feat(omp): update context usage during active turns (#2503)
* feat: add DeepWiki MCP server at project level * feat(omp): poll context window usage during active turns OMP agents previously reported context window usage only at turn completion, unlike Claude and Codex which stream live updates. Add a periodic poll (3s interval) of getSessionStats() that emits usage_updated events while a turn is active, so the client context window meter updates in real-time. * clear context usage poll timer before scheduling next interval * chore: remove .omp/ directory * fix(omp): make live usage polling lifecycle-safe OMP exposes live usage through a stats request rather than stream events. Keep polling scoped to active turns, suppress unchanged snapshots, and invalidate stale completion reads across interrupts, new turns, and session shutdown. --------- Co-authored-by: Mohamed Boudra <boudra.moha@gmail.com>
This commit is contained in:
co-authored by
Mohamed Boudra
parent
5d15e40a2b
commit
2948a435ed
@@ -1,7 +1,9 @@
|
||||
import { describe, expect, test } from "vitest";
|
||||
import { setImmediate as waitForImmediate } from "node:timers/promises";
|
||||
|
||||
import type { PaseoToolCatalog } from "../../tools/types.js";
|
||||
import type { OmpNoTurnScheduler, OmpProviderIdleScheduler } from "./agent.js";
|
||||
import type { OmpUsagePollScheduler } from "./usage-poller.js";
|
||||
import { OmpHarness } from "./test-utils/omp-harness.js";
|
||||
|
||||
class ManualIdleScheduler implements OmpProviderIdleScheduler {
|
||||
@@ -65,6 +67,28 @@ class ManualNoTurnScheduler implements OmpNoTurnScheduler {
|
||||
}
|
||||
}
|
||||
|
||||
class ManualUsagePollScheduler implements OmpUsagePollScheduler {
|
||||
private readonly polls: Array<{ active: boolean; callback: () => void }> = [];
|
||||
|
||||
schedulePoll(callback: () => void): () => void {
|
||||
const poll = { active: true, callback };
|
||||
this.polls.push(poll);
|
||||
return () => {
|
||||
poll.active = false;
|
||||
};
|
||||
}
|
||||
|
||||
poll(): void {
|
||||
const poll = this.polls.shift();
|
||||
if (!poll) throw new Error("OMP has not scheduled a context usage poll");
|
||||
if (poll.active) poll.callback();
|
||||
}
|
||||
|
||||
activePollCount(): number {
|
||||
return this.polls.filter((poll) => poll.active).length;
|
||||
}
|
||||
}
|
||||
|
||||
function createToolCatalog(): PaseoToolCatalog {
|
||||
return {
|
||||
tools: new Map([
|
||||
@@ -209,6 +233,45 @@ describe("OMP agent client and session", () => {
|
||||
expect(omp.completedTurnCount()).toBe(1);
|
||||
});
|
||||
|
||||
test("starts and stops context usage polling with the active turn", async () => {
|
||||
const scheduler = new ManualUsagePollScheduler();
|
||||
const omp = new OmpHarness({ usagePollScheduler: scheduler });
|
||||
await omp.start();
|
||||
omp.runtime().stats = {
|
||||
contextUsage: { tokens: 130, contextWindow: 200_000 },
|
||||
};
|
||||
omp.runtime().state.contextUsage = { tokens: 99, contextWindow: 100_000 };
|
||||
await omp.requireStartTurn("keep working");
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
scheduler.poll();
|
||||
await waitForImmediate();
|
||||
expect(omp.usageUpdates()).toEqual([
|
||||
{
|
||||
inputTokens: 0,
|
||||
cachedInputTokens: 0,
|
||||
outputTokens: 0,
|
||||
totalCostUsd: 0,
|
||||
contextWindowMaxTokens: 200_000,
|
||||
contextWindowUsedTokens: 130,
|
||||
},
|
||||
]);
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
omp.runtime().abortError = new Error("abort unavailable");
|
||||
await expect(omp.interrupt()).rejects.toThrow("abort unavailable");
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
omp.runtime().abortError = null;
|
||||
await omp.interrupt();
|
||||
expect(scheduler.activePollCount()).toBe(0);
|
||||
|
||||
await omp.runPrompt("finish normally", "done");
|
||||
expect(scheduler.activePollCount()).toBe(0);
|
||||
|
||||
await omp.requireStartTurn("close the session");
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
await omp.close();
|
||||
expect(scheduler.activePollCount()).toBe(0);
|
||||
});
|
||||
|
||||
test("does not accept a follow-up until OMP reports stable idle", async () => {
|
||||
const omp = new OmpHarness();
|
||||
await omp.start();
|
||||
|
||||
@@ -27,7 +27,6 @@ import {
|
||||
type AgentSlashCommand,
|
||||
type AgentStreamEvent,
|
||||
type AgentTimelineItem,
|
||||
type AgentUsage,
|
||||
type FetchCatalogOptions,
|
||||
type ImportableProviderSession,
|
||||
type ImportProviderSessionContext,
|
||||
@@ -79,7 +78,6 @@ import type {
|
||||
OmpImageContent,
|
||||
OmpModel,
|
||||
OmpRuntimeEvent,
|
||||
OmpSessionStats,
|
||||
OmpSessionState,
|
||||
OmpThinkingLevel,
|
||||
} from "./rpc-types.js";
|
||||
@@ -102,7 +100,7 @@ import {
|
||||
} from "./host-tools.js";
|
||||
import { OmpSubagentIndex } from "./subagent-index.js";
|
||||
import { mapOmpToolDetail } from "./tool-call-mapper.js";
|
||||
import { mapOmpUsage } from "./usage-mapper.js";
|
||||
import { OmpUsagePoller, type OmpUsagePollScheduler } from "./usage-poller.js";
|
||||
import {
|
||||
buildOmpRpcUiPermissionResponse,
|
||||
mapOmpRpcUiPermissionRequest,
|
||||
@@ -137,6 +135,7 @@ export interface OmpAgentClientOptions {
|
||||
subagentCardScheduler?: OmpSubagentCardScheduler;
|
||||
providerIdleScheduler?: OmpProviderIdleScheduler;
|
||||
noTurnScheduler?: OmpNoTurnScheduler;
|
||||
usagePollScheduler?: OmpUsagePollScheduler;
|
||||
}
|
||||
|
||||
export interface OmpProviderIdleScheduler {
|
||||
@@ -184,6 +183,7 @@ interface OmpAgentSessionOptions {
|
||||
subagentCardScheduler?: OmpSubagentCardScheduler;
|
||||
providerIdleScheduler?: OmpProviderIdleScheduler;
|
||||
noTurnScheduler?: OmpNoTurnScheduler;
|
||||
usagePollScheduler?: OmpUsagePollScheduler;
|
||||
paseoTools?: PaseoToolCatalog;
|
||||
/**
|
||||
* When false (resumed sessions), replayed session events are dropped until
|
||||
@@ -305,35 +305,6 @@ function parseAutoCompactMode(value: string | undefined): AutoCompactMode {
|
||||
return "unknown";
|
||||
}
|
||||
|
||||
function toAgentUsage(stats: OmpSessionStats): AgentUsage | undefined {
|
||||
const inputTokens = stats.tokens?.input ?? 0;
|
||||
const cachedInputTokens = stats.tokens?.cacheRead ?? 0;
|
||||
const outputTokens = stats.tokens?.output ?? 0;
|
||||
const totalCostUsd = stats.cost ?? 0;
|
||||
const contextWindowMaxTokens = stats.contextUsage?.contextWindow ?? undefined;
|
||||
const contextWindowUsedTokens = stats.contextUsage?.tokens ?? undefined;
|
||||
|
||||
if (
|
||||
inputTokens === 0 &&
|
||||
cachedInputTokens === 0 &&
|
||||
outputTokens === 0 &&
|
||||
totalCostUsd === 0 &&
|
||||
contextWindowMaxTokens === undefined &&
|
||||
contextWindowUsedTokens === undefined
|
||||
) {
|
||||
return undefined;
|
||||
}
|
||||
|
||||
return {
|
||||
inputTokens,
|
||||
cachedInputTokens,
|
||||
outputTokens,
|
||||
totalCostUsd,
|
||||
...(typeof contextWindowMaxTokens === "number" ? { contextWindowMaxTokens } : {}),
|
||||
...(typeof contextWindowUsedTokens === "number" ? { contextWindowUsedTokens } : {}),
|
||||
};
|
||||
}
|
||||
|
||||
function ompModelSupportsImageInput(model: OmpModel | null | undefined): boolean {
|
||||
return model?.input?.includes("image") === true;
|
||||
}
|
||||
@@ -898,6 +869,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
private readonly currentModeId: string | null;
|
||||
private readonly providerIdleScheduler: OmpProviderIdleScheduler;
|
||||
private readonly noTurnScheduler: OmpNoTurnScheduler;
|
||||
private readonly usagePoller: OmpUsagePoller;
|
||||
private closed = false;
|
||||
private live: boolean;
|
||||
private readonly emittedUserMessageIds = new Set<string>();
|
||||
@@ -912,6 +884,21 @@ export class OmpAgentSession implements AgentSession {
|
||||
this.live = options.live ?? true;
|
||||
this.providerIdleScheduler = options.providerIdleScheduler ?? createOmpProviderIdleScheduler();
|
||||
this.noTurnScheduler = options.noTurnScheduler ?? createOmpNoTurnScheduler();
|
||||
this.usagePoller = new OmpUsagePoller({
|
||||
scheduler: options.usagePollScheduler,
|
||||
readStats: () => this.runtimeSession.getSessionStats(),
|
||||
onUsage: (usage, turnId) => {
|
||||
this.emit({
|
||||
type: "usage_updated",
|
||||
provider: this.provider,
|
||||
usage,
|
||||
...(turnId === undefined ? {} : { turnId }),
|
||||
});
|
||||
},
|
||||
onPollError: (error) => {
|
||||
this.logger.debug({ err: error }, "OMP context usage poll failed");
|
||||
},
|
||||
});
|
||||
this.subagentCardTracker = new OmpSubagentCardTracker({
|
||||
scheduler: options.subagentCardScheduler,
|
||||
});
|
||||
@@ -976,6 +963,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
this.activePromptRequestId = null;
|
||||
this.clearNoTurnBuffers();
|
||||
this.activeNoTurnPromptText = payload.text;
|
||||
this.usagePoller.startTurn();
|
||||
|
||||
void (async () => {
|
||||
try {
|
||||
@@ -1000,6 +988,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
if (this.activeTurnId !== turnId) {
|
||||
return;
|
||||
}
|
||||
this.usagePoller.stopTurn();
|
||||
this.activeTurnId = null;
|
||||
this.activeClientMessageId = null;
|
||||
this.activeTurnStarted = false;
|
||||
@@ -1129,12 +1118,10 @@ export class OmpAgentSession implements AgentSession {
|
||||
|
||||
async interrupt(): Promise<void> {
|
||||
const turnId = this.activeTurnId;
|
||||
try {
|
||||
await this.runtimeSession.abort();
|
||||
} finally {
|
||||
this.terminalizeActiveWork();
|
||||
}
|
||||
await this.runtimeSession.abort();
|
||||
if (turnId && this.activeTurnId === turnId) {
|
||||
this.terminalizeActiveWork();
|
||||
this.usagePoller.stopTurn();
|
||||
this.activeTurnId = null;
|
||||
this.activeClientMessageId = null;
|
||||
this.activeTurnStarted = false;
|
||||
@@ -1169,6 +1156,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
return;
|
||||
}
|
||||
this.closed = true;
|
||||
this.usagePoller.close();
|
||||
this.cancelNoTurnPromptCompletion();
|
||||
try {
|
||||
await this.runtimeSession.close();
|
||||
@@ -1786,6 +1774,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
}
|
||||
|
||||
private handleProcessExit(error: string): void {
|
||||
this.usagePoller.stopTurn();
|
||||
this.terminalizeActiveWork();
|
||||
this.subagentIndex.clear(this.runtimeSession);
|
||||
if (!this.activeTurnId) {
|
||||
@@ -2142,6 +2131,7 @@ export class OmpAgentSession implements AgentSession {
|
||||
this.clearNoTurnBuffers();
|
||||
const errorMessage = latestOmpErrorMessage(messages);
|
||||
if (typeof errorMessage === "string" && errorMessage.length > 0) {
|
||||
this.usagePoller.stopTurn();
|
||||
this.emit({
|
||||
type: "turn_failed",
|
||||
provider: this.provider,
|
||||
@@ -2150,12 +2140,13 @@ export class OmpAgentSession implements AgentSession {
|
||||
});
|
||||
return;
|
||||
}
|
||||
const finalUsage = this.usagePoller.completeTurn(turnId);
|
||||
this.emit({
|
||||
type: "turn_completed",
|
||||
provider: this.provider,
|
||||
turnId,
|
||||
});
|
||||
void this.refreshAfterTurn(turnId);
|
||||
void this.refreshAfterTurn(finalUsage);
|
||||
}
|
||||
|
||||
private async completeTurnAfterProviderIdle(
|
||||
@@ -2181,23 +2172,8 @@ export class OmpAgentSession implements AgentSession {
|
||||
this.state = await this.runtimeSession.getState();
|
||||
}
|
||||
|
||||
private async refreshAfterTurn(turnId: string | undefined): Promise<void> {
|
||||
await this.refreshState().catch(() => undefined);
|
||||
const usage = await this.runtimeSession
|
||||
.getSessionStats()
|
||||
.then((stats) => {
|
||||
const baseUsage = toAgentUsage(stats);
|
||||
return mapOmpUsage({ stats, state: this.state, baseUsage });
|
||||
})
|
||||
.catch(() => undefined);
|
||||
if (usage) {
|
||||
this.emit({
|
||||
type: "usage_updated",
|
||||
provider: this.provider,
|
||||
turnId,
|
||||
usage,
|
||||
});
|
||||
}
|
||||
private async refreshAfterTurn(finalUsage: Promise<void>): Promise<void> {
|
||||
await Promise.all([this.refreshState().catch(() => undefined), finalUsage]);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -2212,6 +2188,7 @@ export class OmpAgentClient implements AgentClient {
|
||||
private readonly subagentCardScheduler?: OmpSubagentCardScheduler;
|
||||
private readonly providerIdleScheduler?: OmpProviderIdleScheduler;
|
||||
private readonly noTurnScheduler?: OmpNoTurnScheduler;
|
||||
private readonly usagePollScheduler?: OmpUsagePollScheduler;
|
||||
private readonly runtime: OmpRuntime;
|
||||
|
||||
constructor(options: OmpAgentClientOptions) {
|
||||
@@ -2234,6 +2211,7 @@ export class OmpAgentClient implements AgentClient {
|
||||
this.subagentCardScheduler = options.subagentCardScheduler;
|
||||
this.providerIdleScheduler = options.providerIdleScheduler;
|
||||
this.noTurnScheduler = options.noTurnScheduler;
|
||||
this.usagePollScheduler = options.usagePollScheduler;
|
||||
this.runtime = options.runtime ?? createRuntime(options.logger, runtimeSettings);
|
||||
}
|
||||
|
||||
@@ -2274,6 +2252,7 @@ export class OmpAgentClient implements AgentClient {
|
||||
subagentCardScheduler: this.subagentCardScheduler,
|
||||
providerIdleScheduler: this.providerIdleScheduler,
|
||||
noTurnScheduler: this.noTurnScheduler,
|
||||
usagePollScheduler: this.usagePollScheduler,
|
||||
paseoTools: launchContext?.paseoTools,
|
||||
});
|
||||
} catch (error) {
|
||||
@@ -2315,6 +2294,7 @@ export class OmpAgentClient implements AgentClient {
|
||||
subagentCardScheduler: this.subagentCardScheduler,
|
||||
providerIdleScheduler: this.providerIdleScheduler,
|
||||
noTurnScheduler: this.noTurnScheduler,
|
||||
usagePollScheduler: this.usagePollScheduler,
|
||||
paseoTools: launchContext?.paseoTools,
|
||||
live: false,
|
||||
});
|
||||
|
||||
@@ -112,6 +112,7 @@ export class FakeOmpSession implements OmpRuntimeSession {
|
||||
readonly hostToolUpdates: OmpRpcHostToolUpdate[] = [];
|
||||
getStateRequestCount = 0;
|
||||
abortRequested = false;
|
||||
abortError: Error | null = null;
|
||||
readonly canceledExtensionUiRequests: string[] = [];
|
||||
readonly extensionUiResponses: Array<{
|
||||
id: string;
|
||||
@@ -238,6 +239,9 @@ export class FakeOmpSession implements OmpRuntimeSession {
|
||||
}
|
||||
|
||||
async abort(): Promise<void> {
|
||||
if (this.abortError) {
|
||||
throw this.abortError;
|
||||
}
|
||||
this.abortRequested = true;
|
||||
}
|
||||
|
||||
|
||||
@@ -18,6 +18,7 @@ import {
|
||||
type OmpNoTurnScheduler,
|
||||
type OmpProviderIdleScheduler,
|
||||
} from "../agent.js";
|
||||
import type { OmpUsagePollScheduler } from "../usage-poller.js";
|
||||
import type { OmpAgentMessage, OmpRpcSlashCommand } from "../rpc-types.js";
|
||||
import { FakeOmp } from "./fake-omp.js";
|
||||
|
||||
@@ -69,6 +70,7 @@ export class OmpHarness {
|
||||
options: {
|
||||
providerIdleScheduler?: OmpProviderIdleScheduler;
|
||||
noTurnScheduler?: OmpNoTurnScheduler;
|
||||
usagePollScheduler?: OmpUsagePollScheduler;
|
||||
} = {},
|
||||
) {
|
||||
this.client = new OmpAgentClient({
|
||||
@@ -76,6 +78,7 @@ export class OmpHarness {
|
||||
runtime: this.omp,
|
||||
providerIdleScheduler: options.providerIdleScheduler,
|
||||
noTurnScheduler: options.noTurnScheduler,
|
||||
usagePollScheduler: options.usagePollScheduler,
|
||||
});
|
||||
}
|
||||
|
||||
@@ -407,6 +410,10 @@ export class OmpHarness {
|
||||
return this.events.filter((event) => event.type === "turn_completed").length;
|
||||
}
|
||||
|
||||
usageUpdates() {
|
||||
return this.events.flatMap((event) => (event.type === "usage_updated" ? [event.usage] : []));
|
||||
}
|
||||
|
||||
requestToolApproval(input: {
|
||||
id: string;
|
||||
tool: "bash" | "edit" | "write";
|
||||
|
||||
@@ -1,39 +0,0 @@
|
||||
import { describe, expect, test } from "vitest";
|
||||
|
||||
import { mapOmpUsage } from "./usage-mapper.js";
|
||||
|
||||
describe("OMP usage mapper", () => {
|
||||
test("adds OMP context usage to token and cost totals", () => {
|
||||
const usage = mapOmpUsage({
|
||||
stats: {
|
||||
tokens: { input: 28237, output: 548, cacheRead: 269824, cacheWrite: 0, total: 298609 },
|
||||
cost: 0.29253700000000005,
|
||||
},
|
||||
state: {
|
||||
model: null,
|
||||
thinkingLevel: "medium",
|
||||
isStreaming: false,
|
||||
isCompacting: false,
|
||||
sessionId: "session",
|
||||
messageCount: 0,
|
||||
queuedMessageCount: 0,
|
||||
contextUsage: { tokens: 23656, contextWindow: 272000, percent: 8.7 },
|
||||
},
|
||||
baseUsage: {
|
||||
inputTokens: 28237,
|
||||
cachedInputTokens: 269824,
|
||||
outputTokens: 548,
|
||||
totalCostUsd: 0.29253700000000005,
|
||||
},
|
||||
});
|
||||
|
||||
expect(usage).toEqual({
|
||||
inputTokens: 28237,
|
||||
cachedInputTokens: 269824,
|
||||
outputTokens: 548,
|
||||
totalCostUsd: 0.29253700000000005,
|
||||
contextWindowMaxTokens: 272000,
|
||||
contextWindowUsedTokens: 23656,
|
||||
});
|
||||
});
|
||||
});
|
||||
@@ -1,24 +0,0 @@
|
||||
import type { AgentUsage } from "../../agent-sdk-types.js";
|
||||
import type { OmpSessionState, OmpSessionStats } from "./rpc-types.js";
|
||||
|
||||
export function mapOmpUsage(input: {
|
||||
stats: OmpSessionStats;
|
||||
state: OmpSessionState;
|
||||
baseUsage: AgentUsage | undefined;
|
||||
}): AgentUsage | undefined {
|
||||
const contextWindowUsedTokens = finiteNumber(input.state.contextUsage?.tokens);
|
||||
const contextWindowMaxTokens = finiteNumber(input.state.contextUsage?.contextWindow);
|
||||
if (contextWindowUsedTokens === undefined && contextWindowMaxTokens === undefined) {
|
||||
return input.baseUsage;
|
||||
}
|
||||
|
||||
return {
|
||||
...input.baseUsage,
|
||||
...(contextWindowMaxTokens !== undefined ? { contextWindowMaxTokens } : {}),
|
||||
...(contextWindowUsedTokens !== undefined ? { contextWindowUsedTokens } : {}),
|
||||
};
|
||||
}
|
||||
|
||||
function finiteNumber(value: number | null | undefined): number | undefined {
|
||||
return typeof value === "number" && Number.isFinite(value) ? value : undefined;
|
||||
}
|
||||
@@ -0,0 +1,174 @@
|
||||
import { setImmediate as waitForImmediate } from "node:timers/promises";
|
||||
import { describe, expect, test } from "vitest";
|
||||
|
||||
import type { AgentUsage } from "../../agent-sdk-types.js";
|
||||
import type { OmpSessionStats } from "./rpc-types.js";
|
||||
import { OmpUsagePoller, type OmpUsagePollScheduler } from "./usage-poller.js";
|
||||
|
||||
class ManualPollScheduler implements OmpUsagePollScheduler {
|
||||
private readonly polls: Array<{ active: boolean; callback: () => void }> = [];
|
||||
|
||||
schedulePoll(callback: () => void): () => void {
|
||||
const poll = { active: true, callback };
|
||||
this.polls.push(poll);
|
||||
return () => {
|
||||
poll.active = false;
|
||||
};
|
||||
}
|
||||
|
||||
poll(): void {
|
||||
const poll = this.polls.shift();
|
||||
if (!poll) throw new Error("No context usage poll is scheduled");
|
||||
if (poll.active) poll.callback();
|
||||
}
|
||||
|
||||
activePollCount(): number {
|
||||
return this.polls.filter((poll) => poll.active).length;
|
||||
}
|
||||
}
|
||||
|
||||
function deferred<T>() {
|
||||
let resolve!: (value: T) => void;
|
||||
const promise = new Promise<T>((resolvePromise) => {
|
||||
resolve = resolvePromise;
|
||||
});
|
||||
return { promise, resolve };
|
||||
}
|
||||
|
||||
describe("OMP usage poller", () => {
|
||||
test("emits only changed usage while active", async () => {
|
||||
const scheduler = new ManualPollScheduler();
|
||||
const updates: AgentUsage[] = [];
|
||||
let stats: OmpSessionStats = {
|
||||
tokens: { input: 100, cacheRead: 10, output: 20 },
|
||||
cost: 0.01,
|
||||
contextUsage: { contextWindow: 200_000, tokens: 130 },
|
||||
};
|
||||
const poller = new OmpUsagePoller({
|
||||
scheduler,
|
||||
readStats: async () => stats,
|
||||
onUsage: (update) => updates.push(update),
|
||||
onPollError: (error) => {
|
||||
throw error;
|
||||
},
|
||||
});
|
||||
|
||||
poller.startTurn();
|
||||
scheduler.poll();
|
||||
await waitForImmediate();
|
||||
scheduler.poll();
|
||||
await waitForImmediate();
|
||||
stats = { ...stats, contextUsage: { contextWindow: 200_000, tokens: 150 } };
|
||||
scheduler.poll();
|
||||
await waitForImmediate();
|
||||
|
||||
expect(updates).toEqual([
|
||||
{
|
||||
inputTokens: 100,
|
||||
cachedInputTokens: 10,
|
||||
outputTokens: 20,
|
||||
totalCostUsd: 0.01,
|
||||
contextWindowMaxTokens: 200_000,
|
||||
contextWindowUsedTokens: 130,
|
||||
},
|
||||
{
|
||||
inputTokens: 100,
|
||||
cachedInputTokens: 10,
|
||||
outputTokens: 20,
|
||||
totalCostUsd: 0.01,
|
||||
contextWindowMaxTokens: 200_000,
|
||||
contextWindowUsedTokens: 150,
|
||||
},
|
||||
]);
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
poller.stopTurn();
|
||||
expect(scheduler.activePollCount()).toBe(0);
|
||||
});
|
||||
|
||||
test("drops an in-flight poll without hiding the final refresh", async () => {
|
||||
const scheduler = new ManualPollScheduler();
|
||||
const inFlightStats = deferred<OmpSessionStats>();
|
||||
const finalStats = { contextUsage: { contextWindow: 200_000, tokens: 150 } };
|
||||
const finalUsage = {
|
||||
inputTokens: 0,
|
||||
cachedInputTokens: 0,
|
||||
outputTokens: 0,
|
||||
totalCostUsd: 0,
|
||||
contextWindowMaxTokens: 200_000,
|
||||
contextWindowUsedTokens: 150,
|
||||
};
|
||||
const updates: Array<{ usage: AgentUsage; turnId?: string }> = [];
|
||||
let readCount = 0;
|
||||
const poller = new OmpUsagePoller({
|
||||
scheduler,
|
||||
readStats: async () => {
|
||||
readCount += 1;
|
||||
return readCount === 1 ? inFlightStats.promise : finalStats;
|
||||
},
|
||||
onUsage: (usage, turnId) => updates.push({ usage, turnId }),
|
||||
onPollError: (error) => {
|
||||
throw error;
|
||||
},
|
||||
});
|
||||
|
||||
poller.startTurn();
|
||||
scheduler.poll();
|
||||
await expect(poller.completeTurn("turn-1")).resolves.toBeUndefined();
|
||||
inFlightStats.resolve({
|
||||
contextUsage: { contextWindow: 200_000, tokens: 130 },
|
||||
});
|
||||
await waitForImmediate();
|
||||
|
||||
expect(updates).toEqual([{ usage: finalUsage, turnId: "turn-1" }]);
|
||||
await expect(poller.completeTurn()).resolves.toBeUndefined();
|
||||
expect(updates).toEqual([{ usage: finalUsage, turnId: "turn-1" }]);
|
||||
});
|
||||
|
||||
test("drops a final refresh when a newer turn starts", async () => {
|
||||
const scheduler = new ManualPollScheduler();
|
||||
const finalStats = deferred<OmpSessionStats>();
|
||||
const updates: AgentUsage[] = [];
|
||||
const poller = new OmpUsagePoller({
|
||||
scheduler,
|
||||
readStats: () => finalStats.promise,
|
||||
onUsage: (usage) => updates.push(usage),
|
||||
onPollError: (error) => {
|
||||
throw error;
|
||||
},
|
||||
});
|
||||
|
||||
poller.startTurn();
|
||||
const completion = poller.completeTurn();
|
||||
poller.startTurn();
|
||||
finalStats.resolve({ contextUsage: { contextWindow: 200_000, tokens: 150 } });
|
||||
|
||||
await expect(completion).resolves.toBeUndefined();
|
||||
expect(updates).toEqual([]);
|
||||
expect(scheduler.activePollCount()).toBe(1);
|
||||
poller.stopTurn();
|
||||
});
|
||||
|
||||
test("close permanently suppresses pending usage", async () => {
|
||||
const scheduler = new ManualPollScheduler();
|
||||
const finalStats = deferred<OmpSessionStats>();
|
||||
const updates: AgentUsage[] = [];
|
||||
const poller = new OmpUsagePoller({
|
||||
scheduler,
|
||||
readStats: () => finalStats.promise,
|
||||
onUsage: (usage) => updates.push(usage),
|
||||
onPollError: (error) => {
|
||||
throw error;
|
||||
},
|
||||
});
|
||||
|
||||
poller.startTurn();
|
||||
const completion = poller.completeTurn();
|
||||
poller.close();
|
||||
poller.startTurn();
|
||||
finalStats.resolve({ contextUsage: { contextWindow: 200_000, tokens: 150 } });
|
||||
|
||||
await expect(completion).resolves.toBeUndefined();
|
||||
expect(updates).toEqual([]);
|
||||
expect(scheduler.activePollCount()).toBe(0);
|
||||
});
|
||||
});
|
||||
@@ -0,0 +1,169 @@
|
||||
import type { AgentUsage } from "../../agent-sdk-types.js";
|
||||
import type { OmpSessionStats } from "./rpc-types.js";
|
||||
|
||||
export interface OmpUsagePollScheduler {
|
||||
schedulePoll(callback: () => void): () => void;
|
||||
}
|
||||
|
||||
interface OmpUsagePollerOptions {
|
||||
readStats(): Promise<OmpSessionStats>;
|
||||
onUsage(usage: AgentUsage, turnId?: string): void;
|
||||
onPollError(error: unknown): void;
|
||||
scheduler?: OmpUsagePollScheduler;
|
||||
}
|
||||
|
||||
function createScheduler(): OmpUsagePollScheduler {
|
||||
return {
|
||||
schedulePoll: (callback) => {
|
||||
const timer = setTimeout(callback, 3_000);
|
||||
return () => clearTimeout(timer);
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
function toAgentUsage(stats: OmpSessionStats): AgentUsage | undefined {
|
||||
const inputTokens = stats.tokens?.input ?? 0;
|
||||
const cachedInputTokens = stats.tokens?.cacheRead ?? 0;
|
||||
const outputTokens = stats.tokens?.output ?? 0;
|
||||
const totalCostUsd = stats.cost ?? 0;
|
||||
const contextWindowMaxTokens = stats.contextUsage?.contextWindow ?? undefined;
|
||||
const contextWindowUsedTokens = stats.contextUsage?.tokens ?? undefined;
|
||||
|
||||
if (
|
||||
inputTokens === 0 &&
|
||||
cachedInputTokens === 0 &&
|
||||
outputTokens === 0 &&
|
||||
totalCostUsd === 0 &&
|
||||
contextWindowMaxTokens === undefined &&
|
||||
contextWindowUsedTokens === undefined
|
||||
) {
|
||||
return undefined;
|
||||
}
|
||||
|
||||
return {
|
||||
inputTokens,
|
||||
cachedInputTokens,
|
||||
outputTokens,
|
||||
totalCostUsd,
|
||||
...(typeof contextWindowMaxTokens === "number" ? { contextWindowMaxTokens } : {}),
|
||||
...(typeof contextWindowUsedTokens === "number" ? { contextWindowUsedTokens } : {}),
|
||||
};
|
||||
}
|
||||
|
||||
function isSameUsage(left: AgentUsage, right: AgentUsage): boolean {
|
||||
return (
|
||||
left.inputTokens === right.inputTokens &&
|
||||
left.cachedInputTokens === right.cachedInputTokens &&
|
||||
left.outputTokens === right.outputTokens &&
|
||||
left.totalCostUsd === right.totalCostUsd &&
|
||||
left.contextWindowMaxTokens === right.contextWindowMaxTokens &&
|
||||
left.contextWindowUsedTokens === right.contextWindowUsedTokens
|
||||
);
|
||||
}
|
||||
|
||||
export class OmpUsagePoller {
|
||||
private readonly scheduler: OmpUsagePollScheduler;
|
||||
private active = false;
|
||||
private closed = false;
|
||||
private generation = 0;
|
||||
private cancelScheduledPoll: (() => void) | null = null;
|
||||
private lastUsage: AgentUsage | null = null;
|
||||
|
||||
constructor(private readonly options: OmpUsagePollerOptions) {
|
||||
this.scheduler = options.scheduler ?? createScheduler();
|
||||
}
|
||||
|
||||
startTurn(): void {
|
||||
if (this.closed || this.active) {
|
||||
return;
|
||||
}
|
||||
this.active = true;
|
||||
this.generation += 1;
|
||||
this.schedule(this.generation);
|
||||
}
|
||||
|
||||
stopTurn(): void {
|
||||
this.active = false;
|
||||
this.invalidatePendingWork();
|
||||
}
|
||||
|
||||
async completeTurn(turnId?: string): Promise<void> {
|
||||
if (this.closed) {
|
||||
return;
|
||||
}
|
||||
this.active = false;
|
||||
const completionGeneration = this.invalidatePendingWork();
|
||||
let usage: AgentUsage | undefined;
|
||||
try {
|
||||
usage = toAgentUsage(await this.options.readStats());
|
||||
} catch {
|
||||
return;
|
||||
}
|
||||
if (this.closed || this.generation !== completionGeneration) {
|
||||
return;
|
||||
}
|
||||
this.publishUsage(usage, turnId);
|
||||
}
|
||||
|
||||
close(): void {
|
||||
this.closed = true;
|
||||
this.active = false;
|
||||
this.invalidatePendingWork();
|
||||
}
|
||||
|
||||
private invalidatePendingWork(): number {
|
||||
this.generation += 1;
|
||||
this.cancelScheduledPoll?.();
|
||||
this.cancelScheduledPoll = null;
|
||||
return this.generation;
|
||||
}
|
||||
|
||||
private schedule(generation: number): void {
|
||||
this.cancelScheduledPoll = this.scheduler.schedulePoll(() => {
|
||||
this.cancelScheduledPoll = null;
|
||||
void this.poll(generation);
|
||||
});
|
||||
}
|
||||
|
||||
private async poll(generation: number): Promise<void> {
|
||||
let usage: AgentUsage | undefined;
|
||||
try {
|
||||
usage = toAgentUsage(await this.options.readStats());
|
||||
} catch (error) {
|
||||
try {
|
||||
if (this.active && !this.closed && this.generation === generation) {
|
||||
this.options.onPollError(error);
|
||||
}
|
||||
} finally {
|
||||
if (this.active && !this.closed && this.generation === generation) {
|
||||
this.schedule(generation);
|
||||
}
|
||||
}
|
||||
return;
|
||||
}
|
||||
try {
|
||||
if (this.active && !this.closed && this.generation === generation) {
|
||||
this.publishUsage(usage);
|
||||
}
|
||||
} finally {
|
||||
if (this.active && !this.closed && this.generation === generation) {
|
||||
this.schedule(generation);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private takeChangedUsage(usage: AgentUsage | undefined): AgentUsage | undefined {
|
||||
if (!usage || (this.lastUsage && isSameUsage(this.lastUsage, usage))) {
|
||||
return undefined;
|
||||
}
|
||||
this.lastUsage = usage;
|
||||
return usage;
|
||||
}
|
||||
|
||||
private publishUsage(usage: AgentUsage | undefined, turnId?: string): void {
|
||||
const changedUsage = this.takeChangedUsage(usage);
|
||||
if (changedUsage) {
|
||||
this.options.onUsage(changedUsage, turnId);
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user