dsh-token-total 1.1.2 → 1.1.4
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +5 -1
- package/lib/client.js +73 -26
- package/lib/index.js +166 -180
- package/lib/types/client/locales.d.ts +1 -1
- package/lib/types/client/session-usage-locales.d.ts +2 -0
- package/lib/types/session-tree.d.ts +16 -0
- package/lib/types/session-usage.d.ts +5 -33
- package/lib/types/wire.d.ts +16 -0
- package/package.json +1 -1
- package/src/client/SessionUsageView.tsx +6 -1
- package/src/client/StatsLineEnhanced.tsx +69 -38
- package/src/client/locales.ts +8 -0
- package/src/client/session-usage-core.ts +1 -0
- package/src/client/session-usage-locales.ts +3 -0
- package/src/session-detail.ts +7 -36
- package/src/session-tree.ts +146 -0
- package/src/session-usage.ts +32 -181
- package/src/wire.ts +18 -0
package/lib/index.js
CHANGED
|
@@ -1526,6 +1526,134 @@ async function resolveModelNames(modelRefs, llm) {
|
|
|
1526
1526
|
return names;
|
|
1527
1527
|
}
|
|
1528
1528
|
//#endregion
|
|
1529
|
+
//#region src/session-tree.ts
|
|
1530
|
+
function headerOf(item) {
|
|
1531
|
+
if (!item || typeof item !== "object") return null;
|
|
1532
|
+
const header = "header" in item ? item.header : item;
|
|
1533
|
+
return header && typeof header.id === "string" && header.id !== "" ? header : null;
|
|
1534
|
+
}
|
|
1535
|
+
function inheritedCount(value) {
|
|
1536
|
+
return typeof value === "number" && Number.isSafeInteger(value) && value >= 0 ? value : 0;
|
|
1537
|
+
}
|
|
1538
|
+
function throwIfCanceled(error, signal) {
|
|
1539
|
+
signal?.throwIfAborted();
|
|
1540
|
+
if (error && typeof error === "object" && "name" in error && error.name === "AbortError") throw error;
|
|
1541
|
+
}
|
|
1542
|
+
/** Never send raw exceptions to the client: they may contain paths or secrets. */
|
|
1543
|
+
function readIssue(sessionId, error) {
|
|
1544
|
+
const name = error && typeof error === "object" && "name" in error ? error.name : "";
|
|
1545
|
+
const code = error && typeof error === "object" && "code" in error ? error.code : "";
|
|
1546
|
+
return {
|
|
1547
|
+
sessionId,
|
|
1548
|
+
reason: name === "SessionPersistenceCorruptionError" || name === "SessionFormatError" ? "corrupt" : name === "SessionPersistenceNotFoundError" || code === "ENOENT" ? "missing" : "unavailable"
|
|
1549
|
+
};
|
|
1550
|
+
}
|
|
1551
|
+
async function readSession(persistence, id, header, live, signal) {
|
|
1552
|
+
signal?.throwIfAborted();
|
|
1553
|
+
if (typeof live?.snapshotEvents === "function" && typeof live.inheritedEventCount === "number") return {
|
|
1554
|
+
id,
|
|
1555
|
+
header,
|
|
1556
|
+
live,
|
|
1557
|
+
events: live.snapshotEvents(),
|
|
1558
|
+
inheritedCut: inheritedCount(live.inheritedEventCount)
|
|
1559
|
+
};
|
|
1560
|
+
if (typeof persistence.open === "function") {
|
|
1561
|
+
const options = signal === void 0 ? void 0 : { signal };
|
|
1562
|
+
const handle = await persistence.open(id, "read", options);
|
|
1563
|
+
try {
|
|
1564
|
+
const events = [];
|
|
1565
|
+
for (let offset = 0;;) {
|
|
1566
|
+
signal?.throwIfAborted();
|
|
1567
|
+
const page = await handle.read(offset, 500, options);
|
|
1568
|
+
if (!page.events.length) break;
|
|
1569
|
+
for (const event of page.events) events.push(event);
|
|
1570
|
+
offset += page.events.length;
|
|
1571
|
+
}
|
|
1572
|
+
return {
|
|
1573
|
+
id,
|
|
1574
|
+
header: header ?? handle.header,
|
|
1575
|
+
live,
|
|
1576
|
+
events,
|
|
1577
|
+
inheritedCut: inheritedCount(handle.inheritedEventCount)
|
|
1578
|
+
};
|
|
1579
|
+
} finally {
|
|
1580
|
+
await handle.close().catch(() => {});
|
|
1581
|
+
}
|
|
1582
|
+
}
|
|
1583
|
+
if (typeof persistence.inspect === "function") {
|
|
1584
|
+
const inspection = await persistence.inspect(id, signal);
|
|
1585
|
+
signal?.throwIfAborted();
|
|
1586
|
+
return {
|
|
1587
|
+
id,
|
|
1588
|
+
header: header ?? inspection.meta,
|
|
1589
|
+
live,
|
|
1590
|
+
events: inspection.events,
|
|
1591
|
+
inheritedCut: inheritedCount(inspection.inheritedEventCount)
|
|
1592
|
+
};
|
|
1593
|
+
}
|
|
1594
|
+
throw new Error("no readable persistence seam (need open or inspect)");
|
|
1595
|
+
}
|
|
1596
|
+
async function readUsageTree(persistence, sessions, rootId, signal) {
|
|
1597
|
+
signal?.throwIfAborted();
|
|
1598
|
+
const rows = await persistence.list(typeof persistence.open === "function" ? signal === void 0 ? void 0 : { signal } : signal);
|
|
1599
|
+
const headers = /* @__PURE__ */ new Map();
|
|
1600
|
+
for (const row of rows) {
|
|
1601
|
+
const header = headerOf(row);
|
|
1602
|
+
if (header) headers.set(header.id, header);
|
|
1603
|
+
}
|
|
1604
|
+
const live = new Map((sessions?.list() ?? []).map((handle) => [handle.id, handle]));
|
|
1605
|
+
const children = /* @__PURE__ */ new Map();
|
|
1606
|
+
const link = (id, parent) => {
|
|
1607
|
+
if (typeof parent !== "string" || parent === "" || parent === id) return;
|
|
1608
|
+
const bucket = children.get(parent) ?? /* @__PURE__ */ new Set();
|
|
1609
|
+
bucket.add(id);
|
|
1610
|
+
children.set(parent, bucket);
|
|
1611
|
+
};
|
|
1612
|
+
for (const header of headers.values()) link(header.id, header.parentSession);
|
|
1613
|
+
for (const handle of live.values()) link(handle.id, handle.header?.parentSession ?? handle.parentSession);
|
|
1614
|
+
const ids = [];
|
|
1615
|
+
const seen = /* @__PURE__ */ new Set([rootId]);
|
|
1616
|
+
const queue = [rootId];
|
|
1617
|
+
for (let index = 0; index < queue.length; index++) for (const id of children.get(queue[index]) ?? []) {
|
|
1618
|
+
if (seen.has(id)) continue;
|
|
1619
|
+
seen.add(id);
|
|
1620
|
+
ids.push(id);
|
|
1621
|
+
queue.push(id);
|
|
1622
|
+
}
|
|
1623
|
+
const root = await readSession(persistence, rootId, headers.get(rootId), live.get(rootId), signal);
|
|
1624
|
+
const logs = new Array(ids.length);
|
|
1625
|
+
const issues = new Array(ids.length);
|
|
1626
|
+
let next = 0;
|
|
1627
|
+
const workers = await Promise.allSettled(Array.from({ length: Math.min(4, ids.length) }, async () => {
|
|
1628
|
+
for (;;) {
|
|
1629
|
+
signal?.throwIfAborted();
|
|
1630
|
+
const index = next++;
|
|
1631
|
+
if (index >= ids.length) return;
|
|
1632
|
+
const id = ids[index];
|
|
1633
|
+
try {
|
|
1634
|
+
logs[index] = await readSession(persistence, id, headers.get(id), live.get(id), signal);
|
|
1635
|
+
} catch (error) {
|
|
1636
|
+
throwIfCanceled(error, signal);
|
|
1637
|
+
issues[index] = readIssue(id, error);
|
|
1638
|
+
}
|
|
1639
|
+
}
|
|
1640
|
+
}));
|
|
1641
|
+
for (const worker of workers) if (worker.status === "rejected") throw worker.reason;
|
|
1642
|
+
signal?.throwIfAborted();
|
|
1643
|
+
const descendants = logs.filter((log) => log !== void 0);
|
|
1644
|
+
const unreadableSessions = issues.filter((issue) => issue !== void 0);
|
|
1645
|
+
return {
|
|
1646
|
+
root,
|
|
1647
|
+
descendants,
|
|
1648
|
+
coverage: {
|
|
1649
|
+
complete: unreadableSessions.length === 0,
|
|
1650
|
+
totalSessions: ids.length + 1,
|
|
1651
|
+
readableSessions: descendants.length + 1,
|
|
1652
|
+
unreadableSessions
|
|
1653
|
+
}
|
|
1654
|
+
};
|
|
1655
|
+
}
|
|
1656
|
+
//#endregion
|
|
1529
1657
|
//#region src/context-window.ts
|
|
1530
1658
|
/** ASCII-ish characters per token (the host estimator's ratio). */
|
|
1531
1659
|
const ASCII_CHARS_PER_TOKEN = 4;
|
|
@@ -1995,7 +2123,7 @@ function mergeUsageCompositions(parts) {
|
|
|
1995
2123
|
}
|
|
1996
2124
|
//#endregion
|
|
1997
2125
|
//#region src/session-detail.ts
|
|
1998
|
-
const zero = () => ({
|
|
2126
|
+
const zero$1 = () => ({
|
|
1999
2127
|
uncachedInputTokens: 0,
|
|
2000
2128
|
outputTokens: 0,
|
|
2001
2129
|
cacheReadTokens: 0,
|
|
@@ -2082,43 +2210,6 @@ function usageOutputTokens(x) {
|
|
|
2082
2210
|
const value = (x?.usage ?? x)?.outputTokens;
|
|
2083
2211
|
return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : null;
|
|
2084
2212
|
}
|
|
2085
|
-
function headerOf$1(item) {
|
|
2086
|
-
const x = item;
|
|
2087
|
-
if (x?.header?.id && typeof x.header.id === "string") return x.header;
|
|
2088
|
-
return typeof x?.id === "string" && x.id ? x : null;
|
|
2089
|
-
}
|
|
2090
|
-
async function read(p, id, live) {
|
|
2091
|
-
if (live?.snapshotEvents && typeof live.inheritedEventCount === "number") return {
|
|
2092
|
-
events: live.snapshotEvents(),
|
|
2093
|
-
cut: Math.max(0, live.inheritedEventCount)
|
|
2094
|
-
};
|
|
2095
|
-
if (p.open) {
|
|
2096
|
-
const h = await p.open(id, "read");
|
|
2097
|
-
const out = [];
|
|
2098
|
-
try {
|
|
2099
|
-
for (let off = 0;;) {
|
|
2100
|
-
const page = await h.read(off, 500);
|
|
2101
|
-
if (!page.events.length) break;
|
|
2102
|
-
out.push(...page.events);
|
|
2103
|
-
off += page.events.length;
|
|
2104
|
-
}
|
|
2105
|
-
} finally {
|
|
2106
|
-
await h.close().catch(() => {});
|
|
2107
|
-
}
|
|
2108
|
-
return {
|
|
2109
|
-
events: out,
|
|
2110
|
-
cut: Math.max(0, Number(h.inheritedEventCount) || 0)
|
|
2111
|
-
};
|
|
2112
|
-
}
|
|
2113
|
-
if (p.inspect) {
|
|
2114
|
-
const x = await p.inspect(id);
|
|
2115
|
-
return {
|
|
2116
|
-
events: x.events,
|
|
2117
|
-
cut: Math.max(0, Number(x.inheritedEventCount) || 0)
|
|
2118
|
-
};
|
|
2119
|
-
}
|
|
2120
|
-
throw new Error("no readable persistence seam (need open or inspect)");
|
|
2121
|
-
}
|
|
2122
2213
|
function nameOf(h, id) {
|
|
2123
2214
|
const x = h;
|
|
2124
2215
|
if (typeof x?.name === "string" && x.name) return x.name;
|
|
@@ -2126,10 +2217,10 @@ function nameOf(h, id) {
|
|
|
2126
2217
|
if (typeof x?.cwd === "string" && x.cwd) return x.cwd.split(/[\\/]/).filter(Boolean).pop() || id;
|
|
2127
2218
|
return id;
|
|
2128
2219
|
}
|
|
2129
|
-
|
|
2130
|
-
const { events, cut } =
|
|
2220
|
+
function one(log) {
|
|
2221
|
+
const { id, header, live, events, inheritedCut: cut } = log;
|
|
2131
2222
|
const fold = new UsageFold();
|
|
2132
|
-
const totals = zero();
|
|
2223
|
+
const totals = zero$1();
|
|
2133
2224
|
const records = [];
|
|
2134
2225
|
const byKey = /* @__PURE__ */ new Map();
|
|
2135
2226
|
let route = live?.requestContext?.() ?? {};
|
|
@@ -2205,14 +2296,14 @@ async function one(p, id, header, live) {
|
|
|
2205
2296
|
provider: r.provider || route.provider || "",
|
|
2206
2297
|
model: r.model || route.model || "",
|
|
2207
2298
|
status: "running",
|
|
2208
|
-
totals: zero(),
|
|
2299
|
+
totals: zero$1(),
|
|
2209
2300
|
firstTokenMs: null,
|
|
2210
2301
|
decodeMs: null,
|
|
2211
2302
|
tokensPerSecond: null,
|
|
2212
2303
|
turnName: "",
|
|
2213
2304
|
effort,
|
|
2214
2305
|
_attempt: ev.type === "llm/retry-started" ? 1 : 0,
|
|
2215
|
-
_lastUsage: zero()
|
|
2306
|
+
_lastUsage: zero$1()
|
|
2216
2307
|
};
|
|
2217
2308
|
records.push(rec);
|
|
2218
2309
|
byKey.set(key, rec);
|
|
@@ -2232,14 +2323,14 @@ async function one(p, id, header, live) {
|
|
|
2232
2323
|
provider: r.provider || route.provider || "",
|
|
2233
2324
|
model: r.model || route.model || "",
|
|
2234
2325
|
status: "running",
|
|
2235
|
-
totals: zero(),
|
|
2326
|
+
totals: zero$1(),
|
|
2236
2327
|
firstTokenMs: null,
|
|
2237
2328
|
decodeMs: null,
|
|
2238
2329
|
tokensPerSecond: null,
|
|
2239
2330
|
turnName: "",
|
|
2240
2331
|
effort,
|
|
2241
2332
|
_attempt: 0,
|
|
2242
|
-
_lastUsage: zero()
|
|
2333
|
+
_lastUsage: zero$1()
|
|
2243
2334
|
};
|
|
2244
2335
|
records.push(rec);
|
|
2245
2336
|
byKey.set(key, rec);
|
|
@@ -2326,32 +2417,16 @@ async function one(p, id, header, live) {
|
|
|
2326
2417
|
};
|
|
2327
2418
|
}
|
|
2328
2419
|
async function buildSessionUsageDetail(persistence, sessions, rootId, ctx) {
|
|
2329
|
-
const
|
|
2330
|
-
const
|
|
2331
|
-
const
|
|
2332
|
-
for (const h of headers) if (h.parentSession) children.set(h.parentSession, [...children.get(h.parentSession) ?? [], h.id]);
|
|
2333
|
-
for (const h of live.values()) {
|
|
2334
|
-
const p = h.header?.parentSession ?? h.parentSession;
|
|
2335
|
-
if (p) children.set(p, [...(children.get(p) ?? []).filter((x) => x !== h.id), h.id]);
|
|
2336
|
-
}
|
|
2337
|
-
const ids = [];
|
|
2338
|
-
const seen = /* @__PURE__ */ new Set([rootId]);
|
|
2339
|
-
const q = [rootId];
|
|
2340
|
-
while (q.length) for (const c of children.get(q.shift()) ?? []) if (!seen.has(c)) {
|
|
2341
|
-
seen.add(c);
|
|
2342
|
-
ids.push(c);
|
|
2343
|
-
q.push(c);
|
|
2344
|
-
}
|
|
2345
|
-
const all = [rootId, ...ids];
|
|
2346
|
-
const parts = await Promise.all(all.map((id) => one(persistence, id, headers.find((h) => h.id === id), live.get(id))));
|
|
2347
|
-
const totals = zero();
|
|
2420
|
+
const { root, descendants, coverage } = await readUsageTree(persistence, sessions, rootId);
|
|
2421
|
+
const parts = [root, ...descendants].map(one);
|
|
2422
|
+
const totals = zero$1();
|
|
2348
2423
|
const models = /* @__PURE__ */ new Map();
|
|
2349
2424
|
const records = parts.flatMap((x) => x.records);
|
|
2350
2425
|
for (const p of parts) add$1(totals, p.totals);
|
|
2351
2426
|
for (const r of records) {
|
|
2352
2427
|
const key = r.model || "(unknown)";
|
|
2353
2428
|
const m = models.get(key) ?? {
|
|
2354
|
-
totals: zero(),
|
|
2429
|
+
totals: zero$1(),
|
|
2355
2430
|
requests: 0,
|
|
2356
2431
|
providers: /* @__PURE__ */ new Set()
|
|
2357
2432
|
};
|
|
@@ -2405,6 +2480,7 @@ async function buildSessionUsageDetail(persistence, sessions, rootId, ctx) {
|
|
|
2405
2480
|
const firsts = records.filter((x) => x.status === "completed").map((x) => x.firstTokenMs).filter((x) => x !== null);
|
|
2406
2481
|
return {
|
|
2407
2482
|
sessionId: rootId,
|
|
2483
|
+
coverage,
|
|
2408
2484
|
totals,
|
|
2409
2485
|
requests: parts.reduce((n, p) => n + p.requests, 0),
|
|
2410
2486
|
averageFirstTokenMs: firsts.length ? firsts.reduce((a, b) => a + b, 0) / firsts.length : null,
|
|
@@ -2811,82 +2887,25 @@ function buildUsageRoute(ctx, deps) {
|
|
|
2811
2887
|
}
|
|
2812
2888
|
//#endregion
|
|
2813
2889
|
//#region src/session-usage.ts
|
|
2814
|
-
const
|
|
2890
|
+
const zero = () => ({
|
|
2815
2891
|
uncachedInputTokens: 0,
|
|
2816
2892
|
outputTokens: 0,
|
|
2817
2893
|
cacheReadTokens: 0,
|
|
2818
2894
|
cacheWriteTokens: 0
|
|
2819
|
-
};
|
|
2895
|
+
});
|
|
2820
2896
|
function add(into, from) {
|
|
2821
2897
|
into.uncachedInputTokens += from.uncachedInputTokens;
|
|
2822
2898
|
into.outputTokens += from.outputTokens;
|
|
2823
2899
|
into.cacheReadTokens += from.cacheReadTokens;
|
|
2824
2900
|
into.cacheWriteTokens += from.cacheWriteTokens;
|
|
2825
2901
|
}
|
|
2826
|
-
|
|
2827
|
-
* and a live session's numbers move at step granularity, not per frame. */
|
|
2828
|
-
const CACHE_TTL_MS = 4e3;
|
|
2829
|
-
/** Absolute ceiling on cached trees, so a long-lived host with many sessions
|
|
2830
|
-
* cannot grow this map without bound. */
|
|
2831
|
-
const CACHE_MAX = 64;
|
|
2832
|
-
/** Normalize a `list()` row (alpha snapshots or rc.1 header rows). */
|
|
2833
|
-
function headerOf(item) {
|
|
2834
|
-
const snap = item;
|
|
2835
|
-
if (snap !== null && typeof snap === "object" && "header" in snap && snap.header !== null && typeof snap.header === "object" && typeof snap.header.id === "string") return snap.header;
|
|
2836
|
-
const row = item;
|
|
2837
|
-
return typeof row?.id === "string" && row.id !== "" ? row : null;
|
|
2838
|
-
}
|
|
2839
|
-
function inheritedCount(cut) {
|
|
2840
|
-
return typeof cut === "number" && Number.isSafeInteger(cut) && cut >= 0 ? cut : 0;
|
|
2841
|
-
}
|
|
2842
|
-
/** Read one stored session's events through whichever persistence seam the
|
|
2843
|
-
* host line exposes (alpha open/read/close wins; rc.1 inspect is the tail). */
|
|
2844
|
-
async function readEvents(persistence, id, signal) {
|
|
2845
|
-
if (typeof persistence.open === "function") {
|
|
2846
|
-
const handle = await persistence.open(id, "read", signal === void 0 ? void 0 : { signal });
|
|
2847
|
-
try {
|
|
2848
|
-
const inheritedCut = inheritedCount(handle.inheritedEventCount);
|
|
2849
|
-
const collected = [];
|
|
2850
|
-
let offset = 0;
|
|
2851
|
-
const PAGE = 500;
|
|
2852
|
-
for (;;) {
|
|
2853
|
-
signal?.throwIfAborted();
|
|
2854
|
-
const slice = await handle.read(offset, PAGE, signal === void 0 ? void 0 : { signal });
|
|
2855
|
-
const page = slice && slice.events ? slice.events : [];
|
|
2856
|
-
if (page.length === 0) break;
|
|
2857
|
-
for (const ev of page) collected.push(ev);
|
|
2858
|
-
offset += page.length;
|
|
2859
|
-
}
|
|
2860
|
-
return {
|
|
2861
|
-
events: collected,
|
|
2862
|
-
inheritedCut
|
|
2863
|
-
};
|
|
2864
|
-
} finally {
|
|
2865
|
-
await handle.close().catch(() => {});
|
|
2866
|
-
}
|
|
2867
|
-
}
|
|
2868
|
-
if (typeof persistence.inspect === "function") {
|
|
2869
|
-
const inspection = await persistence.inspect(id, signal);
|
|
2870
|
-
return {
|
|
2871
|
-
events: inspection.events,
|
|
2872
|
-
inheritedCut: inheritedCount(inspection.inheritedEventCount)
|
|
2873
|
-
};
|
|
2874
|
-
}
|
|
2875
|
-
throw new Error("no readable persistence seam (need open or inspect)");
|
|
2876
|
-
}
|
|
2877
|
-
/** Sum one session's own usage (its main agent) from its log. */
|
|
2878
|
-
async function sumOneSession(persistence, id, signal, live) {
|
|
2879
|
-
signal?.throwIfAborted();
|
|
2880
|
-
const { events, inheritedCut } = typeof live?.snapshotEvents === "function" && typeof live.inheritedEventCount === "number" ? {
|
|
2881
|
-
events: live.snapshotEvents(),
|
|
2882
|
-
inheritedCut: inheritedCount(live.inheritedEventCount)
|
|
2883
|
-
} : await readEvents(persistence, id, signal);
|
|
2902
|
+
function sumOneSession(log) {
|
|
2884
2903
|
const fold = new UsageFold();
|
|
2885
|
-
const totals =
|
|
2886
|
-
const compaction =
|
|
2887
|
-
for (const
|
|
2888
|
-
if (
|
|
2889
|
-
const sample = fold.fold(
|
|
2904
|
+
const totals = zero();
|
|
2905
|
+
const compaction = zero();
|
|
2906
|
+
for (const event of log.events) {
|
|
2907
|
+
if (event.seq < log.inheritedCut) continue;
|
|
2908
|
+
const sample = fold.fold(event);
|
|
2890
2909
|
if (sample === null || sample.turn) continue;
|
|
2891
2910
|
const buckets = {
|
|
2892
2911
|
uncachedInputTokens: sample.inputTokens,
|
|
@@ -2895,77 +2914,47 @@ async function sumOneSession(persistence, id, signal, live) {
|
|
|
2895
2914
|
cacheWriteTokens: sample.cacheWriteTokens
|
|
2896
2915
|
};
|
|
2897
2916
|
add(totals, buckets);
|
|
2898
|
-
if (
|
|
2917
|
+
if (event.type === "compaction/summary") add(compaction, buckets);
|
|
2899
2918
|
}
|
|
2900
2919
|
return {
|
|
2901
2920
|
totals,
|
|
2902
2921
|
compaction
|
|
2903
2922
|
};
|
|
2904
2923
|
}
|
|
2924
|
+
const CACHE_TTL_MS = 4e3;
|
|
2925
|
+
const CACHE_MAX = 64;
|
|
2905
2926
|
var SessionUsageService = class {
|
|
2906
|
-
/** Short-lived tree reads; any live event can invalidate tree membership. */
|
|
2907
2927
|
cache = /* @__PURE__ */ new Map();
|
|
2908
|
-
/** Any event can introduce a new descendant while an async read is running. */
|
|
2909
2928
|
generation = 0;
|
|
2910
|
-
/**
|
|
2911
|
-
* of ANY cached root, so membership-only invalidation is not safe. */
|
|
2929
|
+
/** New events can introduce descendants of any cached root. */
|
|
2912
2930
|
invalidate(_sessionId) {
|
|
2913
2931
|
this.generation++;
|
|
2914
2932
|
this.cache.clear();
|
|
2915
2933
|
}
|
|
2916
|
-
/**
|
|
2917
|
-
*
|
|
2918
|
-
*
|
|
2919
|
-
* callers can retain their last good display while surfacing the error. */
|
|
2934
|
+
/** A bad descendant leaves the readable totals available with coverage.
|
|
2935
|
+
* Never cache incomplete results: a repaired log must recover on the next read.
|
|
2936
|
+
* Root, catalog and cancellation failures still reject without publishing zero. */
|
|
2920
2937
|
async tree(persistence, sessions, rootId, signal) {
|
|
2921
2938
|
signal?.throwIfAborted();
|
|
2922
2939
|
const hit = this.cache.get(rootId);
|
|
2923
2940
|
if (hit !== void 0 && Date.now() - hit.at < CACHE_TTL_MS) return hit.value;
|
|
2924
2941
|
const generation = this.generation;
|
|
2925
|
-
const
|
|
2926
|
-
const
|
|
2927
|
-
|
|
2928
|
-
|
|
2929
|
-
|
|
2930
|
-
const bucket = children.get(parent);
|
|
2931
|
-
if (bucket === void 0) children.set(parent, [header.id]);
|
|
2932
|
-
else bucket.push(header.id);
|
|
2933
|
-
}
|
|
2934
|
-
const live = new Map((sessions?.list() ?? []).map((handle) => [handle.id, handle]));
|
|
2935
|
-
for (const handle of live.values()) {
|
|
2936
|
-
const parent = handle.header?.parentSession ?? handle.parentSession;
|
|
2937
|
-
if (typeof parent !== "string" || parent === "" || parent === handle.id) continue;
|
|
2938
|
-
const bucket = children.get(parent);
|
|
2939
|
-
if (bucket === void 0) children.set(parent, [handle.id]);
|
|
2940
|
-
else if (!bucket.includes(handle.id)) bucket.push(handle.id);
|
|
2941
|
-
}
|
|
2942
|
-
const descendants = [];
|
|
2943
|
-
const seen = /* @__PURE__ */ new Set([rootId]);
|
|
2944
|
-
const queue = [rootId];
|
|
2945
|
-
while (queue.length > 0) {
|
|
2946
|
-
const current = queue.shift();
|
|
2947
|
-
for (const child of children.get(current) ?? []) {
|
|
2948
|
-
if (seen.has(child)) continue;
|
|
2949
|
-
seen.add(child);
|
|
2950
|
-
descendants.push(child);
|
|
2951
|
-
queue.push(child);
|
|
2952
|
-
}
|
|
2953
|
-
}
|
|
2954
|
-
const root = await sumOneSession(persistence, rootId, signal, live.get(rootId));
|
|
2955
|
-
const self = root.totals;
|
|
2956
|
-
const subagents = { ...ZERO };
|
|
2957
|
-
for (const id of descendants) add(subagents, (await sumOneSession(persistence, id, signal, live.get(id))).totals);
|
|
2958
|
-
const total = { ...self };
|
|
2942
|
+
const { root, descendants, coverage } = await readUsageTree(persistence, sessions, rootId, signal);
|
|
2943
|
+
const own = sumOneSession(root);
|
|
2944
|
+
const subagents = zero();
|
|
2945
|
+
for (const log of descendants) add(subagents, sumOneSession(log).totals);
|
|
2946
|
+
const total = { ...own.totals };
|
|
2959
2947
|
add(total, subagents);
|
|
2960
2948
|
const value = {
|
|
2961
|
-
self,
|
|
2962
|
-
selfCompaction:
|
|
2949
|
+
self: own.totals,
|
|
2950
|
+
selfCompaction: own.compaction,
|
|
2963
2951
|
subagents,
|
|
2964
2952
|
total,
|
|
2965
|
-
subagentSessions: descendants.length
|
|
2953
|
+
subagentSessions: descendants.length,
|
|
2954
|
+
coverage
|
|
2966
2955
|
};
|
|
2967
2956
|
signal?.throwIfAborted();
|
|
2968
|
-
if (generation === this.generation) {
|
|
2957
|
+
if (coverage.complete && generation === this.generation) {
|
|
2969
2958
|
if (this.cache.size >= CACHE_MAX) this.cache.clear();
|
|
2970
2959
|
this.cache.set(rootId, {
|
|
2971
2960
|
at: Date.now(),
|
|
@@ -2974,9 +2963,6 @@ var SessionUsageService = class {
|
|
|
2974
2963
|
}
|
|
2975
2964
|
return value;
|
|
2976
2965
|
}
|
|
2977
|
-
async listHeaders(persistence, signal) {
|
|
2978
|
-
return (typeof persistence.open === "function" ? await persistence.list(signal === void 0 ? void 0 : { signal }) : await persistence.list(signal)).map(headerOf).filter((h) => h !== null);
|
|
2979
|
-
}
|
|
2980
2966
|
};
|
|
2981
2967
|
//#endregion
|
|
2982
2968
|
//#region src/index.ts
|
|
@@ -7,7 +7,7 @@
|
|
|
7
7
|
/** The locale namespace this plugin owns. */
|
|
8
8
|
export declare const LOCALE_NS = "tokenTotal";
|
|
9
9
|
/** The typed translator key union for the panel's locale seat. */
|
|
10
|
-
export type TokenTotalKey = 'nav' | 'back' | 'range' | 'rangePreset.today' | 'rangePreset.3' | 'rangePreset.7' | 'rangePreset.30' | 'rangePreset.90' | 'rangeCustom' | 'from' | 'to' | 'refresh' | 'loading' | 'tokens' | 'tokensHint' | 'tokensUnit' | 'inputTokens' | 'outputTokens' | 'cacheTokens' | 'cachedTokens' | 'sessions' | 'requests' | 'requestsFailed' | 'activeDays' | 'cacheRate' | 'cacheRateHint' | 'cacheHitRate' | 'hitRateLegend' | 'topModel' | 'topModelHint' | 'heatmap' | 'heatLess' | 'heatMore' | 'dow.mon' | 'dow.tue' | 'dow.wed' | 'dow.thu' | 'dow.fri' | 'dow.sat' | 'dow.sun' | 'month.jan' | 'month.feb' | 'month.mar' | 'month.apr' | 'month.may' | 'month.jun' | 'month.jul' | 'month.aug' | 'month.sep' | 'month.oct' | 'month.nov' | 'month.dec' | 'dailyTrend' | 'hourlyTrend' | 'trendLimited' | 'usageRanking' | 'byModel' | 'byProvider' | 'tokensColumn' | 'cacheRateColumn' | 'providerModels' | 'other' | 'total' | 'percent' | 'asOf' | 'empty' | 'stats.counts' | 'stats.tokensPerSecond' | 'stats.cacheHit' | 'stats.cacheMiss' | 'stats.dialog.title' | 'stats.dialog.usageTitle' | 'stats.dialog.llmTime' | 'stats.dialog.toolTime' | 'stats.dialog.ttft' | 'stats.dialog.speed' | 'stats.dialog.main' | 'stats.dialog.subagents' | 'stats.dialog.subagentCount' | 'stats.dialog.treeTotal' | 'message.turnUsage.count' | 'message.turnUsage.cacheHit' | 'message.turnUsage.input' | 'message.turnUsage.cacheRead' | 'message.turnUsage.cacheWrite' | 'message.turnUsage.output';
|
|
10
|
+
export type TokenTotalKey = 'nav' | 'back' | 'range' | 'rangePreset.today' | 'rangePreset.3' | 'rangePreset.7' | 'rangePreset.30' | 'rangePreset.90' | 'rangeCustom' | 'from' | 'to' | 'refresh' | 'loading' | 'tokens' | 'tokensHint' | 'tokensUnit' | 'inputTokens' | 'outputTokens' | 'cacheTokens' | 'cachedTokens' | 'sessions' | 'requests' | 'requestsFailed' | 'activeDays' | 'cacheRate' | 'cacheRateHint' | 'cacheHitRate' | 'hitRateLegend' | 'topModel' | 'topModelHint' | 'heatmap' | 'heatLess' | 'heatMore' | 'dow.mon' | 'dow.tue' | 'dow.wed' | 'dow.thu' | 'dow.fri' | 'dow.sat' | 'dow.sun' | 'month.jan' | 'month.feb' | 'month.mar' | 'month.apr' | 'month.may' | 'month.jun' | 'month.jul' | 'month.aug' | 'month.sep' | 'month.oct' | 'month.nov' | 'month.dec' | 'dailyTrend' | 'hourlyTrend' | 'trendLimited' | 'usageRanking' | 'byModel' | 'byProvider' | 'tokensColumn' | 'cacheRateColumn' | 'providerModels' | 'other' | 'total' | 'percent' | 'asOf' | 'empty' | 'stats.counts' | 'stats.tokensPerSecond' | 'stats.cacheHit' | 'stats.cacheMiss' | 'stats.dialog.title' | 'stats.dialog.usageTitle' | 'stats.partial' | 'stats.incomplete' | 'stats.dialog.llmTime' | 'stats.dialog.toolTime' | 'stats.dialog.ttft' | 'stats.dialog.speed' | 'stats.dialog.main' | 'stats.dialog.subagents' | 'stats.dialog.subagentCount' | 'stats.dialog.treeTotal' | 'message.turnUsage.count' | 'message.turnUsage.cacheHit' | 'message.turnUsage.input' | 'message.turnUsage.cacheRead' | 'message.turnUsage.cacheWrite' | 'message.turnUsage.output';
|
|
11
11
|
export declare const en: Record<TokenTotalKey, string>;
|
|
12
12
|
export declare const zh: Record<TokenTotalKey, string>;
|
|
13
13
|
export declare const zhTW: Record<TokenTotalKey, string>;
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
import type { UsageSessionEvent, UsageSessionHandle, UsageSessionHeader, UsageSessionPersistence, UsageSessionStore } from './context-types.ts';
|
|
2
|
+
import type { SessionUsageCoverage } from './wire.ts';
|
|
3
|
+
/** Shared membership, ownership and failure semantics for both usage APIs.
|
|
4
|
+
* A bad descendant is isolated; root, catalog and cancellation failures reject. */
|
|
5
|
+
export interface UsageSessionLog {
|
|
6
|
+
id: string;
|
|
7
|
+
header?: UsageSessionHeader;
|
|
8
|
+
live?: UsageSessionHandle;
|
|
9
|
+
events: readonly UsageSessionEvent[];
|
|
10
|
+
inheritedCut: number;
|
|
11
|
+
}
|
|
12
|
+
export declare function readUsageTree(persistence: UsageSessionPersistence, sessions: UsageSessionStore | undefined, rootId: string, signal?: AbortSignal): Promise<{
|
|
13
|
+
root: UsageSessionLog;
|
|
14
|
+
descendants: UsageSessionLog[];
|
|
15
|
+
coverage: SessionUsageCoverage;
|
|
16
|
+
}>;
|
|
@@ -1,42 +1,14 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Per-session usage aggregation, INCLUDING subagent sessions.
|
|
3
|
-
*
|
|
4
|
-
* Why this exists: dsh's native `tokenUsage` projection covers only the main
|
|
5
|
-
* agent of the current session — a session that delegates to subagents (whose
|
|
6
|
-
* own logs are separate persisted sessions, linked by `header.parentSession`)
|
|
7
|
-
* shows only a fraction of what it actually spent. This module answers "what
|
|
8
|
-
* did this whole session tree cost".
|
|
9
|
-
*
|
|
10
|
-
* The store has no per-session dimension (it accumulates per day x model), so
|
|
11
|
-
* this is computed ON DEMAND from the session logs: the root session plus every
|
|
12
|
-
* descendant whose `parentSession` chain leads back to it, each replayed
|
|
13
|
-
* through a fresh {@link UsageFold} (the same dedupe rules the backfill uses).
|
|
14
|
-
*
|
|
15
|
-
* Two correctness rules carried over from the backfill engine:
|
|
16
|
-
* - a forked log begins with the events it copied from its parent
|
|
17
|
-
* (`handle.inheritedEventCount`); that prefix is skipped, the parent owns it;
|
|
18
|
-
* - the fold dedupes per (turn, step) within one session, so a streaming
|
|
19
|
-
* sample followed by the final message counts once.
|
|
20
|
-
*
|
|
21
|
-
* Results are cached briefly: the composer pill polls, and replaying a long log
|
|
22
|
-
* per poll would be wasteful. The cache is invalidated for a session as soon as
|
|
23
|
-
* a new event lands (see the collector's live subscription).
|
|
24
|
-
*/
|
|
1
|
+
/** Per-session owned usage, including every readable descendant. */
|
|
25
2
|
import type { UsageSessionPersistence, UsageSessionStore } from './context-types.ts';
|
|
26
3
|
import type { SessionUsageTree } from './wire.ts';
|
|
27
4
|
export type { SessionUsageTotals, SessionUsageTree } from './wire.ts';
|
|
28
5
|
export declare class SessionUsageService {
|
|
29
|
-
/** Short-lived tree reads; any live event can invalidate tree membership. */
|
|
30
6
|
private cache;
|
|
31
|
-
/** Any event can introduce a new descendant while an async read is running. */
|
|
32
7
|
private generation;
|
|
33
|
-
/**
|
|
34
|
-
* of ANY cached root, so membership-only invalidation is not safe. */
|
|
8
|
+
/** New events can introduce descendants of any cached root. */
|
|
35
9
|
invalidate(_sessionId: string): void;
|
|
36
|
-
/**
|
|
37
|
-
*
|
|
38
|
-
*
|
|
39
|
-
* callers can retain their last good display while surfacing the error. */
|
|
10
|
+
/** A bad descendant leaves the readable totals available with coverage.
|
|
11
|
+
* Never cache incomplete results: a repaired log must recover on the next read.
|
|
12
|
+
* Root, catalog and cancellation failures still reject without publishing zero. */
|
|
40
13
|
tree(persistence: UsageSessionPersistence, sessions: UsageSessionStore | undefined, rootId: string, signal?: AbortSignal): Promise<SessionUsageTree>;
|
|
41
|
-
private listHeaders;
|
|
42
14
|
}
|
package/lib/types/wire.d.ts
CHANGED
|
@@ -109,6 +109,20 @@ export interface SessionUsageTotals {
|
|
|
109
109
|
cacheReadTokens: number;
|
|
110
110
|
cacheWriteTokens: number;
|
|
111
111
|
}
|
|
112
|
+
/** A descendant that could not be read. Reasons are stable and sanitized;
|
|
113
|
+
* exception messages and filesystem paths never cross the wire. */
|
|
114
|
+
export interface SessionUsageReadIssue {
|
|
115
|
+
sessionId: string;
|
|
116
|
+
reason: 'corrupt' | 'missing' | 'unavailable';
|
|
117
|
+
}
|
|
118
|
+
/** A partial response is usable, but its totals are explicitly not the whole
|
|
119
|
+
* tree. The root is always readable; older hosts may omit this field. */
|
|
120
|
+
export interface SessionUsageCoverage {
|
|
121
|
+
complete: boolean;
|
|
122
|
+
totalSessions: number;
|
|
123
|
+
readableSessions: number;
|
|
124
|
+
unreadableSessions: SessionUsageReadIssue[];
|
|
125
|
+
}
|
|
112
126
|
/** What one session tree cost, split into the main agent and its subagents.
|
|
113
127
|
* The composer stats row renders this so a delegating session shows its whole
|
|
114
128
|
* spend (the native `tokenUsage` projection only covers the main agent). */
|
|
@@ -124,6 +138,7 @@ export interface SessionUsageTree {
|
|
|
124
138
|
total: SessionUsageTotals;
|
|
125
139
|
/** How many descendant sessions were counted. */
|
|
126
140
|
subagentSessions: number;
|
|
141
|
+
coverage?: SessionUsageCoverage;
|
|
127
142
|
}
|
|
128
143
|
/** Request body of POST /token-total/api/session. */
|
|
129
144
|
export interface SessionUsageRequest {
|
|
@@ -132,6 +147,7 @@ export interface SessionUsageRequest {
|
|
|
132
147
|
/** Detailed per-session usage view used by the Usage tab. */
|
|
133
148
|
export interface SessionUsageDetail {
|
|
134
149
|
sessionId: string;
|
|
150
|
+
coverage?: SessionUsageCoverage;
|
|
135
151
|
totals: SessionUsageTotals;
|
|
136
152
|
requests: number;
|
|
137
153
|
averageFirstTokenMs: number | null;
|
package/package.json
CHANGED
|
@@ -205,7 +205,12 @@ function Records({ detail, t }: { detail: SessionUsageDetail; t: T }): JSX.Eleme
|
|
|
205
205
|
/** Pure presentation export also used by offline visual regression fixtures. */
|
|
206
206
|
export function SessionUsageContent({ detail, t, updatedAt }: { detail: SessionUsageDetail; t: T; updatedAt?: number | null }): JSX.Element {
|
|
207
207
|
return <>
|
|
208
|
-
<p className={css.scope}>{t('scope')}</p
|
|
208
|
+
<p className={css.scope}>{t('scope')}</p>
|
|
209
|
+
{detail.coverage?.complete === false && <div className={css.error} role="status" data-session-usage-partial>
|
|
210
|
+
{t('partialCoverage', { count: String(detail.coverage.unreadableSessions.length) })}
|
|
211
|
+
<span title={detail.coverage.unreadableSessions.map(issue => issue.sessionId).join('\n')}>{t('partialTotals')}</span>
|
|
212
|
+
</div>}
|
|
213
|
+
<Metrics detail={detail} t={t} /><Tables detail={detail} t={t} />
|
|
209
214
|
{/* The two live cards sit between the per-subagent/model tables and the
|
|
210
215
|
request records: the left one describes what the next request carries
|
|
211
216
|
right now, the right one what the whole session spent, while everything
|