fix(dictation): avoid timeout from dangling non-final segments

This commit is contained in:
Mohamed Boudra
2026-02-18 21:21:27 +07:00
parent bcae95c372
commit a1fdeeb7ae
16 changed files with 419 additions and 10 deletions
+4 -1
View File
@@ -58,6 +58,7 @@ import {
import { loadOrCreateDaemonKeyPair } from "./daemon-keypair.js";
import { startRelayTransport, type RelayTransportController } from "./relay-transport.js";
import { getOrCreateServerId } from "./server-id.js";
import { resolveDaemonVersion } from "./daemon-version.js";
import type {
AgentClient,
AgentProvider,
@@ -136,6 +137,7 @@ export async function createPaseoDaemon(
rootLogger: Logger
): Promise<PaseoDaemon> {
const logger = rootLogger.child({ module: "bootstrap" });
const daemonVersion = resolveDaemonVersion(import.meta.url);
// Acquire PID lock before expensive bootstrap work so duplicate starts fail immediately.
await acquirePidLock(config.paseoHome, config.listen);
@@ -496,7 +498,8 @@ export async function createPaseoDaemon(
localModels: localModelConfig ?? undefined,
getSpeechReadiness,
},
config.agentProviderSettings
config.agentProviderSettings,
daemonVersion
);
unsubscribeSpeechReadiness = subscribeSpeechReadiness((snapshot) => {
wsServer?.publishSpeechReadiness(snapshot);
@@ -0,0 +1,71 @@
import { afterEach, describe, expect, it } from "vitest";
import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import path from "node:path";
import { pathToFileURL } from "node:url";
import {
DaemonVersionResolutionError,
resolveDaemonVersion,
} from "./daemon-version.js";
const createdDirs: string[] = [];
function createTempDir(): string {
const dir = mkdtempSync(path.join(tmpdir(), "paseo-daemon-version-"));
createdDirs.push(dir);
return dir;
}
afterEach(() => {
for (const dir of createdDirs.splice(0, createdDirs.length)) {
rmSync(dir, { recursive: true, force: true });
}
});
describe("resolveDaemonVersion", () => {
it("resolves server version by walking up to @getpaseo/server package.json", () => {
const root = createTempDir();
writeFileSync(
path.join(root, "package.json"),
JSON.stringify({ name: "@getpaseo/server", version: "9.8.7" }),
"utf8"
);
const nestedDir = path.join(root, "dist", "server");
mkdirSync(nestedDir, { recursive: true });
const moduleUrl = pathToFileURL(path.join(nestedDir, "index.js")).href;
expect(resolveDaemonVersion(moduleUrl)).toBe("9.8.7");
});
it("throws when @getpaseo/server package metadata cannot be resolved", () => {
const root = createTempDir();
writeFileSync(
path.join(root, "package.json"),
JSON.stringify({ name: "not-getpaseo-server", version: "1.2.3" }),
"utf8"
);
const nestedDir = path.join(root, "dist", "server");
mkdirSync(nestedDir, { recursive: true });
const moduleUrl = pathToFileURL(path.join(nestedDir, "index.js")).href;
expect(() => resolveDaemonVersion(moduleUrl)).toThrow(
DaemonVersionResolutionError
);
});
it("throws when @getpaseo/server version is missing", () => {
const root = createTempDir();
writeFileSync(
path.join(root, "package.json"),
JSON.stringify({ name: "@getpaseo/server" }),
"utf8"
);
const nestedDir = path.join(root, "dist", "server");
mkdirSync(nestedDir, { recursive: true });
const moduleUrl = pathToFileURL(path.join(nestedDir, "index.js")).href;
expect(() => resolveDaemonVersion(moduleUrl)).toThrow(
DaemonVersionResolutionError
);
});
});
@@ -0,0 +1,25 @@
import {
PackageVersionResolutionError,
resolvePackageVersion,
} from "./package-version.js";
const SERVER_PACKAGE_NAME = "@getpaseo/server";
export class DaemonVersionResolutionError extends PackageVersionResolutionError {}
export function resolveDaemonVersion(moduleUrl: string = import.meta.url): string {
try {
return resolvePackageVersion({
moduleUrl,
packageName: SERVER_PACKAGE_NAME,
});
} catch (error) {
if (error instanceof PackageVersionResolutionError) {
throw new DaemonVersionResolutionError({
moduleUrl,
packageName: SERVER_PACKAGE_NAME,
});
}
throw error;
}
}
@@ -233,4 +233,82 @@ describe("DictationStreamManager (provider-agnostic provider)", () => {
expect(finishAccepted).toBeDefined();
expect(finishAccepted?.payload.timeoutMs).toBeGreaterThan(5000);
});
it("adapts finish timeout when only uncommitted non-final transcripts are pending", async () => {
const session = new FakeRealtimeSession();
const emitted: Array<{ type: string; payload: any }> = [];
const manager = new DictationStreamManager({
logger: pino({ level: "silent" }),
emit: (msg) => emitted.push(msg),
sessionId: "s1",
stt: new FakeSttProvider(session),
finalTimeoutMs: 5000,
});
await manager.handleStart("d-uncommitted-timeout", "audio/pcm;rate=24000;bits=16");
await manager.handleChunk({
dictationId: "d-uncommitted-timeout",
seq: 0,
audioBase64: buildPcmBase64(2000, 2400),
format: "audio/pcm;rate=24000;bits=16",
});
session.emitCommitted("seg-1");
session.emitTranscript("seg-1", "hello", true);
session.emitTranscript("seg-dangling", "hel", false);
await manager.handleFinish("d-uncommitted-timeout", 0);
const finishAccepted = emitted.find(
(msg) => msg.type === "dictation_stream_finish_accepted"
);
expect(finishAccepted).toBeDefined();
expect(finishAccepted?.payload.timeoutMs).toBeGreaterThan(5000);
});
it("drops dangling uncommitted non-final transcripts when finishing after silence tail clear", async () => {
vi.useFakeTimers();
try {
const session = new FakeRealtimeSession();
const emitted: Array<{ type: string; payload: any }> = [];
const manager = new DictationStreamManager({
logger: pino({ level: "silent" }),
emit: (msg) => emitted.push(msg),
sessionId: "s1",
stt: new FakeSttProvider(session),
finalTimeoutMs: 5000,
});
await manager.handleStart("d-clear-tail", "audio/pcm;rate=24000;bits=16");
await manager.handleChunk({
dictationId: "d-clear-tail",
seq: 0,
audioBase64: buildPcmBase64(2000, 2400),
format: "audio/pcm;rate=24000;bits=16",
});
session.emitCommitted("seg-1");
session.emitTranscript("seg-1", "hello", true);
await manager.handleChunk({
dictationId: "d-clear-tail",
seq: 1,
audioBase64: buildPcmBase64(0, 2400),
format: "audio/pcm;rate=24000;bits=16",
});
session.emitTranscript("seg-dangling", "", false);
await manager.handleFinish("d-clear-tail", 1);
await tick();
await vi.advanceTimersByTimeAsync(5_100);
const final = emitted.find((msg) => msg.type === "dictation_stream_final");
const error = emitted.find((msg) => msg.type === "dictation_stream_error");
expect(session.clearCalls).toBeGreaterThan(0);
expect(error).toBeUndefined();
expect(final?.payload.text).toBe("hello");
} finally {
vi.useRealTimers();
}
});
});
@@ -552,8 +552,19 @@ export class DictationStreamManager {
const pendingCommittedSegments = state.committedSegmentIds.reduce((count, segmentId) => {
return state.finalTranscriptSegmentIds.has(segmentId) ? count : count + 1;
}, 0);
const committedSet = new Set(state.committedSegmentIds);
const pendingUncommittedTranscriptSegments = Array.from(
state.transcriptsBySegmentId.keys()
).reduce((count, segmentId) => {
if (committedSet.has(segmentId)) {
return count;
}
return state.finalTranscriptSegmentIds.has(segmentId) ? count : count + 1;
}, 0);
const pendingSegments =
pendingCommittedSegments + (state.awaitingFinalCommit ? 1 : 0);
pendingCommittedSegments +
pendingUncommittedTranscriptSegments +
(state.awaitingFinalCommit ? 1 : 0);
const pendingAudioSeconds = Math.ceil(
Math.max(0, state.bytesSinceCommit) / bytesPerSecond
);
@@ -629,6 +640,16 @@ export class DictationStreamManager {
state.bytesSinceCommit = 0;
state.peakSinceCommit = 0;
state.awaitingFinalCommit = false;
const droppedSegments = this.dropUncommittedNonFinalTranscripts(state);
if (droppedSegments > 0) {
this.logger.debug(
{
dictationId,
droppedSegments,
},
"Dictation finish: dropped uncommitted non-final transcript segments after silence clear"
);
}
} else {
state.awaitingFinalCommit = true;
try {
@@ -646,6 +667,22 @@ export class DictationStreamManager {
state.finishSealed = true;
}
private dropUncommittedNonFinalTranscripts(state: DictationStreamState): number {
const committedSet = new Set(state.committedSegmentIds);
let droppedCount = 0;
for (const segmentId of state.transcriptsBySegmentId.keys()) {
if (committedSet.has(segmentId)) {
continue;
}
if (state.finalTranscriptSegmentIds.has(segmentId)) {
continue;
}
state.transcriptsBySegmentId.delete(segmentId);
droppedCount += 1;
}
return droppedCount;
}
private maybeFinalizeDictationStream(dictationId: string): void {
const state = this.streams.get(dictationId);
if (!state) {
@@ -0,0 +1,64 @@
import { existsSync, readFileSync } from "node:fs";
import path from "node:path";
import { fileURLToPath } from "node:url";
type ResolvePackageVersionParams = {
moduleUrl?: string;
packageName: string;
};
type PackageJson = {
name?: unknown;
version?: unknown;
};
export class PackageVersionResolutionError extends Error {
constructor(params: { moduleUrl: string; packageName: string }) {
super(
`Unable to resolve ${params.packageName} version from module URL ${params.moduleUrl}.`
);
this.name = "PackageVersionResolutionError";
}
}
export function resolvePackageVersion(params: ResolvePackageVersionParams): string {
const moduleUrl = params.moduleUrl ?? import.meta.url;
let currentDir = path.dirname(fileURLToPath(moduleUrl));
while (true) {
const packageJsonPath = path.join(currentDir, "package.json");
if (existsSync(packageJsonPath)) {
try {
const packageJson = JSON.parse(readFileSync(packageJsonPath, "utf8")) as PackageJson;
if (packageJson.name === params.packageName) {
if (
typeof packageJson.version === "string" &&
packageJson.version.trim().length > 0
) {
return packageJson.version.trim();
}
throw new PackageVersionResolutionError({
moduleUrl,
packageName: params.packageName,
});
}
} catch (error) {
if (error instanceof PackageVersionResolutionError) {
throw error;
}
// Continue searching parent directories.
}
}
const parentDir = path.dirname(currentDir);
if (parentDir === currentDir) {
break;
}
currentDir = parentDir;
}
throw new PackageVersionResolutionError({
moduleUrl,
packageName: params.packageName,
});
}
@@ -75,7 +75,13 @@ function createServer(agentManagerOverrides?: Record<string, unknown>) {
{} as any,
"/tmp/paseo-test",
async () => ({} as any),
{ allowedOrigins: new Set() }
{ allowedOrigins: new Set() },
undefined,
undefined,
undefined,
undefined,
undefined,
"1.2.3-test"
);
return { server, agentManager };
@@ -79,6 +79,8 @@ import {
import { parseServerInfoStatusPayload } from "./messages.js";
import type { SpeechReadinessSnapshot } from "./speech/speech-runtime.js";
const TEST_DAEMON_VERSION = "1.2.3-test";
class MockSocket {
readyState = 1;
sent: unknown[] = [];
@@ -157,7 +159,9 @@ function createServer(options?: { speechReadiness?: SpeechReadinessSnapshot | nu
? {
getSpeechReadiness: () => speechReadiness,
}
: undefined
: undefined,
undefined,
TEST_DAEMON_VERSION
);
}
@@ -318,6 +322,7 @@ describe("relay external socket reconnect behavior", () => {
const payload = parseServerInfoStatusPayload(envelope.message?.payload);
expect(payload?.status).toBe("server_info");
expect(payload?.version).toBe(TEST_DAEMON_VERSION);
expect(payload?.capabilities?.voice?.dictation.enabled).toBe(
speechReadiness.dictation.enabled
);
+15 -1
View File
@@ -158,6 +158,13 @@ type SessionConnection = {
const EXTERNAL_SESSION_DISCONNECT_GRACE_MS = 90_000;
export class MissingDaemonVersionError extends Error {
constructor() {
super("VoiceAssistantWebSocketServer requires a non-empty daemonVersion.");
this.name = "MissingDaemonVersionError";
}
}
/**
* WebSocket server that only accepts sockets + parses/forwards messages to the session layer.
*/
@@ -168,6 +175,7 @@ export class VoiceAssistantWebSocketServer {
private readonly externalSessionsByKey: Map<string, SessionConnection> = new Map();
private clientIdCounter = 0;
private readonly serverId: string;
private readonly daemonVersion: string;
private readonly agentManager: AgentManager;
private readonly agentStorage: AgentStorage;
private readonly downloadTokenStore: DownloadTokenStore;
@@ -229,10 +237,15 @@ export class VoiceAssistantWebSocketServer {
};
getSpeechReadiness?: () => SpeechReadinessSnapshot;
},
agentProviderRuntimeSettings?: AgentProviderRuntimeSettingsMap
agentProviderRuntimeSettings?: AgentProviderRuntimeSettingsMap,
daemonVersion?: string
) {
this.logger = logger.child({ module: "websocket-server" });
this.serverId = serverId;
if (typeof daemonVersion !== "string" || daemonVersion.trim().length === 0) {
throw new MissingDaemonVersionError();
}
this.daemonVersion = daemonVersion.trim();
this.agentManager = agentManager;
this.agentStorage = agentStorage;
this.downloadTokenStore = downloadTokenStore;
@@ -509,6 +522,7 @@ export class VoiceAssistantWebSocketServer {
status: "server_info",
serverId: this.serverId,
hostname: getHostname(),
version: this.daemonVersion,
...(this.serverCapabilities ? { capabilities: this.serverCapabilities } : {}),
};
}
+10
View File
@@ -1218,6 +1218,14 @@ const ServerInfoHostnameSchema = z.unknown().transform((value): string | null =>
return trimmed.length > 0 ? trimmed : null
})
const ServerInfoVersionSchema = z.unknown().transform((value): string | null => {
if (typeof value !== 'string') {
return null
}
const trimmed = value.trim()
return trimmed.length > 0 ? trimmed : null
})
const ServerCapabilitiesFromUnknownSchema = z
.unknown()
.optional()
@@ -1237,12 +1245,14 @@ export const ServerInfoStatusPayloadSchema = z
status: z.literal('server_info'),
serverId: z.string().trim().min(1),
hostname: ServerInfoHostnameSchema.optional(),
version: ServerInfoVersionSchema.optional(),
capabilities: ServerCapabilitiesFromUnknownSchema,
})
.passthrough()
.transform((payload) => ({
...payload,
hostname: payload.hostname ?? null,
version: payload.version ?? null,
}))
export const StatusMessageSchema = z.object({
+6 -1
View File
@@ -8,10 +8,15 @@ import {
computeExecutionOrder,
buildSortedChildrenMap,
} from "./execution-order.js";
import { resolvePackageVersion } from "../server/package-version.js";
import type { AgentType, ModelName, Task } from "./types.js";
const TASKS_DIR = resolve(process.cwd(), ".tasks");
const store = new FileTaskStore(TASKS_DIR);
const TASK_CLI_VERSION = resolvePackageVersion({
moduleUrl: import.meta.url,
packageName: "@getpaseo/server",
});
async function readStdin(): Promise<string> {
const chunks: Buffer[] = [];
@@ -24,7 +29,7 @@ async function readStdin(): Promise<string> {
const program = new Command()
.name("task")
.description("Minimal task management with dependency tracking")
.version("0.1.0")
.version(TASK_CLI_VERSION)
.addHelpText(
"after",
`