一致
@@ -30,35 +29,6 @@ function parseCreatedAtMs(value: string | null | undefined): number {
  return new Date(normalized).getTime()
 }
 function stopRaf() {
  if (rafId == null) return
  cancelAnimationFrame(rafId)
  rafId = null
 }
 function tick() {
  now.value = Date.now()
  rafId = requestAnimationFrame(tick)
 }
 function startRaf() {
  stopRaf()
  now.value = Date.now()
  rafId = requestAnimationFrame(tick)
 }
 watch(isActive, (active) => {
  if (active) {
    startRaf()
  } else {
    stopRaf()
  }
 }, { immediate: true })
 onUnmounted(() => {
  stopRaf()
 })
 const displayText = computed(() => {
  if (!isActive.value) {
    if (props.responseTimeMs == null) return '-'
@@ -70,7 +40,11 @@ const displayText = computed(() => {
  const createdAtMs = parseCreatedAtMs(props.createdAt)
  if (Number.isNaN(createdAtMs)) return '-'
-  const elapsedMs = Math.max(0, now.value - createdAtMs)
+  // 活跃请求里的 response_time_ms 可能只是首字或中间值；终态才使用后端最终耗时。
  const nowMs = typeof props.displayNowMs === 'number' && Number.isFinite(props.displayNowMs)
    ? props.displayNowMs
    : Date.now()
  const elapsedMs = Math.max(0, nowMs - createdAtMs)
  return `${(elapsedMs / 1000).toFixed(precision.value)}s`
 })
 </script>
@@ -324,6 +324,7 @@
                :created-at="record.created_at"
                :status="getDisplayStatus(record)"
                :response-time-ms="record.response_time_ms ?? null"
                :display-now-ms="displayNowMs ?? null"
              />
            </span>
            <span
@@ -940,6 +941,7 @@
                  :created-at="record.created_at"
                  :status="getDisplayStatus(record)"
                  :response-time-ms="record.response_time_ms ?? null"
                  :display-now-ms="displayNowMs ?? null"
                />
              </span>
            </div>
@@ -1114,6 +1116,7 @@ const props = defineProps<{
  pageSizeOptions: number[]
  // 自动刷新
  autoRefresh: boolean
  displayNowMs?: number | null
 }>()
 const emit = defineEmits<{
@@ -0,0 +1,56 @@
 import { afterEach, describe, expect, it } from 'vitest'
 import { createApp, type App } from 'vue'
 import ElapsedTimeText from '../ElapsedTimeText.vue'
 const mountedApps: Array<{ app: App, root: HTMLElement }> = []
 function mountElapsedTimeText(props: Record<string, unknown>) {
  const root = document.createElement('div')
  document.body.appendChild(root)
  const app = createApp(ElapsedTimeText, props)
  app.mount(root)
  mountedApps.push({ app, root })
  return root
 }
 afterEach(() => {
  for (const { app, root } of mountedApps.splice(0)) {
    app.unmount()
    root.remove()
  }
 })
 describe('ElapsedTimeText', () => {
  it('uses the supplied display clock for active requests', () => {
    const root = mountElapsedTimeText({
      createdAt: '2026-05-28T12:00:00Z',
      status: 'streaming',
      responseTimeMs: 10_000,
      displayNowMs: Date.parse('2026-05-28T12:00:40Z'),
    })
    expect(root.textContent).toBe('40.00s')
  })
  it('keeps terminal requests pinned to the backend final duration', () => {
    const root = mountElapsedTimeText({
      createdAt: '2026-05-28T12:00:00Z',
      status: 'completed',
      responseTimeMs: 42_340,
      displayNowMs: Date.parse('2026-05-28T12:01:30Z'),
    })
    expect(root.textContent).toBe('42.34s')
  })
  it('clamps active elapsed time at zero when the display clock is behind', () => {
    const root = mountElapsedTimeText({
      createdAt: '2026-05-28T12:00:40Z',
      status: 'pending',
      displayNowMs: Date.parse('2026-05-28T12:00:00Z'),
    })
    expect(root.textContent).toBe('0.00s')
  })
 })
@@ -91,8 +91,16 @@ vi.mock('lucide-vue-next', async () => {
 vi.mock('../ElapsedTimeText.vue', () => ({
  default: defineComponent({
    name: 'ElapsedTimeTextStub',
-    setup() {
+    props: {
-      return () => h('span', 'elapsed')
+      displayNowMs: {
        type: Number,
        default: null,
      },
    },
    setup(props) {
      return () => h('span', {
        'data-display-now-ms': props.displayNowMs == null ? '' : String(props.displayNowMs),
      }, 'elapsed')
    },
  }),
 }))
@@ -243,6 +251,16 @@ describe('UsageRecordsTable', () => {
    expect(root.querySelector('[data-active-latency-state="waiting-first-byte"]')).toBeNull()
  })
  it('passes the shared display clock to active elapsed text', () => {
    const root = mountUsageRecordsTable([buildRecord({
      status: 'streaming',
      response_time_ms: null,
      first_byte_time_ms: 500,
    })], { displayNowMs: 1_779_999_000_000 })
    expect(root.querySelector('[data-display-now-ms="1779999000000"]')).not.toBeNull()
  })
  it('shows failed when Codex image progress fails before the usage record finalizes', () => {
    const root = mountUsageRecordsTable([buildRecord({
      status: 'pending',
@@ -0,0 +1,101 @@
 import { afterEach, describe, expect, it, vi } from 'vitest'
 import { nextTick, ref } from 'vue'
 import { useActiveElapsedDisplayClock } from '../useActiveElapsedDisplayClock'
 type TestRecord = {
  status: string
 }
 afterEach(() => {
  vi.useRealTimers()
 })
 describe('useActiveElapsedDisplayClock', () => {
  it('ticks only while visible active records exist', async () => {
    vi.useFakeTimers()
    let nowMs = 1_000
    const records = ref<TestRecord[]>([])
    const isPageVisible = ref(true)
    const serverClockOffsetMs = ref(0)
    const hasServerClockOffset = ref(false)
    const clock = useActiveElapsedDisplayClock({
      records,
      isPageVisible,
      serverClockOffsetMs,
      hasServerClockOffset,
      resolveStatus: record => record.status,
      intervalMs: 250,
      now: () => nowMs,
    })
    expect(clock.hasVisibleActiveRecords.value).toBe(false)
    expect(clock.displayNowMs.value).toBe(1_000)
    nowMs = 1_250
    vi.advanceTimersByTime(250)
    expect(clock.displayNowMs.value).toBe(1_000)
    records.value = [{ status: 'streaming' }]
    await nextTick()
    expect(clock.hasVisibleActiveRecords.value).toBe(true)
    expect(clock.displayNowMs.value).toBe(1_250)
    nowMs = 1_500
    vi.advanceTimersByTime(250)
    expect(clock.displayNowMs.value).toBe(1_500)
    isPageVisible.value = false
    await nextTick()
    nowMs = 1_750
    vi.advanceTimersByTime(250)
    expect(clock.displayNowMs.value).toBe(1_500)
    clock.stopActiveElapsedDisplayClock()
  })
  it('applies server clock offset to the display time', () => {
    vi.useFakeTimers()
    const clock = useActiveElapsedDisplayClock({
      records: ref<TestRecord[]>([{ status: 'pending' }]),
      isPageVisible: ref(true),
      serverClockOffsetMs: ref(-11_000),
      hasServerClockOffset: ref(true),
      resolveStatus: record => record.status,
      intervalMs: 250,
      now: () => 2_000,
    })
    expect(clock.calibratedDisplayNowMs.value).toBe(-9_000)
    clock.stopActiveElapsedDisplayClock()
  })
  it('stops when active records disappear', async () => {
    vi.useFakeTimers()
    let nowMs = 3_000
    const records = ref<TestRecord[]>([{ status: 'pending' }])
    const clock = useActiveElapsedDisplayClock({
      records,
      isPageVisible: ref(true),
      serverClockOffsetMs: ref(0),
      hasServerClockOffset: ref(false),
      resolveStatus: record => record.status,
      intervalMs: 250,
      now: () => nowMs,
    })
    nowMs = 3_250
    vi.advanceTimersByTime(250)
    expect(clock.displayNowMs.value).toBe(3_250)
    records.value = [{ status: 'completed' }]
    await nextTick()
    nowMs = 3_500
    vi.advanceTimersByTime(250)
    expect(clock.displayNowMs.value).toBe(3_250)
    clock.stopActiveElapsedDisplayClock()
  })
 })
@@ -0,0 +1,102 @@
 import { describe, expect, it } from 'vitest'
 import {
  calculateServerClockOffsetMs,
  shouldUseServerClockSample,
  useServerClock
 } from '../useServerClock'
 describe('useServerClock', () => {
  it('calculates offset from the response receive time', () => {
    const offset = calculateServerClockOffsetMs({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_200,
      round_trip_ms: 200,
    })
    expect(offset).toBe(-9_700)
  })
  it('ignores missing or invalid timing samples', () => {
    expect(calculateServerClockOffsetMs(undefined)).toBeNull()
    expect(calculateServerClockOffsetMs({
      server_now_unix_ms: Number.NaN,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_200,
      round_trip_ms: 200,
    })).toBeNull()
    expect(calculateServerClockOffsetMs({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_200,
      client_receive_unix_ms: 20_000,
      round_trip_ms: 200,
    })).toBeNull()
    expect(calculateServerClockOffsetMs({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_200,
      round_trip_ms: Number.NaN,
    })).toBeNull()
  })
  it('keeps the previous offset when a response has no server timing', () => {
    const clock = useServerClock()
    clock.updateServerClockOffset({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_200,
      round_trip_ms: 200,
    })
    clock.updateServerClockOffset(undefined)
    expect(clock.hasServerClockOffset.value).toBe(true)
    expect(clock.serverClockOffsetMs.value).toBe(-9_700)
    expect(clock.serverClockSampleRoundTripMs.value).toBe(200)
  })
  it('does not let a much slower sample overwrite a better clock offset', () => {
    const clock = useServerClock()
    clock.updateServerClockOffset({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_050,
      round_trip_ms: 50,
    })
    clock.updateServerClockOffset({
      server_now_unix_ms: 20_500,
      client_send_unix_ms: 30_000,
      client_receive_unix_ms: 30_500,
      round_trip_ms: 500,
    })
    expect(clock.serverClockOffsetMs.value).toBe(-9_550)
    expect(clock.serverClockSampleRoundTripMs.value).toBe(50)
  })
  it('accepts a faster sample after an initial slow sample', () => {
    const clock = useServerClock()
    clock.updateServerClockOffset({
      server_now_unix_ms: 10_500,
      client_send_unix_ms: 20_000,
      client_receive_unix_ms: 20_500,
      round_trip_ms: 500,
    })
    clock.updateServerClockOffset({
      server_now_unix_ms: 20_500,
      client_send_unix_ms: 30_000,
      client_receive_unix_ms: 30_050,
      round_trip_ms: 50,
    })
    expect(clock.serverClockOffsetMs.value).toBe(-9_550)
    expect(clock.serverClockSampleRoundTripMs.value).toBe(50)
  })
  it('allows small RTT regressions so the offset can stay fresh', () => {
    expect(shouldUseServerClockSample(140, 50)).toBe(true)
    expect(shouldUseServerClockSample(151, 50)).toBe(false)
  })
 })
@@ -1,4 +1,5 @@
 export { useUsageData } from './useUsageData'
 export { useActiveElapsedDisplayClock } from './useActiveElapsedDisplayClock'
 export { useUsageFilters } from './useUsageFilters'
 export { useUsagePagination } from './useUsagePagination'
 export { getDateRangeFromPeriod, formatDateTime, getSuccessRateColor } from './useDateRange'
@@ -0,0 +1,84 @@
 import { computed, ref, watch, type ComputedRef, type Ref } from 'vue'
 type ActiveElapsedStatus = string | null | undefined
 export interface UseActiveElapsedDisplayClockOptions<TRecord> {
  records: Ref<TRecord[]> | ComputedRef<TRecord[]>
  isPageVisible: Ref<boolean> | ComputedRef<boolean>
  serverClockOffsetMs: Ref<number> | ComputedRef<number>
  hasServerClockOffset: Ref<boolean> | ComputedRef<boolean>
  resolveStatus: (record: TRecord) => ActiveElapsedStatus
  intervalMs?: number
  now?: () => number
 }
 export function useActiveElapsedDisplayClock<TRecord>(
  options: UseActiveElapsedDisplayClockOptions<TRecord>
 ) {
  const intervalMs = options.intervalMs ?? 250
  const now = options.now ?? Date.now
  const displayNowMs = ref(now())
  let activeElapsedDisplayTimer: ReturnType<typeof setInterval> | null = null
  const hasVisibleActiveRecords = computed(() => {
    return options.records.value.some((record) => {
      const displayStatus = options.resolveStatus(record)
      return displayStatus === 'pending' || displayStatus === 'streaming'
    })
  })
  const calibratedDisplayNowMs = computed(() => {
    return options.hasServerClockOffset.value
      ? displayNowMs.value + options.serverClockOffsetMs.value
      : displayNowMs.value
  })
  function tickActiveElapsedDisplay() {
    displayNowMs.value = now()
  }
  function startActiveElapsedDisplayTimer() {
    if (activeElapsedDisplayTimer) return
    if (!options.isPageVisible.value || !hasVisibleActiveRecords.value) return
    tickActiveElapsedDisplay()
    activeElapsedDisplayTimer = setInterval(tickActiveElapsedDisplay, intervalMs)
  }
  function stopActiveElapsedDisplayTimer() {
    if (activeElapsedDisplayTimer) {
      clearInterval(activeElapsedDisplayTimer)
      activeElapsedDisplayTimer = null
    }
  }
  function syncActiveElapsedDisplayTimer() {
    if (options.isPageVisible.value && hasVisibleActiveRecords.value) {
      startActiveElapsedDisplayTimer()
    } else {
      stopActiveElapsedDisplayTimer()
    }
  }
  const stopActiveElapsedDisplayWatch = watch(
    [hasVisibleActiveRecords, options.isPageVisible],
    syncActiveElapsedDisplayTimer,
    { immediate: true }
  )
  function stopActiveElapsedDisplayClock() {
    stopActiveElapsedDisplayWatch()
    stopActiveElapsedDisplayTimer()
  }
  return {
    displayNowMs,
    calibratedDisplayNowMs,
    hasVisibleActiveRecords,
    tickActiveElapsedDisplay,
    startActiveElapsedDisplayTimer,
    stopActiveElapsedDisplayTimer,
    syncActiveElapsedDisplayTimer,
    stopActiveElapsedDisplayClock,
  }
 }
@@ -0,0 +1,62 @@
 import { ref } from 'vue'
 import type { ServerTimingMetadata } from '@/api/serverTiming'
 const SERVER_CLOCK_RTT_REGRESSION_TOLERANCE_MS = 100
 export function calculateServerClockOffsetMs(timing: ServerTimingMetadata | null | undefined): number | null {
  if (!timing) return null
  const {
    server_now_unix_ms: serverNowUnixMs,
    client_send_unix_ms: clientSendUnixMs,
    client_receive_unix_ms: clientReceiveUnixMs,
    round_trip_ms: roundTripMs,
  } = timing
  if (
    !Number.isFinite(serverNowUnixMs) ||
    !Number.isFinite(clientSendUnixMs) ||
    !Number.isFinite(clientReceiveUnixMs) ||
    !Number.isFinite(roundTripMs)
  ) {
    return null
  }
  if (clientReceiveUnixMs < clientSendUnixMs || roundTripMs < 0) {
    return null
  }
  return serverNowUnixMs - clientReceiveUnixMs
 }
 export function shouldUseServerClockSample(
  nextRoundTripMs: number,
  currentRoundTripMs: number | null | undefined
 ): boolean {
  if (!Number.isFinite(nextRoundTripMs) || nextRoundTripMs < 0) return false
  if (currentRoundTripMs == null || !Number.isFinite(currentRoundTripMs)) return true
  return nextRoundTripMs <= currentRoundTripMs + SERVER_CLOCK_RTT_REGRESSION_TOLERANCE_MS
 }
 export function useServerClock() {
  const serverClockOffsetMs = ref(0)
  const hasServerClockOffset = ref(false)
  const serverClockSampleRoundTripMs = ref<number | null>(null)
  function updateServerClockOffset(timing: ServerTimingMetadata | null | undefined): void {
    const offsetMs = calculateServerClockOffsetMs(timing)
    if (offsetMs == null) return
    if (!shouldUseServerClockSample(timing?.round_trip_ms ?? Number.NaN, serverClockSampleRoundTripMs.value)) {
      return
    }
    serverClockOffsetMs.value = offsetMs
    serverClockSampleRoundTripMs.value = timing?.round_trip_ms ?? null
    hasServerClockOffset.value = true
  }
  return {
    serverClockOffsetMs,
    hasServerClockOffset,
    serverClockSampleRoundTripMs,
    updateServerClockOffset,
  }
 }
@@ -14,6 +14,7 @@ import { createDefaultStats } from '../types'
 import { log } from '@/utils/logger'
 import { getErrorStatus } from '@/types/api-error'
 import { isUsageProviderVisible, normalizeUsageProviderStats } from '../utils/providerStats'
 import { useServerClock } from './useServerClock'
 export interface UseUsageDataOptions {
  isAdminPage: Ref<boolean>
@@ -65,6 +66,11 @@ export function useUsageData(options: UseUsageDataOptions) {
  // 可用的筛选选项（从统计数据获取，而不是从记录中）
  const availableModels = ref<string[]>([])
  const availableProviders = ref<string[]>([])
  const {
    serverClockOffsetMs,
    hasServerClockOffset,
    updateServerClockOffset,
  } = useServerClock()
  // 增强的模型统计（包含效率分析）
  const enhancedModelStats = computed<EnhancedModelStatsItem[]>(() => {
@@ -213,6 +219,7 @@ export function useUsageData(options: UseUsageDataOptions) {
      if (requestId !== loadStatsRequestId) {
        return false
      }
      updateServerClockOffset(userData.server_timing)
      stats.value = {
        total_requests: userData.total_requests || 0,
@@ -366,6 +373,7 @@ export function useUsageData(options: UseUsageDataOptions) {
        if (requestId !== loadRecordsRequestId) {
          return
        }
        updateServerClockOffset(response.server_timing)
        const nextRecords = (response.records || []) as UsageRecord[]
        currentRecords.value = mergeRecordStatus(currentRecords.value, nextRecords)
        totalRecords.value = response.total || 0
@@ -375,6 +383,7 @@ export function useUsageData(options: UseUsageDataOptions) {
        if (requestId !== loadRecordsRequestId) {
          return
        }
        updateServerClockOffset(userData.server_timing)
        const nextRecords = (userData.records || []) as UsageRecord[]
        currentRecords.value = mergeRecordStatus(currentRecords.value, nextRecords)
        totalRecords.value = userData.pagination?.total || currentRecords.value.length
@@ -557,6 +566,8 @@ export function useUsageData(options: UseUsageDataOptions) {
    apiFormatStats,
    currentRecords,
    totalRecords,
    serverClockOffsetMs,
    hasServerClockOffset,
    // 筛选选项
    availableModels,
@@ -568,6 +579,7 @@ export function useUsageData(options: UseUsageDataOptions) {
    // 方法
    loadStats,
    loadRecords,
-    refreshData
+    refreshData,
    updateServerClockOffset
  }
 }
@@ -100,6 +100,7 @@
      :total-records="effectiveTotalRecords"
      :page-size-options="pageSizeOptions"
      :auto-refresh="globalAutoRefresh"
      :display-now-ms="calibratedDisplayNowMs"
      @update:time-range="handleTimeRangeChange"
      @update:filter-search="handleFilterSearchChange"
      @update:filter-user="handleFilterUserChange"
@@ -148,6 +149,7 @@ import {
  IntervalTimelineCard
 } from '@/features/usage/components'
 import {
  useActiveElapsedDisplayClock,
  useUsageData,
  getDateRangeFromPeriod
 } from '@/features/usage/composables'
@@ -246,7 +248,10 @@ const {
  availableModels,
  availableProviders,
  loadStats,
-  loadRecords
+  loadRecords,
  serverClockOffsetMs,
  hasServerClockOffset,
  updateServerClockOffset
 } = useUsageData({ isAdminPage })
 // 热力图状态
@@ -434,6 +439,7 @@ const AUTO_REFRESH_INTERVAL = 1000 // 1秒刷新一次（用于活跃请求）
 const ACTIVE_DISCOVERY_HOT_INTERVAL = 1000 // 有活跃请求时 1 秒扫描一次
 const ACTIVE_DISCOVERY_IDLE_INTERVAL = 5000 // 空闲时降频，避免后台持续刷日志
 const GLOBAL_AUTO_REFRESH_INTERVAL = 3000 // 3秒刷新一次（全局自动刷新）
 const ACTIVE_ELAPSED_DISPLAY_INTERVAL = 250 // 共享显示时钟，避免每行单独动画
 const globalAutoRefresh = ref(false) // 全局自动刷新开关（默认关闭）
 const isPageVisible = ref(typeof document === 'undefined' ? true : !document.hidden)
@@ -445,10 +451,14 @@ const discoveredActiveRequestIds = new Set<string>()
 async function loadActiveRequestUpdates(ids?: string[]) {
  if (isAdminPage.value) {
-    return usageApi.getActiveRequests(ids, timeRange.value)
+    const result = await usageApi.getActiveRequests(ids, timeRange.value)
    updateServerClockOffset(result.server_timing)
    return result
  }
  const idsParam = ids?.length ? ids.join(',') : undefined
-  return meApi.getActiveRequests(idsParam)
+  const result = await meApi.getActiveRequests(idsParam)
  updateServerClockOffset(result.server_timing)
  return result
 }
 async function pollActiveRequests() {
@@ -714,8 +724,10 @@ function handleVisibilityChange() {
    stopAutoRefresh()
    stopActiveDiscovery()
    stopGlobalAutoRefresh()
    stopActiveElapsedDisplayTimer()
    return
  }
  syncActiveElapsedDisplayTimer()
  if (hasActiveRequests.value) {
    startAutoRefresh()
  }
@@ -732,6 +744,7 @@ onUnmounted(() => {
  stopAutoRefresh()
  stopActiveDiscovery()
  stopGlobalAutoRefresh()
  stopActiveElapsedDisplayClock()
 })
 // 用户页面的前端分页（后端一次性返回所有记录，前端分页+筛选）
@@ -755,6 +768,20 @@ const effectiveTotalRecords = computed(() => {
 // 显示的记录
 const displayRecords = computed(() => paginatedRecords.value)
 const {
  calibratedDisplayNowMs,
  stopActiveElapsedDisplayTimer,
  syncActiveElapsedDisplayTimer,
  stopActiveElapsedDisplayClock,
 } = useActiveElapsedDisplayClock({
  records: displayRecords,
  isPageVisible,
  serverClockOffsetMs,
  hasServerClockOffset,
  resolveStatus: resolveDisplayRequestStatus,
  intervalMs: ACTIVE_ELAPSED_DISPLAY_INTERVAL,
 })
 const availableClientFamilies = computed(() => {
  const families = new Set<string>()
  currentRecords.value.forEach((record) => {