@zq-silk/yui 0.6.5 → 0.6.7
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/dist/cli/commandCatalog.js +6 -1
- package/dist/cli/invocationRouter.js +2 -1
- package/dist/cli/updatePorts.js +63 -9
- package/dist/cli.js +23 -0
- package/dist/controller/agentRuntimeObserver.js +76 -16
- package/dist/controller/controller.js +398 -76
- package/dist/controller/fileSchedulerStoreAdapter.js +60 -51
- package/dist/controller/jobSupervisor.js +2 -16
- package/dist/controller/resourceInventoryLinux.js +73 -20
- package/dist/controller/runtime.js +1 -0
- package/dist/controller/runtimeEventProcessor.js +171 -31
- package/dist/controller/updateReconciliation.js +113 -0
- package/dist/coordination/keyedWorkQueue.js +189 -0
- package/dist/core/controllerServer.js +7 -1
- package/dist/runtime/runtimeSessionCandidate.js +43 -0
- package/dist/scheduler/activeRoleRunDelivery.js +2 -4
- package/dist/scheduler/activeTaskProgress.js +2 -4
- package/dist/scheduler/leaderWakeupProcessor.js +3 -1
- package/dist/scheduler/ports.js +35 -2
- package/dist/scheduler/roleRunLiveness.js +44 -22
- package/dist/scheduler/roleRunStall.js +49 -25
- package/dist/storage/sqliteSchema.js +363 -14
- package/dist/storage/sqliteStore.js +403 -30
- package/dist/storage/storeRpc.js +5 -0
- package/dist/storage/taskStore.js +104 -6
- package/package.json +1 -1
|
@@ -849,7 +849,7 @@ export const ROOT_COMMAND = buildNode({
|
|
|
849
849
|
sections: [{
|
|
850
850
|
id: "lifecycle",
|
|
851
851
|
title: "Commands",
|
|
852
|
-
entries: ["status", "cleanup", "identity", "stop", "restart"]
|
|
852
|
+
entries: ["status", "cleanup", "identity", "live-identity", "stop", "restart"]
|
|
853
853
|
}],
|
|
854
854
|
children: [
|
|
855
855
|
{
|
|
@@ -869,6 +869,11 @@ export const ROOT_COMMAND = buildNode({
|
|
|
869
869
|
summary: "Read the stable runtime identity receipt (build, backend, worker).",
|
|
870
870
|
hidden: true
|
|
871
871
|
},
|
|
872
|
+
{
|
|
873
|
+
name: "live-identity",
|
|
874
|
+
summary: "Read the authenticated live Controller launch identity.",
|
|
875
|
+
hidden: true
|
|
876
|
+
},
|
|
872
877
|
{ name: "stop", summary: "Stop the Controller." },
|
|
873
878
|
{ name: "restart", summary: "Restart internal services without stopping tmux sessions." }
|
|
874
879
|
]
|
|
@@ -33,7 +33,8 @@ function resolveExecutionPath(args) {
|
|
|
33
33
|
const internalExecutable = child !== undefined && ((node === ROOT_COMMAND && child.name === "internal")
|
|
34
34
|
|| (node.path.join(" ") === "yui completion" && child.name === "candidates")
|
|
35
35
|
|| (node.path.join(" ") === "yui task run" && child.name === "checkpoint")
|
|
36
|
-
|| (node.path.join(" ") === "yui controller"
|
|
36
|
+
|| (node.path.join(" ") === "yui controller"
|
|
37
|
+
&& (child.name === "identity" || child.name === "live-identity")));
|
|
37
38
|
if (child === undefined || (child.hidden && !internalExecutable)) {
|
|
38
39
|
if (node.kind === "hybrid" && node.acceptsArguments)
|
|
39
40
|
break;
|
package/dist/cli/updatePorts.js
CHANGED
|
@@ -261,10 +261,12 @@ export function createUpdatePorts(environment, spawn = spawnSync, stagingRoot =
|
|
|
261
261
|
}
|
|
262
262
|
},
|
|
263
263
|
// `runUpdate` is intentionally synchronous because the npm/staged-binary
|
|
264
|
-
// ports use spawnSync.
|
|
265
|
-
//
|
|
266
|
-
//
|
|
264
|
+
// ports use spawnSync. A short-lived child owns the async, exactly fenced
|
|
265
|
+
// reconciliation first; lifecycle capture then uses structured live
|
|
266
|
+
// Controller commands. Tests can replace these seams with deterministic
|
|
267
|
+
// fakes without touching a real Controller.
|
|
267
268
|
controllerStatus(home) {
|
|
269
|
+
reconcileControllerResourcesForUpdate(home, environment, spawn);
|
|
268
270
|
return readControllerLifecycle(home, environment, spawn);
|
|
269
271
|
},
|
|
270
272
|
stopController(home, expectedPid) {
|
|
@@ -285,6 +287,7 @@ export function createUpdatePorts(environment, spawn = spawnSync, stagingRoot =
|
|
|
285
287
|
}
|
|
286
288
|
const UPDATE_CLI_PATH = fileURLToPath(new URL("../cli.js", import.meta.url));
|
|
287
289
|
const UPDATE_CLIENT_RUNTIME_PATH = fileURLToPath(new URL("../controller/clientRuntime.js", import.meta.url));
|
|
290
|
+
const UPDATE_CONTROLLER_RECONCILIATION_PATH = fileURLToPath(new URL("../controller/updateReconciliation.js", import.meta.url));
|
|
288
291
|
/**
|
|
289
292
|
* Capture running state plus an authenticated exact process identity before
|
|
290
293
|
* stopping. Public `controller status` is an inventory and intentionally
|
|
@@ -330,18 +333,18 @@ function readControllerLifecycle(home, environment, spawn, cliBinary) {
|
|
|
330
333
|
if (current !== undefined) {
|
|
331
334
|
throw new Error("Controller inventory is current but has no process proof; treating ownership as unknown-active.");
|
|
332
335
|
}
|
|
333
|
-
return proveControllerAbsent(home, environment, spawn);
|
|
336
|
+
return proveControllerAbsent(home, environment, spawn, cliBinary);
|
|
334
337
|
}
|
|
335
|
-
const identity = parseControllerIdentity(runControllerCommand(home, environment, spawn, "identity", cliBinary));
|
|
338
|
+
const identity = parseControllerIdentity(runControllerCommand(home, environment, spawn, "live-identity", cliBinary));
|
|
336
339
|
return {
|
|
337
340
|
running: true,
|
|
338
341
|
...(isPositivePid(processInfo.pid) ? { pid: processInfo.pid } : {}),
|
|
339
342
|
identity
|
|
340
343
|
};
|
|
341
344
|
}
|
|
342
|
-
function proveControllerAbsent(home, environment, spawn) {
|
|
345
|
+
function proveControllerAbsent(home, environment, spawn, cliBinary) {
|
|
343
346
|
try {
|
|
344
|
-
runControllerCommand(home, environment, spawn, "identity");
|
|
347
|
+
runControllerCommand(home, environment, spawn, "live-identity", cliBinary);
|
|
345
348
|
}
|
|
346
349
|
catch (error) {
|
|
347
350
|
if (controllerErrorCode(error) === "CONTROLLER_NOT_RUNNING") {
|
|
@@ -349,7 +352,7 @@ function proveControllerAbsent(home, environment, spawn) {
|
|
|
349
352
|
}
|
|
350
353
|
throw new Error(`Controller absence could not be authenticated: ${messageOf(error)}`, { cause: error });
|
|
351
354
|
}
|
|
352
|
-
throw new Error("Controller identity is reachable but inventory is currentless; treating ownership as unknown-active.");
|
|
355
|
+
throw new Error("A live Controller identity is reachable but inventory is currentless; treating ownership as unknown-active.");
|
|
353
356
|
}
|
|
354
357
|
function stopControllerForUpdate(home, expectedPid, environment, spawn) {
|
|
355
358
|
// Parent update owns this lifecycle boundary. Invoke the runtime client
|
|
@@ -464,7 +467,7 @@ function restartControllerForUpdate(home, environment, spawn, activatedBinary, a
|
|
|
464
467
|
const replacementPid = parseReplacementPid(restart.data);
|
|
465
468
|
let identityFailure;
|
|
466
469
|
try {
|
|
467
|
-
const identity = parseControllerIdentity(runControllerCommand(home, environment, spawn, "identity", activatedBinary));
|
|
470
|
+
const identity = parseControllerIdentity(runControllerCommand(home, environment, spawn, "live-identity", activatedBinary));
|
|
468
471
|
assertActivatedControllerIdentity(identity, activatedBinary, activatedVersion);
|
|
469
472
|
}
|
|
470
473
|
catch (error) {
|
|
@@ -586,6 +589,57 @@ function runControllerCommand(home, environment, spawn, method, cliBinary) {
|
|
|
586
589
|
}
|
|
587
590
|
return parsed.data;
|
|
588
591
|
}
|
|
592
|
+
function reconcileControllerResourcesForUpdate(home, environment, spawn) {
|
|
593
|
+
const helper = [
|
|
594
|
+
"const values = process.argv.slice(1);",
|
|
595
|
+
"const home = values.pop();",
|
|
596
|
+
"const reconciliationModule = values.pop();",
|
|
597
|
+
"(async () => {",
|
|
598
|
+
" const { reconcileControllerResourcesForUpdate } = await import(reconciliationModule);",
|
|
599
|
+
" const data = await reconcileControllerResourcesForUpdate(home, process.env);",
|
|
600
|
+
" process.stdout.write(JSON.stringify({ ok: true, data }));",
|
|
601
|
+
"})().catch((error) => {",
|
|
602
|
+
" const message = error instanceof Error ? error.message : String(error);",
|
|
603
|
+
" process.stderr.write(JSON.stringify({ ok: false, code: 'RUNTIME_ERROR', message }));",
|
|
604
|
+
" process.exitCode = 5;",
|
|
605
|
+
"});"
|
|
606
|
+
].join(" ");
|
|
607
|
+
const result = spawn(process.execPath, ["-e", helper, UPDATE_CONTROLLER_RECONCILIATION_PATH, home], { cwd: process.cwd(), env: { ...environment, YUI_HOME: home }, shell: false });
|
|
608
|
+
if (result.error !== undefined || result.status !== 0) {
|
|
609
|
+
const detail = structuredErrorMessage(result) ?? result.stderr.toString("utf8").trim();
|
|
610
|
+
throw new Error(`Controller reconciliation failed (exit ${result.status ?? "null"})${detail.length === 0 ? "." : `: ${detail}`}`);
|
|
611
|
+
}
|
|
612
|
+
let parsed;
|
|
613
|
+
try {
|
|
614
|
+
parsed = JSON.parse(result.stdout.toString("utf8"));
|
|
615
|
+
}
|
|
616
|
+
catch (error) {
|
|
617
|
+
throw new Error("Controller reconciliation returned an invalid structured result.", {
|
|
618
|
+
cause: error
|
|
619
|
+
});
|
|
620
|
+
}
|
|
621
|
+
if (!isRecord(parsed)
|
|
622
|
+
|| parsed.ok !== true
|
|
623
|
+
|| !isRecord(parsed.data)
|
|
624
|
+
|| !Array.isArray(parsed.data.cleaned)
|
|
625
|
+
|| parsed.data.cleaned.some((id) => typeof id !== "string")) {
|
|
626
|
+
throw new Error("Controller reconciliation returned an invalid structured result.");
|
|
627
|
+
}
|
|
628
|
+
}
|
|
629
|
+
function structuredErrorMessage(result) {
|
|
630
|
+
for (const buffer of [result.stderr, result.stdout]) {
|
|
631
|
+
try {
|
|
632
|
+
const value = JSON.parse(buffer.toString("utf8"));
|
|
633
|
+
if (isRecord(value) && typeof value.message === "string" && value.message.length > 0) {
|
|
634
|
+
return value.message;
|
|
635
|
+
}
|
|
636
|
+
}
|
|
637
|
+
catch {
|
|
638
|
+
// Fall back to the child's raw stderr below.
|
|
639
|
+
}
|
|
640
|
+
}
|
|
641
|
+
return undefined;
|
|
642
|
+
}
|
|
589
643
|
function parseControllerIdentity(value) {
|
|
590
644
|
if (typeof value.executablePath !== "string"
|
|
591
645
|
|| value.executablePath.length === 0
|
package/dist/cli.js
CHANGED
|
@@ -301,6 +301,29 @@ export async function main() {
|
|
|
301
301
|
}
|
|
302
302
|
if (args[0] === "controller") {
|
|
303
303
|
const method = args[1];
|
|
304
|
+
if (method === "live-identity" && args.length === 2) {
|
|
305
|
+
try {
|
|
306
|
+
const identity = await callController(home, "controller.identity", {});
|
|
307
|
+
emit("", false, identity);
|
|
308
|
+
}
|
|
309
|
+
catch (error) {
|
|
310
|
+
if (!(error instanceof ControllerClientError))
|
|
311
|
+
throw error;
|
|
312
|
+
if (jsonOutput) {
|
|
313
|
+
process.stderr.write(`${JSON.stringify({
|
|
314
|
+
ok: false,
|
|
315
|
+
code: error.code,
|
|
316
|
+
message: error.message,
|
|
317
|
+
details: {}
|
|
318
|
+
})}\n`);
|
|
319
|
+
}
|
|
320
|
+
else {
|
|
321
|
+
process.stderr.write(`RUNTIME_ERROR: ${error.message}\n`);
|
|
322
|
+
}
|
|
323
|
+
process.exitCode = 5;
|
|
324
|
+
}
|
|
325
|
+
return;
|
|
326
|
+
}
|
|
304
327
|
if (method === "identity" && args.length === 2) {
|
|
305
328
|
// Issue 02: the stable, read-only runtime identity receipt. It survives
|
|
306
329
|
// a Controller stop and answers build ID, package digest, backend, and
|
|
@@ -1,6 +1,8 @@
|
|
|
1
1
|
import { createHash } from "node:crypto";
|
|
2
2
|
import { builtinAgentDriverRegistry } from "../runtime/builtinAgentDrivers.js";
|
|
3
3
|
import { createRuntimeObservation, runtimeObservationFenceMatches, runtimeObservationFromTaskEvent } from "../runtime/runtimeObservation.js";
|
|
4
|
+
const DEFAULT_MAX_CONCURRENT_SAMPLES = 8;
|
|
5
|
+
const MAX_CONCURRENT_SAMPLES = 64;
|
|
4
6
|
/**
|
|
5
7
|
* Controller-owned, provider-independent sampler. Drivers own source parsing;
|
|
6
8
|
* this component owns active-Run discovery, cursor lifetime, canonical event
|
|
@@ -11,11 +13,13 @@ export class AgentRuntimeObserver {
|
|
|
11
13
|
inbox;
|
|
12
14
|
drivers;
|
|
13
15
|
#states = new Map();
|
|
16
|
+
#maxConcurrentSamples;
|
|
14
17
|
#sequence = 0;
|
|
15
|
-
constructor(store, inbox, drivers = builtinAgentDriverRegistry()) {
|
|
18
|
+
constructor(store, inbox, drivers = builtinAgentDriverRegistry(), options = {}) {
|
|
16
19
|
this.store = store;
|
|
17
20
|
this.inbox = inbox;
|
|
18
21
|
this.drivers = drivers;
|
|
22
|
+
this.#maxConcurrentSamples = sampleConcurrency(options.maxConcurrentSamples);
|
|
19
23
|
}
|
|
20
24
|
async sample(now = new Date()) {
|
|
21
25
|
const active = this.activeSources();
|
|
@@ -25,7 +29,9 @@ export class AgentRuntimeObserver {
|
|
|
25
29
|
this.#states.delete(key);
|
|
26
30
|
}
|
|
27
31
|
const dirty = new Set();
|
|
28
|
-
|
|
32
|
+
const sequenceBase = this.#sequence;
|
|
33
|
+
this.#sequence += active.length;
|
|
34
|
+
await forEachConcurrent(active, this.#maxConcurrentSamples, async ({ key, fence, source, freshSession, persistedState }, index) => {
|
|
29
35
|
const existingState = this.#states.get(key);
|
|
30
36
|
// Cursor state is intentionally process-local, but the latest canonical
|
|
31
37
|
// usage/activity baseline is durable. Rehydrate it after Controller
|
|
@@ -49,7 +55,10 @@ export class AgentRuntimeObserver {
|
|
|
49
55
|
}
|
|
50
56
|
state.cursor = sample.cursor;
|
|
51
57
|
const at = now.toISOString();
|
|
52
|
-
|
|
58
|
+
// Reserve sequence numbers in source-key order before sampling. Provider
|
|
59
|
+
// latency can change completion order without changing observation
|
|
60
|
+
// identity or the canonical sequence assigned to a source.
|
|
61
|
+
const sequence = sequenceBase + index;
|
|
53
62
|
if (existingState === undefined && freshSession && state.usage === undefined) {
|
|
54
63
|
const zero = Object.freeze({ inputTokens: 0, outputTokens: 0 });
|
|
55
64
|
this.inbox.enqueueObservation(createRuntimeObservation({
|
|
@@ -114,28 +123,52 @@ export class AgentRuntimeObserver {
|
|
|
114
123
|
if (sample.activityId !== undefined)
|
|
115
124
|
state.activityId = sample.activityId;
|
|
116
125
|
this.#states.set(key, state);
|
|
117
|
-
})
|
|
118
|
-
return Object.freeze([...dirty].sort());
|
|
126
|
+
});
|
|
127
|
+
return Object.freeze([...dirty].sort(numericCompare));
|
|
119
128
|
}
|
|
120
129
|
activeSources() {
|
|
121
130
|
const result = [];
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
127
|
-
|
|
131
|
+
const indexedTaskIds = this.store.listActiveTaskIds?.();
|
|
132
|
+
// FileTaskStore remains a development/compatibility fallback. The normal
|
|
133
|
+
// Controller store is SQLite and must discover only its indexed hot set.
|
|
134
|
+
const activeTasks = indexedTaskIds === undefined
|
|
135
|
+
? this.store.listTasks().filter((task) => task.status === "active")
|
|
136
|
+
: [...new Set(indexedTaskIds)]
|
|
137
|
+
.sort(numericCompare)
|
|
138
|
+
.map((taskId) => this.store.getTask(taskId))
|
|
139
|
+
.filter((task) => (task !== null && task.status === "active"));
|
|
140
|
+
for (const task of activeTasks) {
|
|
141
|
+
// A Task still incurs one O(E) event projection. Group those observations
|
|
142
|
+
// by Run and sort each group once so every active Run can reuse the same
|
|
143
|
+
// ordered slice instead of repeatedly filtering/sorting all E events.
|
|
144
|
+
const observationsByRunId = new Map();
|
|
145
|
+
for (const event of this.store.listEvents(task.id)) {
|
|
146
|
+
const observation = runtimeObservationFromTaskEvent(event);
|
|
147
|
+
const runId = observation?.fence.runId;
|
|
148
|
+
if (observation === null || runId === undefined)
|
|
149
|
+
continue;
|
|
150
|
+
const grouped = observationsByRunId.get(runId);
|
|
151
|
+
if (grouped === undefined) {
|
|
152
|
+
observationsByRunId.set(runId, [observation]);
|
|
153
|
+
}
|
|
154
|
+
else {
|
|
155
|
+
grouped.push(observation);
|
|
156
|
+
}
|
|
157
|
+
}
|
|
158
|
+
for (const observations of observationsByRunId.values()) {
|
|
159
|
+
observations.sort(compareObservations);
|
|
160
|
+
}
|
|
128
161
|
for (const run of this.store.listAgentRuns(task.id)) {
|
|
129
162
|
if (run.status !== "active"
|
|
130
163
|
|| this.store.getActiveAgentRun(task.id, run.roleName)?.id !== run.id)
|
|
131
164
|
continue;
|
|
165
|
+
const observations = observationsByRunId.get(run.id) ?? [];
|
|
132
166
|
const accepted = observations
|
|
133
167
|
.filter((observation) => observation.kind === "turn.accepted"
|
|
134
168
|
&& observation.fence.runId === run.id
|
|
135
169
|
&& observation.fence.roleName === run.roleName
|
|
136
170
|
&& observation.fence.agentId === run.effective.agentId
|
|
137
171
|
&& observation.payload.observerSource !== undefined)
|
|
138
|
-
.sort(compareObservations)
|
|
139
172
|
.at(-1);
|
|
140
173
|
const source = accepted?.payload.observerSource;
|
|
141
174
|
if (accepted === undefined || source === undefined
|
|
@@ -150,19 +183,23 @@ export class AgentRuntimeObserver {
|
|
|
150
183
|
}
|
|
151
184
|
const fence = accepted.fence;
|
|
152
185
|
const exact = observations
|
|
153
|
-
.filter((observation) => runtimeObservationFenceMatches(fence, observation.fence))
|
|
154
|
-
.sort(compareObservations);
|
|
186
|
+
.filter((observation) => runtimeObservationFenceMatches(fence, observation.fence));
|
|
155
187
|
const persistedUsage = exact.filter((observation) => (observation.kind === "activity.observed"
|
|
156
188
|
&& observation.payload.usage !== undefined)).at(-1);
|
|
157
189
|
const persistedHealth = exact.filter((observation) => (observation.kind === "observer.health"
|
|
158
190
|
&& observation.payload.sourceId === source.sourceId)).at(-1);
|
|
159
191
|
result.push(Object.freeze({
|
|
160
192
|
key: JSON.stringify([
|
|
193
|
+
fence.taskId,
|
|
194
|
+
fence.roleName,
|
|
195
|
+
fence.runId,
|
|
196
|
+
fence.agentId,
|
|
161
197
|
fence.driverId,
|
|
198
|
+
fence.launchId,
|
|
162
199
|
fence.sessionGenerationId,
|
|
163
200
|
fence.nativeSessionId,
|
|
164
201
|
fence.nativeTurnId,
|
|
165
|
-
fence.
|
|
202
|
+
fence.receiptId,
|
|
166
203
|
source.sourceId
|
|
167
204
|
]),
|
|
168
205
|
fence,
|
|
@@ -187,8 +224,31 @@ export class AgentRuntimeObserver {
|
|
|
187
224
|
}));
|
|
188
225
|
}
|
|
189
226
|
}
|
|
190
|
-
return result;
|
|
227
|
+
return Object.freeze(result.sort((left, right) => numericCompare(left.key, right.key)));
|
|
228
|
+
}
|
|
229
|
+
}
|
|
230
|
+
async function forEachConcurrent(values, limit, visit) {
|
|
231
|
+
let nextIndex = 0;
|
|
232
|
+
const workers = Math.min(limit, values.length);
|
|
233
|
+
await Promise.all(Array.from({ length: workers }, async () => {
|
|
234
|
+
while (nextIndex < values.length) {
|
|
235
|
+
const index = nextIndex;
|
|
236
|
+
nextIndex += 1;
|
|
237
|
+
await visit(values[index], index);
|
|
238
|
+
}
|
|
239
|
+
}));
|
|
240
|
+
}
|
|
241
|
+
function sampleConcurrency(value) {
|
|
242
|
+
const resolved = value ?? DEFAULT_MAX_CONCURRENT_SAMPLES;
|
|
243
|
+
if (!Number.isSafeInteger(resolved)
|
|
244
|
+
|| resolved < 1
|
|
245
|
+
|| resolved > MAX_CONCURRENT_SAMPLES) {
|
|
246
|
+
throw new Error(`Agent runtime observer maxConcurrentSamples must be between 1 and ${MAX_CONCURRENT_SAMPLES}.`);
|
|
191
247
|
}
|
|
248
|
+
return resolved;
|
|
249
|
+
}
|
|
250
|
+
function numericCompare(left, right) {
|
|
251
|
+
return left.localeCompare(right, undefined, { numeric: true });
|
|
192
252
|
}
|
|
193
253
|
function compareObservations(left, right) {
|
|
194
254
|
return left.receivedAt.localeCompare(right.receivedAt)
|