@evident-ai/cli 3.1.1-dev.2f78b33 → 3.1.1-dev.35e0a15
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 +15 -11
- package/dist/index.js +1233 -96
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -11,8 +11,8 @@ import chalk2 from "chalk";
|
|
|
11
11
|
|
|
12
12
|
// src/lib/config.ts
|
|
13
13
|
import Conf from "conf";
|
|
14
|
-
import {
|
|
15
|
-
import {
|
|
14
|
+
import { chmodSync, existsSync, statSync } from "fs";
|
|
15
|
+
import { dirname } from "path";
|
|
16
16
|
var PRODUCTION_API_URL = "https://api.production.evident.run/v1";
|
|
17
17
|
var PRODUCTION_TUNNEL_URL = "wss://tunnel.production.evident.run";
|
|
18
18
|
var defaults = {
|
|
@@ -47,8 +47,35 @@ var credentials = new Conf({
|
|
|
47
47
|
projectName: "evident",
|
|
48
48
|
projectSuffix: "",
|
|
49
49
|
configName: "credentials",
|
|
50
|
-
defaults: {}
|
|
50
|
+
defaults: {},
|
|
51
|
+
configFileMode: 384
|
|
51
52
|
});
|
|
53
|
+
var CREDENTIALS_FILE_MODE = 384;
|
|
54
|
+
var CREDENTIALS_DIR_MODE = 448;
|
|
55
|
+
var permissionWarningEmitted = false;
|
|
56
|
+
function hardenCredentialsPermissions() {
|
|
57
|
+
if (process.platform === "win32") {
|
|
58
|
+
return;
|
|
59
|
+
}
|
|
60
|
+
const file = credentials.path;
|
|
61
|
+
for (const [path, mode] of [
|
|
62
|
+
[file, CREDENTIALS_FILE_MODE],
|
|
63
|
+
[dirname(file), CREDENTIALS_DIR_MODE]
|
|
64
|
+
]) {
|
|
65
|
+
try {
|
|
66
|
+
if (existsSync(path) && (statSync(path).mode & 511) !== mode) {
|
|
67
|
+
chmodSync(path, mode);
|
|
68
|
+
}
|
|
69
|
+
} catch (err) {
|
|
70
|
+
if (!permissionWarningEmitted) {
|
|
71
|
+
permissionWarningEmitted = true;
|
|
72
|
+
console.error(
|
|
73
|
+
`[config] could not restrict permissions on ${path}; the credentials file may be readable by other users on this machine: ${err instanceof Error ? err.message : String(err)}`
|
|
74
|
+
);
|
|
75
|
+
}
|
|
76
|
+
}
|
|
77
|
+
}
|
|
78
|
+
}
|
|
52
79
|
function getApiUrlConfig() {
|
|
53
80
|
return getApiUrl();
|
|
54
81
|
}
|
|
@@ -59,6 +86,7 @@ function credentialsKey() {
|
|
|
59
86
|
return getApiUrl();
|
|
60
87
|
}
|
|
61
88
|
function getCredentials() {
|
|
89
|
+
hardenCredentialsPermissions();
|
|
62
90
|
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
63
91
|
return byEndpoint[credentialsKey()] ?? {};
|
|
64
92
|
}
|
|
@@ -70,14 +98,17 @@ function setCredentials(creds) {
|
|
|
70
98
|
expiresAt: creds.expiresAt
|
|
71
99
|
};
|
|
72
100
|
credentials.set("byEndpoint", byEndpoint);
|
|
101
|
+
hardenCredentialsPermissions();
|
|
73
102
|
}
|
|
74
103
|
function clearCredentials() {
|
|
75
104
|
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
76
105
|
delete byEndpoint[credentialsKey()];
|
|
77
106
|
credentials.set("byEndpoint", byEndpoint);
|
|
107
|
+
hardenCredentialsPermissions();
|
|
78
108
|
}
|
|
79
109
|
function clearAllCredentials() {
|
|
80
110
|
credentials.clear();
|
|
111
|
+
hardenCredentialsPermissions();
|
|
81
112
|
}
|
|
82
113
|
function getCliName() {
|
|
83
114
|
const argv1 = process.argv[1] || "";
|
|
@@ -285,14 +316,14 @@ function blank() {
|
|
|
285
316
|
console.log();
|
|
286
317
|
}
|
|
287
318
|
function waitForEnter(prompt = "Press Enter to continue...") {
|
|
288
|
-
return new Promise((
|
|
319
|
+
return new Promise((resolve3) => {
|
|
289
320
|
process.stdout.write(chalk.dim(prompt));
|
|
290
321
|
const handler = () => {
|
|
291
322
|
process.stdin.removeListener("data", handler);
|
|
292
323
|
process.stdin.setRawMode?.(false);
|
|
293
324
|
process.stdin.pause();
|
|
294
325
|
console.log();
|
|
295
|
-
|
|
326
|
+
resolve3();
|
|
296
327
|
};
|
|
297
328
|
if (process.stdin.isTTY) {
|
|
298
329
|
process.stdin.setRawMode?.(true);
|
|
@@ -302,7 +333,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
|
|
|
302
333
|
});
|
|
303
334
|
}
|
|
304
335
|
function sleep(ms) {
|
|
305
|
-
return new Promise((
|
|
336
|
+
return new Promise((resolve3) => setTimeout(resolve3, ms));
|
|
306
337
|
}
|
|
307
338
|
|
|
308
339
|
// src/commands/login.ts
|
|
@@ -373,22 +404,23 @@ async function deviceFlowLogin(options) {
|
|
|
373
404
|
}
|
|
374
405
|
async function tokenLogin() {
|
|
375
406
|
console.log("Token login mode.");
|
|
376
|
-
console.log("
|
|
407
|
+
console.log("Run `evident login` on a machine with a browser to get a token.");
|
|
408
|
+
console.log("Manage or revoke existing tokens under Settings \u2192 CLI tokens.");
|
|
377
409
|
blank();
|
|
378
410
|
process.stdout.write("Paste token: ");
|
|
379
|
-
const token = await new Promise((
|
|
411
|
+
const token = await new Promise((resolve3) => {
|
|
380
412
|
let data = "";
|
|
381
413
|
process.stdin.setEncoding("utf8");
|
|
382
414
|
process.stdin.on("data", (chunk) => {
|
|
383
415
|
data += chunk;
|
|
384
416
|
});
|
|
385
417
|
process.stdin.on("end", () => {
|
|
386
|
-
|
|
418
|
+
resolve3(data.trim());
|
|
387
419
|
});
|
|
388
420
|
if (process.stdin.isTTY) {
|
|
389
421
|
process.stdin.once("data", (chunk) => {
|
|
390
422
|
process.stdin.pause();
|
|
391
|
-
|
|
423
|
+
resolve3(chunk.toString().trim());
|
|
392
424
|
});
|
|
393
425
|
process.stdin.resume();
|
|
394
426
|
}
|
|
@@ -467,9 +499,9 @@ async function whoami() {
|
|
|
467
499
|
}
|
|
468
500
|
|
|
469
501
|
// src/commands/run.ts
|
|
502
|
+
import { homedir as homedir2 } from "os";
|
|
503
|
+
import { isAbsolute as isAbsolute2, join as join2, parse, resolve as resolvePath } from "path";
|
|
470
504
|
import chalk6 from "chalk";
|
|
471
|
-
import ora3 from "ora";
|
|
472
|
-
import { select as select3 } from "@inquirer/prompts";
|
|
473
505
|
|
|
474
506
|
// ../../packages/types/src/telemetry/index.ts
|
|
475
507
|
var TelemetryEventTypes = {
|
|
@@ -485,6 +517,10 @@ var TelemetryEventTypes = {
|
|
|
485
517
|
var MAX_FRAME_BYTES = 256 * 1024;
|
|
486
518
|
var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
|
|
487
519
|
|
|
520
|
+
// ../../packages/types/src/runner-files.ts
|
|
521
|
+
var MAX_FILE_PUSH_BYTES = 64 * 1024;
|
|
522
|
+
var MAX_FILE_SYNC_DIRECTORIES = 16;
|
|
523
|
+
|
|
488
524
|
// ../../packages/types/src/logging/index.ts
|
|
489
525
|
var CORRELATION_ID_HEADER = "x-evident-correlation-id";
|
|
490
526
|
function log(level, event, fields) {
|
|
@@ -514,6 +550,10 @@ function stripQuery(url) {
|
|
|
514
550
|
}
|
|
515
551
|
}
|
|
516
552
|
|
|
553
|
+
// src/commands/run.ts
|
|
554
|
+
import ora3 from "ora";
|
|
555
|
+
import { select as select3 } from "@inquirer/prompts";
|
|
556
|
+
|
|
517
557
|
// src/lib/telemetry.ts
|
|
518
558
|
var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
|
|
519
559
|
function getCliVersion() {
|
|
@@ -725,7 +765,7 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
|
725
765
|
if (health.healthy) {
|
|
726
766
|
return health;
|
|
727
767
|
}
|
|
728
|
-
await new Promise((
|
|
768
|
+
await new Promise((resolve3) => setTimeout(resolve3, 1e3));
|
|
729
769
|
}
|
|
730
770
|
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
731
771
|
}
|
|
@@ -1364,7 +1404,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
|
|
|
1364
1404
|
}
|
|
1365
1405
|
}
|
|
1366
1406
|
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1367
|
-
await new Promise((
|
|
1407
|
+
await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
|
|
1368
1408
|
}
|
|
1369
1409
|
}
|
|
1370
1410
|
return null;
|
|
@@ -1492,6 +1532,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1492
1532
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1493
1533
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1494
1534
|
}
|
|
1535
|
+
function isB2AbandonmentConfirmed(params) {
|
|
1536
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
1537
|
+
}
|
|
1495
1538
|
function messageError(messages, userMessageId) {
|
|
1496
1539
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1497
1540
|
const error2 = errorOf(reply);
|
|
@@ -1505,6 +1548,42 @@ function messageError(messages, userMessageId) {
|
|
|
1505
1548
|
}
|
|
1506
1549
|
return "The agent run failed.";
|
|
1507
1550
|
}
|
|
1551
|
+
function messageFailure(messages, userMessageId) {
|
|
1552
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1553
|
+
const error2 = errorOf(reply);
|
|
1554
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
1555
|
+
const e = error2;
|
|
1556
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
1557
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
1558
|
+
if (e.name === "ProviderAuthError") {
|
|
1559
|
+
const data = e.data;
|
|
1560
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
1561
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
1562
|
+
}
|
|
1563
|
+
if (e.name === "APIError") {
|
|
1564
|
+
const data = e.data;
|
|
1565
|
+
const statusCode = data?.statusCode;
|
|
1566
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
1567
|
+
return {
|
|
1568
|
+
kind: "model_auth",
|
|
1569
|
+
providerId: replyProviderId,
|
|
1570
|
+
modelId: replyModelId,
|
|
1571
|
+
reason: "rejected"
|
|
1572
|
+
};
|
|
1573
|
+
}
|
|
1574
|
+
}
|
|
1575
|
+
return null;
|
|
1576
|
+
}
|
|
1577
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
1578
|
+
if (classified != null) return classified;
|
|
1579
|
+
if (hasConfiguredProvider !== false) return null;
|
|
1580
|
+
return {
|
|
1581
|
+
kind: "model_auth",
|
|
1582
|
+
providerId: replyProviderId,
|
|
1583
|
+
modelId: replyModelId,
|
|
1584
|
+
reason: "missing"
|
|
1585
|
+
};
|
|
1586
|
+
}
|
|
1508
1587
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1509
1588
|
if (!messages || messages.length === 0) return false;
|
|
1510
1589
|
return messages.some(
|
|
@@ -1738,12 +1817,12 @@ var StreamForwarder = class {
|
|
|
1738
1817
|
let endBody;
|
|
1739
1818
|
if (has_body) {
|
|
1740
1819
|
const chunks = [];
|
|
1741
|
-
bodyPromise = new Promise((
|
|
1820
|
+
bodyPromise = new Promise((resolve3) => {
|
|
1742
1821
|
pushBody = (buf) => {
|
|
1743
1822
|
chunks.push(buf);
|
|
1744
1823
|
};
|
|
1745
1824
|
endBody = () => {
|
|
1746
|
-
|
|
1825
|
+
resolve3(Buffer.concat(chunks));
|
|
1747
1826
|
};
|
|
1748
1827
|
});
|
|
1749
1828
|
}
|
|
@@ -1860,7 +1939,7 @@ function connectTunnel(options) {
|
|
|
1860
1939
|
} = options;
|
|
1861
1940
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1862
1941
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1863
|
-
return new Promise((
|
|
1942
|
+
return new Promise((resolve3, reject) => {
|
|
1864
1943
|
const ws = new WebSocket2(url, {
|
|
1865
1944
|
headers: {
|
|
1866
1945
|
Authorization: authHeader
|
|
@@ -1915,7 +1994,7 @@ function connectTunnel(options) {
|
|
|
1915
1994
|
clearTimeout(connectionTimeout);
|
|
1916
1995
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1917
1996
|
onConnected?.(connectedAgentId);
|
|
1918
|
-
|
|
1997
|
+
resolve3({
|
|
1919
1998
|
ws,
|
|
1920
1999
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1921
2000
|
});
|
|
@@ -2033,6 +2112,416 @@ var RunnerConnection = class {
|
|
|
2033
2112
|
}
|
|
2034
2113
|
};
|
|
2035
2114
|
|
|
2115
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2116
|
+
import { writeFileSync } from "fs";
|
|
2117
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2118
|
+
try {
|
|
2119
|
+
writeFileSync(path, `${agentId}
|
|
2120
|
+
`);
|
|
2121
|
+
return { ok: true };
|
|
2122
|
+
} catch (error2) {
|
|
2123
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2124
|
+
}
|
|
2125
|
+
}
|
|
2126
|
+
|
|
2127
|
+
// src/lib/channels/driver.ts
|
|
2128
|
+
import { homedir } from "os";
|
|
2129
|
+
|
|
2130
|
+
// src/lib/file-push.ts
|
|
2131
|
+
import { randomUUID } from "crypto";
|
|
2132
|
+
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2133
|
+
import { basename, dirname as dirname2, isAbsolute, join, relative, resolve as resolve2, sep } from "path";
|
|
2134
|
+
var FILE_MODE = 384;
|
|
2135
|
+
var DIRECTORY_MODE = 448;
|
|
2136
|
+
async function writePushedFile(request) {
|
|
2137
|
+
const { requestedPath, content, allowedDirectories, homeDir } = request;
|
|
2138
|
+
const bytes = content.byteLength;
|
|
2139
|
+
if (allowedDirectories.length === 0) {
|
|
2140
|
+
return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
|
|
2141
|
+
path: requestedPath,
|
|
2142
|
+
bytes
|
|
2143
|
+
});
|
|
2144
|
+
}
|
|
2145
|
+
if (bytes > MAX_FILE_PUSH_BYTES) {
|
|
2146
|
+
return refuse(
|
|
2147
|
+
"file_too_large",
|
|
2148
|
+
`File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
|
|
2149
|
+
{
|
|
2150
|
+
path: requestedPath,
|
|
2151
|
+
bytes
|
|
2152
|
+
}
|
|
2153
|
+
);
|
|
2154
|
+
}
|
|
2155
|
+
const candidate = expandAndValidate(requestedPath, homeDir);
|
|
2156
|
+
if (candidate === null) {
|
|
2157
|
+
return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
|
|
2158
|
+
path: requestedPath,
|
|
2159
|
+
bytes
|
|
2160
|
+
});
|
|
2161
|
+
}
|
|
2162
|
+
try {
|
|
2163
|
+
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2164
|
+
dirname2(candidate)
|
|
2165
|
+
);
|
|
2166
|
+
const realTarget = join(existingAncestor, ...missingSegments, basename(candidate));
|
|
2167
|
+
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2168
|
+
if (allowedDirectory === null) {
|
|
2169
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2170
|
+
path: realTarget,
|
|
2171
|
+
bytes
|
|
2172
|
+
});
|
|
2173
|
+
}
|
|
2174
|
+
if (missingSegments.length > 0) {
|
|
2175
|
+
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2176
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2177
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2178
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2179
|
+
path: realTarget,
|
|
2180
|
+
bytes,
|
|
2181
|
+
reason: "parent_changed_after_create"
|
|
2182
|
+
});
|
|
2183
|
+
}
|
|
2184
|
+
}
|
|
2185
|
+
await writeAtomically(realTarget, content);
|
|
2186
|
+
log("info", "file_push_written", { path: realTarget, bytes });
|
|
2187
|
+
return { ok: true, path: realTarget };
|
|
2188
|
+
} catch (err) {
|
|
2189
|
+
const errno = err.code ?? "UNKNOWN";
|
|
2190
|
+
return refuse("write_failed", `The runner could not write the file (${errno}).`, {
|
|
2191
|
+
path: candidate,
|
|
2192
|
+
bytes,
|
|
2193
|
+
errno,
|
|
2194
|
+
...errorFields(err)
|
|
2195
|
+
});
|
|
2196
|
+
}
|
|
2197
|
+
}
|
|
2198
|
+
function expandAndValidate(requestedPath, homeDir) {
|
|
2199
|
+
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2200
|
+
return null;
|
|
2201
|
+
}
|
|
2202
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2203
|
+
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2204
|
+
return null;
|
|
2205
|
+
}
|
|
2206
|
+
if (!isAbsolute(expanded)) {
|
|
2207
|
+
return null;
|
|
2208
|
+
}
|
|
2209
|
+
const candidate = resolve2(expanded);
|
|
2210
|
+
const name = basename(candidate);
|
|
2211
|
+
return name === "" || name === "." || name === ".." ? null : candidate;
|
|
2212
|
+
}
|
|
2213
|
+
async function resolveNearestExistingAncestor(directory) {
|
|
2214
|
+
const missingSegments = [];
|
|
2215
|
+
let current = directory;
|
|
2216
|
+
for (; ; ) {
|
|
2217
|
+
try {
|
|
2218
|
+
return { existingAncestor: await realpath(current), missingSegments };
|
|
2219
|
+
} catch (err) {
|
|
2220
|
+
const parent = dirname2(current);
|
|
2221
|
+
if (err.code !== "ENOENT" || parent === current) {
|
|
2222
|
+
throw err;
|
|
2223
|
+
}
|
|
2224
|
+
missingSegments.unshift(basename(current));
|
|
2225
|
+
current = parent;
|
|
2226
|
+
}
|
|
2227
|
+
}
|
|
2228
|
+
}
|
|
2229
|
+
async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
|
|
2230
|
+
for (const directory of allowedDirectories) {
|
|
2231
|
+
if (!isAbsolute(directory)) {
|
|
2232
|
+
log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
|
|
2233
|
+
continue;
|
|
2234
|
+
}
|
|
2235
|
+
const realDirectory = await realpathCreatingIfMissing(directory);
|
|
2236
|
+
if (realDirectory !== null && contains(realDirectory, realTarget)) {
|
|
2237
|
+
return realDirectory;
|
|
2238
|
+
}
|
|
2239
|
+
}
|
|
2240
|
+
return null;
|
|
2241
|
+
}
|
|
2242
|
+
async function realpathCreatingIfMissing(directory) {
|
|
2243
|
+
try {
|
|
2244
|
+
return await realpath(directory);
|
|
2245
|
+
} catch (err) {
|
|
2246
|
+
if (err.code !== "ENOENT") {
|
|
2247
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2248
|
+
directory,
|
|
2249
|
+
reason: "unresolvable",
|
|
2250
|
+
...errorFields(err)
|
|
2251
|
+
});
|
|
2252
|
+
return null;
|
|
2253
|
+
}
|
|
2254
|
+
}
|
|
2255
|
+
try {
|
|
2256
|
+
await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
|
|
2257
|
+
await chmod(directory, DIRECTORY_MODE);
|
|
2258
|
+
return await realpath(directory);
|
|
2259
|
+
} catch (err) {
|
|
2260
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2261
|
+
directory,
|
|
2262
|
+
reason: "create_failed",
|
|
2263
|
+
...errorFields(err)
|
|
2264
|
+
});
|
|
2265
|
+
return null;
|
|
2266
|
+
}
|
|
2267
|
+
}
|
|
2268
|
+
function contains(realDirectory, realTarget) {
|
|
2269
|
+
const rel = relative(realDirectory, realTarget);
|
|
2270
|
+
return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
|
|
2271
|
+
}
|
|
2272
|
+
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2273
|
+
let current = existingAncestor;
|
|
2274
|
+
for (const segment of missingSegments) {
|
|
2275
|
+
current = join(current, segment);
|
|
2276
|
+
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2277
|
+
await chmod(current, DIRECTORY_MODE);
|
|
2278
|
+
}
|
|
2279
|
+
}
|
|
2280
|
+
async function writeAtomically(realTarget, content) {
|
|
2281
|
+
const temporaryPath = join(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2282
|
+
let handle;
|
|
2283
|
+
try {
|
|
2284
|
+
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
2285
|
+
await handle.writeFile(content);
|
|
2286
|
+
await handle.chmod(FILE_MODE);
|
|
2287
|
+
await handle.close();
|
|
2288
|
+
handle = void 0;
|
|
2289
|
+
await rename(temporaryPath, realTarget);
|
|
2290
|
+
} catch (err) {
|
|
2291
|
+
await discardTemporaryFile(temporaryPath, handle);
|
|
2292
|
+
throw err;
|
|
2293
|
+
}
|
|
2294
|
+
}
|
|
2295
|
+
async function discardTemporaryFile(temporaryPath, handle) {
|
|
2296
|
+
try {
|
|
2297
|
+
await handle?.close();
|
|
2298
|
+
} catch (err) {
|
|
2299
|
+
log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2300
|
+
}
|
|
2301
|
+
try {
|
|
2302
|
+
await unlink(temporaryPath);
|
|
2303
|
+
} catch (err) {
|
|
2304
|
+
const errno = err.code;
|
|
2305
|
+
if (errno !== "ENOENT" && errno !== "ENOTDIR") {
|
|
2306
|
+
log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2307
|
+
}
|
|
2308
|
+
}
|
|
2309
|
+
}
|
|
2310
|
+
function refuse(code, message, fields) {
|
|
2311
|
+
log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
|
|
2312
|
+
return { ok: false, code, message };
|
|
2313
|
+
}
|
|
2314
|
+
|
|
2315
|
+
// src/lib/runner-file-sync.ts
|
|
2316
|
+
var MAX_ACK_ATTEMPTS = 5;
|
|
2317
|
+
async function syncPendingRunnerFiles(options) {
|
|
2318
|
+
const pending = await listPendingFiles(options);
|
|
2319
|
+
const pendingIds = new Set(pending.map((file) => file.id));
|
|
2320
|
+
for (const id of options.ackFailures.keys()) {
|
|
2321
|
+
if (!pendingIds.has(id)) options.ackFailures.delete(id);
|
|
2322
|
+
}
|
|
2323
|
+
if (pending.length === 0) return 0;
|
|
2324
|
+
options.log({
|
|
2325
|
+
level: "info",
|
|
2326
|
+
message: `Runner file sync: ${pending.length} file(s) queued for this runner`
|
|
2327
|
+
});
|
|
2328
|
+
let applied = 0;
|
|
2329
|
+
for (const file of pending) {
|
|
2330
|
+
if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
|
|
2331
|
+
if (await applyOne(options, file)) applied += 1;
|
|
2332
|
+
}
|
|
2333
|
+
return applied;
|
|
2334
|
+
}
|
|
2335
|
+
async function listPendingFiles(options) {
|
|
2336
|
+
let res;
|
|
2337
|
+
try {
|
|
2338
|
+
res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
|
|
2339
|
+
headers: { Authorization: options.getAuthHeader() }
|
|
2340
|
+
});
|
|
2341
|
+
} catch (err) {
|
|
2342
|
+
options.log({
|
|
2343
|
+
level: "warn",
|
|
2344
|
+
message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
|
|
2345
|
+
});
|
|
2346
|
+
return [];
|
|
2347
|
+
}
|
|
2348
|
+
if (!res.ok) {
|
|
2349
|
+
options.log({
|
|
2350
|
+
level: res.status === 404 ? "debug" : "warn",
|
|
2351
|
+
message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
2352
|
+
});
|
|
2353
|
+
return [];
|
|
2354
|
+
}
|
|
2355
|
+
let body;
|
|
2356
|
+
try {
|
|
2357
|
+
body = await res.json();
|
|
2358
|
+
} catch (err) {
|
|
2359
|
+
options.log({
|
|
2360
|
+
level: "warn",
|
|
2361
|
+
message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
|
|
2362
|
+
});
|
|
2363
|
+
return [];
|
|
2364
|
+
}
|
|
2365
|
+
if (!Array.isArray(body)) {
|
|
2366
|
+
options.log({
|
|
2367
|
+
level: "warn",
|
|
2368
|
+
message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
|
|
2369
|
+
});
|
|
2370
|
+
return [];
|
|
2371
|
+
}
|
|
2372
|
+
const files = [];
|
|
2373
|
+
for (const entry of body) {
|
|
2374
|
+
const file = asPendingFile(entry);
|
|
2375
|
+
if (file === null) {
|
|
2376
|
+
options.log({
|
|
2377
|
+
level: "warn",
|
|
2378
|
+
message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
|
|
2379
|
+
});
|
|
2380
|
+
continue;
|
|
2381
|
+
}
|
|
2382
|
+
files.push(file);
|
|
2383
|
+
}
|
|
2384
|
+
return files;
|
|
2385
|
+
}
|
|
2386
|
+
function asPendingFile(entry) {
|
|
2387
|
+
if (entry === null || typeof entry !== "object") return null;
|
|
2388
|
+
const { id, path, size } = entry;
|
|
2389
|
+
if (typeof id !== "string" || id === "") return null;
|
|
2390
|
+
if (typeof path !== "string" || path === "") return null;
|
|
2391
|
+
if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
|
|
2392
|
+
return { id, path, size };
|
|
2393
|
+
}
|
|
2394
|
+
async function applyOne(options, file) {
|
|
2395
|
+
const label = `${file.id.slice(0, 8)} (${file.path})`;
|
|
2396
|
+
if (options.allowedDirectories.length === 0) {
|
|
2397
|
+
options.log({
|
|
2398
|
+
level: "warn",
|
|
2399
|
+
message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
|
|
2400
|
+
});
|
|
2401
|
+
await ack(options, file, "rejected", "file_sync_disabled");
|
|
2402
|
+
return false;
|
|
2403
|
+
}
|
|
2404
|
+
if (file.size > MAX_FILE_PUSH_BYTES) {
|
|
2405
|
+
options.log({
|
|
2406
|
+
level: "warn",
|
|
2407
|
+
message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
|
|
2408
|
+
});
|
|
2409
|
+
await ack(options, file, "rejected", "file_too_large");
|
|
2410
|
+
return false;
|
|
2411
|
+
}
|
|
2412
|
+
const download = await downloadContent(options, file, label);
|
|
2413
|
+
if (!download.ok) {
|
|
2414
|
+
if (download.terminal) await ack(options, file, "rejected", download.code);
|
|
2415
|
+
return false;
|
|
2416
|
+
}
|
|
2417
|
+
let outcome;
|
|
2418
|
+
try {
|
|
2419
|
+
outcome = await writePushedFile({
|
|
2420
|
+
requestedPath: file.path,
|
|
2421
|
+
content: download.content,
|
|
2422
|
+
allowedDirectories: options.allowedDirectories,
|
|
2423
|
+
homeDir: options.homeDir
|
|
2424
|
+
});
|
|
2425
|
+
} catch (err) {
|
|
2426
|
+
options.log({
|
|
2427
|
+
level: "error",
|
|
2428
|
+
message: `Runner file ${label} could not be written: ${describe(err)}`
|
|
2429
|
+
});
|
|
2430
|
+
await ack(options, file, "rejected", "write_failed");
|
|
2431
|
+
return false;
|
|
2432
|
+
}
|
|
2433
|
+
if (!outcome.ok) {
|
|
2434
|
+
options.log({
|
|
2435
|
+
level: "warn",
|
|
2436
|
+
message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
|
|
2437
|
+
});
|
|
2438
|
+
await ack(options, file, "rejected", outcome.code);
|
|
2439
|
+
return false;
|
|
2440
|
+
}
|
|
2441
|
+
options.log({
|
|
2442
|
+
level: "info",
|
|
2443
|
+
message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
|
|
2444
|
+
});
|
|
2445
|
+
await ack(options, file, "applied");
|
|
2446
|
+
return true;
|
|
2447
|
+
}
|
|
2448
|
+
function durableDownloadCode(status) {
|
|
2449
|
+
return status === 413 ? "file_too_large" : "write_failed";
|
|
2450
|
+
}
|
|
2451
|
+
async function downloadContent(options, file, label) {
|
|
2452
|
+
try {
|
|
2453
|
+
const res = await options.fetchImpl(
|
|
2454
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
|
|
2455
|
+
{ headers: { Authorization: options.getAuthHeader() } }
|
|
2456
|
+
);
|
|
2457
|
+
if (!res.ok) {
|
|
2458
|
+
const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
|
|
2459
|
+
if (!terminal) {
|
|
2460
|
+
options.log({
|
|
2461
|
+
level: "warn",
|
|
2462
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
2463
|
+
});
|
|
2464
|
+
return { ok: false, terminal: false };
|
|
2465
|
+
}
|
|
2466
|
+
const code = durableDownloadCode(res.status);
|
|
2467
|
+
options.log({
|
|
2468
|
+
level: "error",
|
|
2469
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
|
|
2470
|
+
});
|
|
2471
|
+
return { ok: false, terminal: true, code };
|
|
2472
|
+
}
|
|
2473
|
+
return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
|
|
2474
|
+
} catch (err) {
|
|
2475
|
+
options.log({
|
|
2476
|
+
level: "warn",
|
|
2477
|
+
message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
|
|
2478
|
+
});
|
|
2479
|
+
return { ok: false, terminal: false };
|
|
2480
|
+
}
|
|
2481
|
+
}
|
|
2482
|
+
async function ack(options, file, status, reason) {
|
|
2483
|
+
const outcome = `${status}${reason ? ` (${reason})` : ""}`;
|
|
2484
|
+
try {
|
|
2485
|
+
const res = await options.fetchImpl(
|
|
2486
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
2487
|
+
{
|
|
2488
|
+
method: "POST",
|
|
2489
|
+
headers: {
|
|
2490
|
+
Authorization: options.getAuthHeader(),
|
|
2491
|
+
"Content-Type": "application/json"
|
|
2492
|
+
},
|
|
2493
|
+
body: JSON.stringify(reason ? { status, reason } : { status })
|
|
2494
|
+
}
|
|
2495
|
+
);
|
|
2496
|
+
if (!res.ok) {
|
|
2497
|
+
recordAckFailure(
|
|
2498
|
+
options,
|
|
2499
|
+
file,
|
|
2500
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
|
|
2501
|
+
);
|
|
2502
|
+
return;
|
|
2503
|
+
}
|
|
2504
|
+
options.ackFailures.delete(file.id);
|
|
2505
|
+
} catch (err) {
|
|
2506
|
+
recordAckFailure(
|
|
2507
|
+
options,
|
|
2508
|
+
file,
|
|
2509
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
|
|
2510
|
+
);
|
|
2511
|
+
}
|
|
2512
|
+
}
|
|
2513
|
+
function recordAckFailure(options, file, what) {
|
|
2514
|
+
const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
|
|
2515
|
+
options.ackFailures.set(file.id, attempts);
|
|
2516
|
+
options.log({
|
|
2517
|
+
level: "error",
|
|
2518
|
+
message: attempts >= MAX_ACK_ATTEMPTS ? `${what} \u2014 giving up after ${attempts} attempts. It stays pending until the server expires it; restart the runner to retry.` : `${what} \u2014 it stays pending until a later drain re-acks it (attempt ${attempts} of ${MAX_ACK_ATTEMPTS})`
|
|
2519
|
+
});
|
|
2520
|
+
}
|
|
2521
|
+
function describe(err) {
|
|
2522
|
+
return err instanceof Error ? err.message : String(err);
|
|
2523
|
+
}
|
|
2524
|
+
|
|
2036
2525
|
// src/lib/channels/driver.ts
|
|
2037
2526
|
function messageIdOf(m) {
|
|
2038
2527
|
if (!m || typeof m !== "object") return void 0;
|
|
@@ -2061,7 +2550,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2061
2550
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2062
2551
|
var HEARTBEAT_MS = 6e4;
|
|
2063
2552
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
2553
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
2554
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2064
2555
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
2556
|
+
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
2065
2557
|
var ChannelAuthError = class extends Error {
|
|
2066
2558
|
constructor(message) {
|
|
2067
2559
|
super(message);
|
|
@@ -2098,8 +2590,38 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2098
2590
|
pausedMaxWaitMs;
|
|
2099
2591
|
stuckQueuedMs;
|
|
2100
2592
|
now;
|
|
2593
|
+
fileSyncDirectories;
|
|
2594
|
+
homeDir;
|
|
2101
2595
|
/** Cache of conversationId → opencode sessionId. */
|
|
2102
2596
|
sessions = /* @__PURE__ */ new Map();
|
|
2597
|
+
/**
|
|
2598
|
+
* conversationId → the opencode session this runner has ABANDONED as that
|
|
2599
|
+
* conversation's binding (#553), after a genuine (`sessionExists === true`)
|
|
2600
|
+
* dispatch failure: the session still exists but is wedged, so #485's self-heal
|
|
2601
|
+
* must bind a fresh one.
|
|
2602
|
+
*
|
|
2603
|
+
* Dropping the local binding + clearing the server row is not enough on its own:
|
|
2604
|
+
* a SIBLING message dispatched earlier in the same drain is still in-flight under
|
|
2605
|
+
* the same session, and its watcher's routine status writes carry
|
|
2606
|
+
* `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
|
|
2607
|
+
* and `ensureSession`'s persisted-id fallback then reuses it, defeating the
|
|
2608
|
+
* self-heal. This map makes the runner authoritative instead of racing those
|
|
2609
|
+
* writes: *`ensureSession` never reuses an abandoned id for that conversation,
|
|
2610
|
+
* whatever the server row says* — which holds even when the resurrecting write
|
|
2611
|
+
* is one we deliberately keep (see `markDone`).
|
|
2612
|
+
*
|
|
2613
|
+
* Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
|
|
2614
|
+
* one conversation hold ONE entry (the newest abandonment replaces the older), and
|
|
2615
|
+
* hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
|
|
2616
|
+
* NEWEST abandoned id per conversation is guarded: after a second abandonment a
|
|
2617
|
+
* late sibling of the FIRST session can write that id back and `ensureSession`
|
|
2618
|
+
* will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
|
|
2619
|
+
* NOT dropped when the session's watcher tears down: `markDone` still writes the
|
|
2620
|
+
* abandoned id back (it must, or the reply is lost), so the guard has to outlive
|
|
2621
|
+
* the turn that resurrects it. In-memory only — a restart forgets it, at the same
|
|
2622
|
+
* bounded cost.
|
|
2623
|
+
*/
|
|
2624
|
+
supersededSessions = /* @__PURE__ */ new Map();
|
|
2103
2625
|
/**
|
|
2104
2626
|
* Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
|
|
2105
2627
|
* longer idempotent (no caller-supplied `messageID`), and its read-back picks
|
|
@@ -2222,6 +2744,24 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2222
2744
|
sessionTitles = /* @__PURE__ */ new Map();
|
|
2223
2745
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
2224
2746
|
draining = false;
|
|
2747
|
+
/**
|
|
2748
|
+
* Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
|
|
2749
|
+
* drain ping don't download, write and ack the same file twice.
|
|
2750
|
+
*/
|
|
2751
|
+
syncingFiles = false;
|
|
2752
|
+
/**
|
|
2753
|
+
* Consecutive failed acks per pending file (#559). Lives on the driver so it
|
|
2754
|
+
* survives across drains — without it, a file whose ack keeps failing is
|
|
2755
|
+
* re-downloaded and re-written every ~2s until the server expires it.
|
|
2756
|
+
*/
|
|
2757
|
+
fileAckFailures = /* @__PURE__ */ new Map();
|
|
2758
|
+
/**
|
|
2759
|
+
* Monotonic count of files this runner has pulled and written (#559). Only
|
|
2760
|
+
* ever increases, so `run.ts` detects work by comparing it against the value
|
|
2761
|
+
* it saw on the previous cycle — including work that landed mid-sleep, the
|
|
2762
|
+
* same trick `lastProxiedActivityAt` uses.
|
|
2763
|
+
*/
|
|
2764
|
+
appliedFileCount = 0;
|
|
2225
2765
|
/**
|
|
2226
2766
|
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
2227
2767
|
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
@@ -2252,6 +2792,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2252
2792
|
this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
2253
2793
|
this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
2254
2794
|
this.now = config2.now ?? (() => Date.now());
|
|
2795
|
+
this.fileSyncDirectories = config2.fileSyncDirectories ?? [];
|
|
2796
|
+
this.homeDir = config2.homeDir ?? homedir();
|
|
2255
2797
|
}
|
|
2256
2798
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2257
2799
|
get opencodeBase() {
|
|
@@ -2279,6 +2821,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2279
2821
|
);
|
|
2280
2822
|
return run2;
|
|
2281
2823
|
}
|
|
2824
|
+
/**
|
|
2825
|
+
* Pull-and-apply any files Evident has queued for this runner (#559), riding
|
|
2826
|
+
* the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
|
|
2827
|
+
* and drain ping that call `drainPending()`. There is deliberately no channel,
|
|
2828
|
+
* control frame or poll loop of its own: worst-case latency is one poll tick.
|
|
2829
|
+
*
|
|
2830
|
+
* NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
|
|
2831
|
+
* cost a conversation turn. Failures are logged and either acked as a terminal
|
|
2832
|
+
* outcome or left pending for the next drain (see `runner-file-sync.ts`).
|
|
2833
|
+
*
|
|
2834
|
+
* Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
|
|
2835
|
+
*
|
|
2836
|
+
* @returns the number of files written to disk.
|
|
2837
|
+
*/
|
|
2838
|
+
async syncPendingFiles() {
|
|
2839
|
+
if (this.stopped) return 0;
|
|
2840
|
+
if (this.syncingFiles) return 0;
|
|
2841
|
+
this.syncingFiles = true;
|
|
2842
|
+
try {
|
|
2843
|
+
const applied = await syncPendingRunnerFiles({
|
|
2844
|
+
agentId: this.agentId,
|
|
2845
|
+
apiUrl: this.apiUrl,
|
|
2846
|
+
getAuthHeader: this.getAuthHeader,
|
|
2847
|
+
fetchImpl: this.fetchImpl,
|
|
2848
|
+
allowedDirectories: this.fileSyncDirectories,
|
|
2849
|
+
homeDir: this.homeDir,
|
|
2850
|
+
ackFailures: this.fileAckFailures,
|
|
2851
|
+
log: this.log
|
|
2852
|
+
});
|
|
2853
|
+
this.appliedFileCount += applied;
|
|
2854
|
+
return applied;
|
|
2855
|
+
} catch (err) {
|
|
2856
|
+
this.log({
|
|
2857
|
+
level: "error",
|
|
2858
|
+
message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
|
|
2859
|
+
});
|
|
2860
|
+
return 0;
|
|
2861
|
+
} finally {
|
|
2862
|
+
this.syncingFiles = false;
|
|
2863
|
+
}
|
|
2864
|
+
}
|
|
2282
2865
|
async runDrain() {
|
|
2283
2866
|
let dispatched = 0;
|
|
2284
2867
|
try {
|
|
@@ -2312,6 +2895,28 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2312
2895
|
}
|
|
2313
2896
|
return false;
|
|
2314
2897
|
}
|
|
2898
|
+
/**
|
|
2899
|
+
* File-pull work, for `run.ts`'s idle accounting (#559).
|
|
2900
|
+
*
|
|
2901
|
+
* Pulling a file is real work that `drainPending()` knows nothing about, so
|
|
2902
|
+
* without this a near-idle runner counts a credential pull as an empty tick
|
|
2903
|
+
* and `--idle-timeout` can `process.exit` mid-pull — leaving a
|
|
2904
|
+
* `.evident-push-*.tmp` behind — or immediately after the write, before the
|
|
2905
|
+
* browser has run the authorize/callback that activates it (the user then sees
|
|
2906
|
+
* `saved_not_activated` for a runner that was fine).
|
|
2907
|
+
*
|
|
2908
|
+
* Two signals because one cannot cover both cases: `inFlight` is the pull
|
|
2909
|
+
* happening RIGHT NOW (it may outlive the tick that started it), and
|
|
2910
|
+
* `appliedFiles` is monotonic so a pull that started AND finished between two
|
|
2911
|
+
* idle checks still shows up as an advance.
|
|
2912
|
+
*
|
|
2913
|
+
* CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
|
|
2914
|
+
* the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
|
|
2915
|
+
* samples afterwards reads `true` every single cycle and can never idle out.
|
|
2916
|
+
*/
|
|
2917
|
+
fileSyncActivity() {
|
|
2918
|
+
return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
|
|
2919
|
+
}
|
|
2315
2920
|
/**
|
|
2316
2921
|
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2317
2922
|
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
@@ -2373,7 +2978,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2373
2978
|
await this.sleep(step);
|
|
2374
2979
|
}
|
|
2375
2980
|
}
|
|
2376
|
-
while (this.hasInFlightWatchers()) {
|
|
2981
|
+
while (this.hasInFlightWatchers() || this.syncingFiles) {
|
|
2377
2982
|
if (this.now() >= deadline) return false;
|
|
2378
2983
|
await this.sleep(step);
|
|
2379
2984
|
}
|
|
@@ -2408,10 +3013,15 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2408
3013
|
* @returns the count of messages NEWLY dispatched (not already in-flight).
|
|
2409
3014
|
*/
|
|
2410
3015
|
async processConversation(conv) {
|
|
2411
|
-
const sessionId = await this.ensureSession(conv);
|
|
3016
|
+
const { sessionId, refusedSessionId } = await this.ensureSession(conv);
|
|
2412
3017
|
const messages = await this.getPendingMessages(conv.id);
|
|
2413
3018
|
let dispatched = 0;
|
|
2414
3019
|
let skippedAlreadyDispatched = 0;
|
|
3020
|
+
if (refusedSessionId && messages.length > 0) {
|
|
3021
|
+
void this.postSignal(conv.id, messages[0].id, "session_superseded", {
|
|
3022
|
+
superseded_session_id: refusedSessionId
|
|
3023
|
+
});
|
|
3024
|
+
}
|
|
2415
3025
|
for (const message of messages) {
|
|
2416
3026
|
if (this.stopped) break;
|
|
2417
3027
|
if (this.dispatched.has(message.id)) {
|
|
@@ -2460,6 +3070,13 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2460
3070
|
}
|
|
2461
3071
|
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
2462
3072
|
this.sessions.delete(conv.id);
|
|
3073
|
+
this.supersede(conv.id, sessionId);
|
|
3074
|
+
this.log({
|
|
3075
|
+
level: "warn",
|
|
3076
|
+
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3077
|
+
conversation_id: conv.id,
|
|
3078
|
+
message_id: message.id
|
|
3079
|
+
});
|
|
2463
3080
|
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
2464
3081
|
this.log({
|
|
2465
3082
|
level: "warn",
|
|
@@ -2500,8 +3117,42 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2500
3117
|
this.ensureWatcherRunning(sessionId);
|
|
2501
3118
|
return dispatched;
|
|
2502
3119
|
}
|
|
3120
|
+
/**
|
|
3121
|
+
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
3122
|
+
* (#553). Keyed by conversation and hard-capped, so it cannot grow with the
|
|
3123
|
+
* number of failures — see the `supersededSessions` field doc.
|
|
3124
|
+
*/
|
|
3125
|
+
supersede(conversationId, sessionId) {
|
|
3126
|
+
this.supersededSessions.delete(conversationId);
|
|
3127
|
+
this.supersededSessions.set(conversationId, sessionId);
|
|
3128
|
+
while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
|
|
3129
|
+
const oldest = this.supersededSessions.keys().next().value;
|
|
3130
|
+
if (oldest === void 0) return;
|
|
3131
|
+
this.supersededSessions.delete(oldest);
|
|
3132
|
+
}
|
|
3133
|
+
}
|
|
3134
|
+
/** Whether `sessionId` is the session this conversation has abandoned (#553). */
|
|
3135
|
+
isSuperseded(conversationId, sessionId) {
|
|
3136
|
+
return this.supersededSessions.get(conversationId) === sessionId;
|
|
3137
|
+
}
|
|
3138
|
+
/**
|
|
3139
|
+
* Resolve the opencode session to run this conversation's turns in.
|
|
3140
|
+
*
|
|
3141
|
+
* `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
|
|
3142
|
+
* binding was an id this runner had abandoned, so a resurrection genuinely
|
|
3143
|
+
* happened and a fresh session was bound instead. The caller reports it.
|
|
3144
|
+
*/
|
|
2503
3145
|
async ensureSession(conv) {
|
|
2504
3146
|
const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
|
|
3147
|
+
if (bound && this.isSuperseded(conv.id, bound)) {
|
|
3148
|
+
this.log({
|
|
3149
|
+
level: "warn",
|
|
3150
|
+
message: `OpenCode session ${bound.slice(0, 8)} was abandoned for conversation ${conv.id.slice(0, 8)} after a failed dispatch but is still bound to it (the persisted id was written back by a turn already in flight) \u2014 ignoring it and binding a fresh session.`,
|
|
3151
|
+
conversation_id: conv.id
|
|
3152
|
+
});
|
|
3153
|
+
this.sessions.delete(conv.id);
|
|
3154
|
+
return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
|
|
3155
|
+
}
|
|
2505
3156
|
if (bound) {
|
|
2506
3157
|
const exists = await sessionExists(this.port, bound);
|
|
2507
3158
|
if (exists === false) {
|
|
@@ -2511,12 +3162,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2511
3162
|
conversation_id: conv.id
|
|
2512
3163
|
});
|
|
2513
3164
|
this.sessions.delete(conv.id);
|
|
2514
|
-
return this.createAndBindSession(conv.id);
|
|
3165
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2515
3166
|
}
|
|
2516
3167
|
this.sessions.set(conv.id, bound);
|
|
2517
|
-
return bound;
|
|
3168
|
+
return { sessionId: bound };
|
|
2518
3169
|
}
|
|
2519
|
-
return this.createAndBindSession(conv.id);
|
|
3170
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2520
3171
|
}
|
|
2521
3172
|
/**
|
|
2522
3173
|
* Create a fresh OpenCode session for a conversation, cache the binding, and
|
|
@@ -2719,12 +3370,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2719
3370
|
stuckReported: false,
|
|
2720
3371
|
lastAliveAt: 0,
|
|
2721
3372
|
aliveInFlight: false,
|
|
3373
|
+
titleSynced: false,
|
|
3374
|
+
titleSyncInFlight: false,
|
|
2722
3375
|
awaitingHumanLatched: false,
|
|
2723
3376
|
pausedOnQuestion: false,
|
|
2724
3377
|
pausedOnPermission: false,
|
|
2725
3378
|
pausedClearConfirmed: false,
|
|
2726
3379
|
pausedInFlight: false,
|
|
2727
|
-
deliveryDeadlineAnchored: false
|
|
3380
|
+
deliveryDeadlineAnchored: false,
|
|
3381
|
+
b2PinnedSinceMs: 0,
|
|
3382
|
+
b2LastDescendantCheckMs: 0,
|
|
3383
|
+
b2AbandonedSignalled: false
|
|
2728
3384
|
});
|
|
2729
3385
|
}
|
|
2730
3386
|
/**
|
|
@@ -2792,12 +3448,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2792
3448
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2793
3449
|
lastAliveAt: 0,
|
|
2794
3450
|
aliveInFlight: false,
|
|
3451
|
+
titleSynced: false,
|
|
3452
|
+
titleSyncInFlight: false,
|
|
2795
3453
|
awaitingHumanLatched: false,
|
|
2796
3454
|
pausedOnQuestion: false,
|
|
2797
3455
|
pausedOnPermission: false,
|
|
2798
3456
|
pausedClearConfirmed: false,
|
|
2799
3457
|
pausedInFlight: false,
|
|
2800
|
-
deliveryDeadlineAnchored: false
|
|
3458
|
+
deliveryDeadlineAnchored: false,
|
|
3459
|
+
b2PinnedSinceMs: 0,
|
|
3460
|
+
b2LastDescendantCheckMs: 0,
|
|
3461
|
+
b2AbandonedSignalled: false
|
|
2801
3462
|
});
|
|
2802
3463
|
}
|
|
2803
3464
|
/**
|
|
@@ -2959,58 +3620,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2959
3620
|
}
|
|
2960
3621
|
}
|
|
2961
3622
|
if (state === "done") {
|
|
2962
|
-
this.
|
|
2963
|
-
if (!inFlight.done) {
|
|
2964
|
-
this.log({
|
|
2965
|
-
level: "info",
|
|
2966
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
2967
|
-
conversation_id: conv.id,
|
|
2968
|
-
message_id: inFlight.evidentMessageId
|
|
2969
|
-
});
|
|
2970
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
2971
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
2972
|
-
try {
|
|
2973
|
-
await this.markDone(
|
|
2974
|
-
conv.id,
|
|
2975
|
-
inFlight.evidentMessageId,
|
|
2976
|
-
sessionId,
|
|
2977
|
-
inFlight.opencodeMessageId,
|
|
2978
|
-
title,
|
|
2979
|
-
usage
|
|
2980
|
-
);
|
|
2981
|
-
} catch (err) {
|
|
2982
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2983
|
-
if (err instanceof ChannelTerminalError) {
|
|
2984
|
-
this.log({
|
|
2985
|
-
level: "warn",
|
|
2986
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2987
|
-
conversation_id: conv.id,
|
|
2988
|
-
message_id: inFlight.evidentMessageId
|
|
2989
|
-
});
|
|
2990
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2991
|
-
return;
|
|
2992
|
-
}
|
|
2993
|
-
if (this.now() >= inFlight.deadline) {
|
|
2994
|
-
this.log({
|
|
2995
|
-
level: "warn",
|
|
2996
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
2997
|
-
conversation_id: conv.id,
|
|
2998
|
-
message_id: inFlight.evidentMessageId
|
|
2999
|
-
});
|
|
3000
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3001
|
-
return;
|
|
3002
|
-
}
|
|
3003
|
-
this.log({
|
|
3004
|
-
level: "warn",
|
|
3005
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
3006
|
-
conversation_id: conv.id,
|
|
3007
|
-
message_id: inFlight.evidentMessageId
|
|
3008
|
-
});
|
|
3009
|
-
return;
|
|
3010
|
-
}
|
|
3011
|
-
inFlight.done = true;
|
|
3012
|
-
}
|
|
3013
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3623
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3014
3624
|
return;
|
|
3015
3625
|
}
|
|
3016
3626
|
if (state === "failed") {
|
|
@@ -3024,8 +3634,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3024
3634
|
message_id: inFlight.evidentMessageId
|
|
3025
3635
|
});
|
|
3026
3636
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
3637
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3027
3638
|
try {
|
|
3028
|
-
await this.markFailed(
|
|
3639
|
+
await this.markFailed(
|
|
3640
|
+
conv.id,
|
|
3641
|
+
inFlight.evidentMessageId,
|
|
3642
|
+
sessionId,
|
|
3643
|
+
error2,
|
|
3644
|
+
usage,
|
|
3645
|
+
failure
|
|
3646
|
+
);
|
|
3029
3647
|
} catch (err) {
|
|
3030
3648
|
if (err instanceof ChannelAuthError) throw err;
|
|
3031
3649
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3070,6 +3688,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3070
3688
|
});
|
|
3071
3689
|
}
|
|
3072
3690
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
3691
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
3692
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
3693
|
+
if (!pinnedNow) {
|
|
3694
|
+
if (snapshotReadable) {
|
|
3695
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
3696
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
3697
|
+
inFlight.b2AbandonedSignalled = false;
|
|
3698
|
+
}
|
|
3699
|
+
} else {
|
|
3700
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
3701
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3702
|
+
return;
|
|
3703
|
+
}
|
|
3704
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
3705
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
3706
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
3707
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
3708
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
3709
|
+
if (isB2AbandonmentConfirmed({
|
|
3710
|
+
pinnedForMs,
|
|
3711
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
3712
|
+
descendantOngoing
|
|
3713
|
+
})) {
|
|
3714
|
+
inFlight.b2AbandonedSignalled = true;
|
|
3715
|
+
this.log({
|
|
3716
|
+
level: "warn",
|
|
3717
|
+
message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
|
|
3718
|
+
conversation_id: conv.id,
|
|
3719
|
+
message_id: id
|
|
3720
|
+
});
|
|
3721
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
3722
|
+
watched_for_ms: pinnedForMs
|
|
3723
|
+
});
|
|
3724
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3725
|
+
return;
|
|
3726
|
+
}
|
|
3727
|
+
}
|
|
3728
|
+
}
|
|
3073
3729
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3074
3730
|
this.log({
|
|
3075
3731
|
level: "warn",
|
|
@@ -3089,6 +3745,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3089
3745
|
inFlight.aliveInFlight = false;
|
|
3090
3746
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3091
3747
|
});
|
|
3748
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
3749
|
+
inFlight.titleSyncInFlight = true;
|
|
3750
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
3751
|
+
if (!title) {
|
|
3752
|
+
inFlight.titleSyncInFlight = false;
|
|
3753
|
+
return;
|
|
3754
|
+
}
|
|
3755
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
3756
|
+
inFlight.titleSyncInFlight = false;
|
|
3757
|
+
if (ok) inFlight.titleSynced = true;
|
|
3758
|
+
});
|
|
3759
|
+
}
|
|
3092
3760
|
}
|
|
3093
3761
|
if (awaitingHuman) {
|
|
3094
3762
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3126,6 +3794,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3126
3794
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3127
3795
|
}
|
|
3128
3796
|
}
|
|
3797
|
+
/**
|
|
3798
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
3799
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
3800
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
3801
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
3802
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
3803
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
3804
|
+
* and risking the two copies silently drifting apart.
|
|
3805
|
+
*/
|
|
3806
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
3807
|
+
const conv = watcher.conv;
|
|
3808
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
3809
|
+
if (!inFlight.done) {
|
|
3810
|
+
this.log({
|
|
3811
|
+
level: "info",
|
|
3812
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
3813
|
+
conversation_id: conv.id,
|
|
3814
|
+
message_id: inFlight.evidentMessageId
|
|
3815
|
+
});
|
|
3816
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
3817
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
3818
|
+
try {
|
|
3819
|
+
await this.markDone(
|
|
3820
|
+
conv.id,
|
|
3821
|
+
inFlight.evidentMessageId,
|
|
3822
|
+
sessionId,
|
|
3823
|
+
inFlight.opencodeMessageId,
|
|
3824
|
+
title,
|
|
3825
|
+
usage
|
|
3826
|
+
);
|
|
3827
|
+
} catch (err) {
|
|
3828
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3829
|
+
if (err instanceof ChannelTerminalError) {
|
|
3830
|
+
this.log({
|
|
3831
|
+
level: "warn",
|
|
3832
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
3833
|
+
conversation_id: conv.id,
|
|
3834
|
+
message_id: inFlight.evidentMessageId
|
|
3835
|
+
});
|
|
3836
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3837
|
+
return;
|
|
3838
|
+
}
|
|
3839
|
+
if (this.now() >= inFlight.deadline) {
|
|
3840
|
+
this.log({
|
|
3841
|
+
level: "warn",
|
|
3842
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
3843
|
+
conversation_id: conv.id,
|
|
3844
|
+
message_id: inFlight.evidentMessageId
|
|
3845
|
+
});
|
|
3846
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3847
|
+
return;
|
|
3848
|
+
}
|
|
3849
|
+
this.log({
|
|
3850
|
+
level: "warn",
|
|
3851
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
3852
|
+
conversation_id: conv.id,
|
|
3853
|
+
message_id: inFlight.evidentMessageId
|
|
3854
|
+
});
|
|
3855
|
+
return;
|
|
3856
|
+
}
|
|
3857
|
+
inFlight.done = true;
|
|
3858
|
+
}
|
|
3859
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3860
|
+
}
|
|
3129
3861
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3130
3862
|
/**
|
|
3131
3863
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3292,6 +4024,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3292
4024
|
if (state === "failed") {
|
|
3293
4025
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3294
4026
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4027
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3295
4028
|
this.log({
|
|
3296
4029
|
level: "error",
|
|
3297
4030
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3299,7 +4032,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3299
4032
|
message_id: row.id
|
|
3300
4033
|
});
|
|
3301
4034
|
try {
|
|
3302
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4035
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3303
4036
|
} catch (err) {
|
|
3304
4037
|
if (err instanceof ChannelAuthError) throw err;
|
|
3305
4038
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3705,6 +4438,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3705
4438
|
}
|
|
3706
4439
|
return false;
|
|
3707
4440
|
}
|
|
4441
|
+
/**
|
|
4442
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
4443
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
4444
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
4445
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
4446
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
4447
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
4448
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
4449
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
4450
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
4451
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
4452
|
+
* not ongoing".
|
|
4453
|
+
*
|
|
4454
|
+
* Return contract:
|
|
4455
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
4456
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
4457
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
4458
|
+
* CONFIRMED NOT a descendant of it.
|
|
4459
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
4460
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
4461
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
4462
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
4463
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
4464
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
4465
|
+
* here.
|
|
4466
|
+
*
|
|
4467
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
4468
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
4469
|
+
* by interaction attribution or the recovery path.
|
|
4470
|
+
*/
|
|
4471
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
4472
|
+
let current = sessionId;
|
|
4473
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
4474
|
+
if (current === rootSessionId) return true;
|
|
4475
|
+
const parent = await this.resolveSessionParent(current);
|
|
4476
|
+
if (parent === void 0) return null;
|
|
4477
|
+
if (parent === null) return false;
|
|
4478
|
+
current = parent;
|
|
4479
|
+
}
|
|
4480
|
+
return null;
|
|
4481
|
+
}
|
|
3708
4482
|
/**
|
|
3709
4483
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3710
4484
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -3789,6 +4563,54 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3789
4563
|
}
|
|
3790
4564
|
return null;
|
|
3791
4565
|
}
|
|
4566
|
+
/**
|
|
4567
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
4568
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
4569
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
4570
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
4571
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
4572
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
4573
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
4574
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
4575
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
4576
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
4577
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
4578
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
4579
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
4580
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
4581
|
+
*
|
|
4582
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
4583
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
4584
|
+
* caller only latches `titleSynced` on `true`).
|
|
4585
|
+
*/
|
|
4586
|
+
async patchConversationTitle(conversationId, title) {
|
|
4587
|
+
try {
|
|
4588
|
+
const res = await this.fetchImpl(
|
|
4589
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
4590
|
+
{
|
|
4591
|
+
method: "PATCH",
|
|
4592
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
4593
|
+
body: JSON.stringify({ title })
|
|
4594
|
+
}
|
|
4595
|
+
);
|
|
4596
|
+
if (!res.ok) {
|
|
4597
|
+
this.log({
|
|
4598
|
+
level: "debug",
|
|
4599
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
4600
|
+
conversation_id: conversationId
|
|
4601
|
+
});
|
|
4602
|
+
return false;
|
|
4603
|
+
}
|
|
4604
|
+
return true;
|
|
4605
|
+
} catch (err) {
|
|
4606
|
+
this.log({
|
|
4607
|
+
level: "debug",
|
|
4608
|
+
message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
|
|
4609
|
+
conversation_id: conversationId
|
|
4610
|
+
});
|
|
4611
|
+
return false;
|
|
4612
|
+
}
|
|
4613
|
+
}
|
|
3792
4614
|
/**
|
|
3793
4615
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3794
4616
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -3847,6 +4669,84 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3847
4669
|
}
|
|
3848
4670
|
return false;
|
|
3849
4671
|
}
|
|
4672
|
+
/**
|
|
4673
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
4674
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
4675
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
4676
|
+
*
|
|
4677
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
4678
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
4679
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
4680
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
4681
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
4682
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
4683
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
4684
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
4685
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
4686
|
+
* executing, between its step's completion and the next generation step"
|
|
4687
|
+
* transcript gap that a transcript-based check would need a second,
|
|
4688
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
4689
|
+
* message timestamps at all.
|
|
4690
|
+
*
|
|
4691
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
4692
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
4693
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
4694
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
4695
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
4696
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
4697
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
4698
|
+
* delegation — which the root's status is not.
|
|
4699
|
+
*
|
|
4700
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
4701
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
4702
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
4703
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
4704
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
4705
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
4706
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
4707
|
+
* instead.
|
|
4708
|
+
*
|
|
4709
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
4710
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
4711
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
4712
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
4713
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
4714
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
4715
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
4716
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
4717
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
4718
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
4719
|
+
* or pathological chain partway through the parent walk), OR at
|
|
4720
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
4721
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
4722
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
4723
|
+
* (unlike the recovery cross-check's contract) — see
|
|
4724
|
+
* `isB2AbandonmentConfirmed`.
|
|
4725
|
+
*/
|
|
4726
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
4727
|
+
const sessions = await listSessions(this.port);
|
|
4728
|
+
if (!sessions) {
|
|
4729
|
+
this.log({
|
|
4730
|
+
level: "warn",
|
|
4731
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
4732
|
+
});
|
|
4733
|
+
return null;
|
|
4734
|
+
}
|
|
4735
|
+
let indeterminate = false;
|
|
4736
|
+
for (const candidate of sessions) {
|
|
4737
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
4738
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
4739
|
+
if (membership === null) {
|
|
4740
|
+
indeterminate = true;
|
|
4741
|
+
continue;
|
|
4742
|
+
}
|
|
4743
|
+
if (membership === false) continue;
|
|
4744
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
4745
|
+
if (ongoing === true) return true;
|
|
4746
|
+
if (ongoing === null) indeterminate = true;
|
|
4747
|
+
}
|
|
4748
|
+
return indeterminate ? null : false;
|
|
4749
|
+
}
|
|
3850
4750
|
/**
|
|
3851
4751
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
3852
4752
|
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
@@ -3975,6 +4875,32 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3975
4875
|
}
|
|
3976
4876
|
return messages;
|
|
3977
4877
|
}
|
|
4878
|
+
/**
|
|
4879
|
+
* The `opencode_session_id` fragment of a status PATCH body — `{}` when this
|
|
4880
|
+
* conversation has ABANDONED that session (#553). The field is optional
|
|
4881
|
+
* server-side and an absent one leaves the persisted binding untouched, so
|
|
4882
|
+
* omitting it is how a routine status write stops resurrecting it.
|
|
4883
|
+
*
|
|
4884
|
+
* ONLY for writes whose sole cost is a lost deep link. The `processing` notice
|
|
4885
|
+
* degrades to no "View in Evident" link (the reaction swap still fires) and the
|
|
4886
|
+
* turn-failure notice is built from the PATCH's own `error` text with a link off
|
|
4887
|
+
* the persisted row — neither loses content the user came for. `markDone`
|
|
4888
|
+
* deliberately does NOT use this helper: the server fetches the reply text
|
|
4889
|
+
* THROUGH the session id it is given, so suppressing there would replace the
|
|
4890
|
+
* agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
|
|
4891
|
+
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4892
|
+
* stick.
|
|
4893
|
+
*/
|
|
4894
|
+
sessionIdBody(sessionId, conversationId, messageId, status) {
|
|
4895
|
+
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4896
|
+
this.log({
|
|
4897
|
+
level: "debug",
|
|
4898
|
+
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
|
|
4899
|
+
conversation_id: conversationId,
|
|
4900
|
+
message_id: messageId
|
|
4901
|
+
});
|
|
4902
|
+
return {};
|
|
4903
|
+
}
|
|
3978
4904
|
/**
|
|
3979
4905
|
* EXISTING combinedAuth route — now fired by the watcher on queued→running
|
|
3980
4906
|
* (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
|
|
@@ -4004,7 +4930,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4004
4930
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
4005
4931
|
body: JSON.stringify({
|
|
4006
4932
|
status: "processing",
|
|
4007
|
-
|
|
4933
|
+
...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
|
|
4008
4934
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
4009
4935
|
...title ? { title } : {}
|
|
4010
4936
|
})
|
|
@@ -4053,6 +4979,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4053
4979
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
4054
4980
|
body: JSON.stringify({
|
|
4055
4981
|
status: "done",
|
|
4982
|
+
// ALWAYS sent, even for a session this conversation has abandoned
|
|
4983
|
+
// (#553): the server reads the reply text back out of THIS session id
|
|
4984
|
+
// to deliver it. Omitting it would leave the user with "✅ Done!"
|
|
4985
|
+
// instead of the answer — a worse regression than the resurrection it
|
|
4986
|
+
// would prevent, which `ensureSession`'s guard handles anyway.
|
|
4056
4987
|
opencode_session_id: sessionId,
|
|
4057
4988
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
4058
4989
|
...title ? { title } : {},
|
|
@@ -4079,11 +5010,21 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4079
5010
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4080
5011
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4081
5012
|
*/
|
|
4082
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5013
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4083
5014
|
const body = { status: "failed" };
|
|
4084
|
-
if (sessionId
|
|
5015
|
+
if (sessionId === null) {
|
|
5016
|
+
body.opencode_session_id = null;
|
|
5017
|
+
} else if (sessionId !== void 0) {
|
|
5018
|
+
Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
|
|
5019
|
+
}
|
|
4085
5020
|
if (error2 !== void 0) body.error = error2;
|
|
4086
5021
|
if (usage) Object.assign(body, usage);
|
|
5022
|
+
if (failure) {
|
|
5023
|
+
body.failure_kind = failure.kind;
|
|
5024
|
+
body.failure_provider_id = failure.providerId;
|
|
5025
|
+
body.failure_model_id = failure.modelId;
|
|
5026
|
+
body.failure_reason = failure.reason;
|
|
5027
|
+
}
|
|
4087
5028
|
await this.callWithRetry(
|
|
4088
5029
|
"marking message as failed",
|
|
4089
5030
|
() => this.fetchImpl(
|
|
@@ -4096,6 +5037,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4096
5037
|
)
|
|
4097
5038
|
);
|
|
4098
5039
|
}
|
|
5040
|
+
/**
|
|
5041
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5042
|
+
*
|
|
5043
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5044
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
5045
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
5046
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
5047
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
5048
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
5049
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
5050
|
+
*/
|
|
5051
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
5052
|
+
const classified = messageFailure(messages, userMessageId);
|
|
5053
|
+
if (classified != null) return classified;
|
|
5054
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
5055
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
5056
|
+
return applyZeroProviderFallback(
|
|
5057
|
+
classified,
|
|
5058
|
+
hasProvider,
|
|
5059
|
+
reply?.info?.providerID ?? null,
|
|
5060
|
+
reply?.info?.modelID ?? null
|
|
5061
|
+
);
|
|
5062
|
+
}
|
|
4099
5063
|
/**
|
|
4100
5064
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4101
5065
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4272,7 +5236,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4272
5236
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4273
5237
|
console.log(
|
|
4274
5238
|
chalk5.dim(
|
|
4275
|
-
` ${getCliName()} run --
|
|
5239
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4276
5240
|
)
|
|
4277
5241
|
);
|
|
4278
5242
|
}
|
|
@@ -4402,19 +5366,21 @@ async function resolveAgentIdFromKey(authHeader) {
|
|
|
4402
5366
|
return { agent_id: data.agent_id };
|
|
4403
5367
|
}
|
|
4404
5368
|
return {
|
|
4405
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --
|
|
5369
|
+
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
|
|
4406
5370
|
};
|
|
4407
5371
|
} catch (error2) {
|
|
4408
5372
|
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4409
5373
|
return { error: `Failed to resolve runner from key: ${message}` };
|
|
4410
5374
|
}
|
|
4411
5375
|
}
|
|
5376
|
+
var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
|
|
4412
5377
|
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
4413
5378
|
const apiUrl = getApiUrlConfig();
|
|
4414
5379
|
try {
|
|
4415
5380
|
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
4416
5381
|
method: "POST",
|
|
4417
|
-
headers: { Authorization: authHeader }
|
|
5382
|
+
headers: { Authorization: authHeader },
|
|
5383
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
4418
5384
|
});
|
|
4419
5385
|
if (!response.ok) {
|
|
4420
5386
|
const serverMessage = await readErrorMessage(response);
|
|
@@ -4425,7 +5391,35 @@ async function notifyAgentDisconnected(agentId, authHeader) {
|
|
|
4425
5391
|
}
|
|
4426
5392
|
return { ok: true };
|
|
4427
5393
|
} catch (error2) {
|
|
4428
|
-
return { ok: false, error:
|
|
5394
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
5395
|
+
}
|
|
5396
|
+
}
|
|
5397
|
+
function describeBestEffortError(error2) {
|
|
5398
|
+
const name = error2?.name;
|
|
5399
|
+
if (name === "TimeoutError" || name === "AbortError") {
|
|
5400
|
+
return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
|
|
5401
|
+
}
|
|
5402
|
+
return error2 instanceof Error ? error2.message : String(error2);
|
|
5403
|
+
}
|
|
5404
|
+
async function reportMicrovmId(agentId, authHeader, microvmId) {
|
|
5405
|
+
try {
|
|
5406
|
+
const apiUrl = getApiUrlConfig();
|
|
5407
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
|
|
5408
|
+
method: "POST",
|
|
5409
|
+
headers: { Authorization: authHeader, "Content-Type": "application/json" },
|
|
5410
|
+
body: JSON.stringify({ microvm_id: microvmId }),
|
|
5411
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
5412
|
+
});
|
|
5413
|
+
if (!response.ok) {
|
|
5414
|
+
const serverMessage = await readErrorMessage(response);
|
|
5415
|
+
return {
|
|
5416
|
+
ok: false,
|
|
5417
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5418
|
+
};
|
|
5419
|
+
}
|
|
5420
|
+
return { ok: true };
|
|
5421
|
+
} catch (error2) {
|
|
5422
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
4429
5423
|
}
|
|
4430
5424
|
}
|
|
4431
5425
|
async function getAgentInfo(agentId, authHeader) {
|
|
@@ -4475,6 +5469,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
4475
5469
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
4476
5470
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4477
5471
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
5472
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
4478
5473
|
function resolveLogLevel(options) {
|
|
4479
5474
|
const accepted = Object.keys(LOG_LEVELS);
|
|
4480
5475
|
const validate = (value, source) => {
|
|
@@ -4498,6 +5493,34 @@ function resolveLogLevel(options) {
|
|
|
4498
5493
|
}
|
|
4499
5494
|
return "info";
|
|
4500
5495
|
}
|
|
5496
|
+
function resolveFileSyncDirectories(raw, homeDir) {
|
|
5497
|
+
const directories = [];
|
|
5498
|
+
for (const entry of raw ?? []) {
|
|
5499
|
+
const trimmed = entry.trim();
|
|
5500
|
+
if (trimmed === "") {
|
|
5501
|
+
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
5502
|
+
}
|
|
5503
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join2(homeDir, trimmed.slice(2)) : trimmed;
|
|
5504
|
+
if (!isAbsolute2(expanded)) {
|
|
5505
|
+
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
5506
|
+
}
|
|
5507
|
+
const normalized = resolvePath(expanded);
|
|
5508
|
+
if (parse(normalized).root === normalized) {
|
|
5509
|
+
throw new Error(
|
|
5510
|
+
`--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
|
|
5511
|
+
);
|
|
5512
|
+
}
|
|
5513
|
+
if (!directories.includes(normalized)) {
|
|
5514
|
+
directories.push(normalized);
|
|
5515
|
+
}
|
|
5516
|
+
}
|
|
5517
|
+
if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
|
|
5518
|
+
throw new Error(
|
|
5519
|
+
`--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
|
|
5520
|
+
);
|
|
5521
|
+
}
|
|
5522
|
+
return directories;
|
|
5523
|
+
}
|
|
4501
5524
|
function meetsThreshold(state, level) {
|
|
4502
5525
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
4503
5526
|
}
|
|
@@ -4619,18 +5642,29 @@ async function handleAuthError(state, error2) {
|
|
|
4619
5642
|
async function driveChannels(state, driver) {
|
|
4620
5643
|
let idlePolls = 0;
|
|
4621
5644
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5645
|
+
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
4622
5646
|
while (state.running) {
|
|
4623
5647
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
4624
5648
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
4625
5649
|
if (state.interactive) displayStatus(state);
|
|
4626
5650
|
await state.connection.reconnectPromise;
|
|
4627
5651
|
}
|
|
5652
|
+
const carriedOverFileSync = driver.fileSyncActivity().inFlight;
|
|
5653
|
+
void driver.syncPendingFiles().catch(
|
|
5654
|
+
(error2) => logActivity(state, {
|
|
5655
|
+
type: "error",
|
|
5656
|
+
error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
5657
|
+
})
|
|
5658
|
+
);
|
|
4628
5659
|
try {
|
|
4629
5660
|
const processed = await driver.drainPending();
|
|
4630
5661
|
state.messageCount += processed;
|
|
4631
5662
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4632
5663
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4633
|
-
|
|
5664
|
+
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5665
|
+
const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
|
|
5666
|
+
lastSeenAppliedFiles = appliedFiles;
|
|
5667
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
4634
5668
|
idlePolls = 0;
|
|
4635
5669
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
4636
5670
|
} else if (state.idleTimeout !== null) {
|
|
@@ -4659,7 +5693,7 @@ async function driveChannels(state, driver) {
|
|
|
4659
5693
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
4660
5694
|
if (state.interactive) displayStatus(state);
|
|
4661
5695
|
}
|
|
4662
|
-
await new Promise((
|
|
5696
|
+
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
4663
5697
|
if (state.idleTimeout !== null && idlePolls >= 2) {
|
|
4664
5698
|
const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
|
|
4665
5699
|
if (idleMs > state.idleTimeout * 1e3) {
|
|
@@ -4762,7 +5796,18 @@ async function notifyOffline(state) {
|
|
|
4762
5796
|
if (state.interactive) displayStatus(state);
|
|
4763
5797
|
}
|
|
4764
5798
|
}
|
|
5799
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
5800
|
+
const startedAt = Date.now();
|
|
5801
|
+
try {
|
|
5802
|
+
return await run2();
|
|
5803
|
+
} finally {
|
|
5804
|
+
const elapsedMs = Date.now() - startedAt;
|
|
5805
|
+
durations[name] = elapsedMs;
|
|
5806
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
5807
|
+
}
|
|
5808
|
+
}
|
|
4765
5809
|
async function cleanup(state, opts = {}) {
|
|
5810
|
+
const durations = {};
|
|
4766
5811
|
state.running = false;
|
|
4767
5812
|
for (const timer of state.sessionCleanupTimers) {
|
|
4768
5813
|
clearInterval(timer);
|
|
@@ -4776,7 +5821,13 @@ async function cleanup(state, opts = {}) {
|
|
|
4776
5821
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4777
5822
|
displayStatus(state);
|
|
4778
5823
|
}
|
|
4779
|
-
const
|
|
5824
|
+
const driver = state.channelDriver;
|
|
5825
|
+
const settled = await timeShutdownPhase(
|
|
5826
|
+
state,
|
|
5827
|
+
durations,
|
|
5828
|
+
"drain",
|
|
5829
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
5830
|
+
);
|
|
4780
5831
|
if (!settled) {
|
|
4781
5832
|
logActivity(state, {
|
|
4782
5833
|
type: "info",
|
|
@@ -4785,13 +5836,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4785
5836
|
if (state.interactive) displayStatus(state);
|
|
4786
5837
|
}
|
|
4787
5838
|
}
|
|
4788
|
-
await notifyOffline(state);
|
|
5839
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
4789
5840
|
if (state.connection) {
|
|
4790
|
-
state.connection
|
|
5841
|
+
const connection = state.connection;
|
|
5842
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
4791
5843
|
state.connection = null;
|
|
4792
5844
|
}
|
|
4793
5845
|
if (state.opencodeProcess) {
|
|
4794
|
-
|
|
5846
|
+
const opencodeProcess = state.opencodeProcess;
|
|
5847
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
4795
5848
|
if (state.interactive) {
|
|
4796
5849
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
4797
5850
|
displayStatus(state);
|
|
@@ -4800,12 +5853,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4800
5853
|
}
|
|
4801
5854
|
state.opencodeProcess = null;
|
|
4802
5855
|
}
|
|
5856
|
+
return durations;
|
|
4803
5857
|
}
|
|
4804
5858
|
async function run(options) {
|
|
4805
5859
|
const interactive = isInteractive(options.json);
|
|
4806
5860
|
let logLevel;
|
|
5861
|
+
let fileSyncDirectories;
|
|
4807
5862
|
try {
|
|
4808
5863
|
logLevel = resolveLogLevel(options);
|
|
5864
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir2());
|
|
4809
5865
|
} catch (error2) {
|
|
4810
5866
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4811
5867
|
if (options.json) {
|
|
@@ -4840,6 +5896,11 @@ async function run(options) {
|
|
|
4840
5896
|
sessionCleanupTimers: [],
|
|
4841
5897
|
authHeader: ""
|
|
4842
5898
|
};
|
|
5899
|
+
if (fileSyncDirectories.length > 0) {
|
|
5900
|
+
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
5901
|
+
} else {
|
|
5902
|
+
log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
|
|
5903
|
+
}
|
|
4843
5904
|
if (!options.runner && options.agent) {
|
|
4844
5905
|
telemetry.info(
|
|
4845
5906
|
EventTypes.DEPRECATED_AGENT_FLAG_USED,
|
|
@@ -4863,14 +5924,38 @@ async function run(options) {
|
|
|
4863
5924
|
const handleSignal = async () => {
|
|
4864
5925
|
if (state.shuttingDown) return;
|
|
4865
5926
|
state.shuttingDown = true;
|
|
5927
|
+
const shutdownStartedAt = Date.now();
|
|
4866
5928
|
if (state.interactive) {
|
|
4867
5929
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
4868
5930
|
displayStatus(state);
|
|
4869
5931
|
} else {
|
|
4870
5932
|
log2(state, "Shutting down...");
|
|
4871
5933
|
}
|
|
4872
|
-
await cleanup(state, { graceful: true });
|
|
4873
|
-
|
|
5934
|
+
const durations = await cleanup(state, { graceful: true });
|
|
5935
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
5936
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
5937
|
+
let timer;
|
|
5938
|
+
const flushed = shutdownTelemetry().then(
|
|
5939
|
+
() => true,
|
|
5940
|
+
(error2) => {
|
|
5941
|
+
log2(
|
|
5942
|
+
state,
|
|
5943
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
5944
|
+
"warn"
|
|
5945
|
+
);
|
|
5946
|
+
return true;
|
|
5947
|
+
}
|
|
5948
|
+
);
|
|
5949
|
+
const timedOut = new Promise((resolve3) => {
|
|
5950
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
5951
|
+
});
|
|
5952
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
5953
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
5954
|
+
}
|
|
5955
|
+
clearTimeout(timer);
|
|
5956
|
+
});
|
|
5957
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
5958
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
4874
5959
|
process.exit(0);
|
|
4875
5960
|
};
|
|
4876
5961
|
process.on("SIGINT", handleSignal);
|
|
@@ -4984,6 +6069,21 @@ async function run(options) {
|
|
|
4984
6069
|
}
|
|
4985
6070
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
4986
6071
|
state.agentName = validation.agent.name;
|
|
6072
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
6073
|
+
if (microvmId) {
|
|
6074
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
6075
|
+
if (reported.ok) {
|
|
6076
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
6077
|
+
} else {
|
|
6078
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
6079
|
+
log2(state, message, "warn");
|
|
6080
|
+
if (state.interactive && !state.json) {
|
|
6081
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6082
|
+
}
|
|
6083
|
+
}
|
|
6084
|
+
} else {
|
|
6085
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
6086
|
+
}
|
|
4987
6087
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
4988
6088
|
try {
|
|
4989
6089
|
const oc = await ensureOpenCodeRunning({
|
|
@@ -5032,6 +6132,10 @@ async function run(options) {
|
|
|
5032
6132
|
getAuthHeader: () => state.authHeader,
|
|
5033
6133
|
conversationFilter: state.conversationFilter,
|
|
5034
6134
|
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
6135
|
+
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
6136
|
+
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
6137
|
+
fileSyncDirectories,
|
|
6138
|
+
homeDir: homedir2(),
|
|
5035
6139
|
log: (entry) => (
|
|
5036
6140
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
5037
6141
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -5058,6 +6162,18 @@ async function run(options) {
|
|
|
5058
6162
|
type: "info",
|
|
5059
6163
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5060
6164
|
});
|
|
6165
|
+
if (options.tunnelReadyFile) {
|
|
6166
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
6167
|
+
if (marker.ok) {
|
|
6168
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
6169
|
+
} else {
|
|
6170
|
+
log2(
|
|
6171
|
+
state,
|
|
6172
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
6173
|
+
"error"
|
|
6174
|
+
);
|
|
6175
|
+
}
|
|
6176
|
+
}
|
|
5061
6177
|
emitAgentConnected(state.agentId, {
|
|
5062
6178
|
port: state.port,
|
|
5063
6179
|
cli_version: getCliVersion(),
|
|
@@ -5113,6 +6229,12 @@ async function run(options) {
|
|
|
5113
6229
|
onDrainPing: () => {
|
|
5114
6230
|
if (!state.running) return;
|
|
5115
6231
|
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
6232
|
+
void channelDriver.syncPendingFiles().catch(
|
|
6233
|
+
(error2) => logActivity(state, {
|
|
6234
|
+
type: "error",
|
|
6235
|
+
error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
6236
|
+
})
|
|
6237
|
+
);
|
|
5116
6238
|
channelDriver.drainPending().then((processed) => {
|
|
5117
6239
|
if (processed > 0) {
|
|
5118
6240
|
state.messageCount += processed;
|
|
@@ -5196,7 +6318,10 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5196
6318
|
program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
|
|
5197
6319
|
program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
|
|
5198
6320
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
5199
|
-
program.command("run").description("Connect to Evident and process messages").option("
|
|
6321
|
+
program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
|
|
6322
|
+
"-a, --agent [id]",
|
|
6323
|
+
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
6324
|
+
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5200
6325
|
"--log-level <level>",
|
|
5201
6326
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5202
6327
|
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").option(
|
|
@@ -5208,6 +6333,14 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5208
6333
|
).option(
|
|
5209
6334
|
"--session-cleanup-interval <duration>",
|
|
5210
6335
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
6336
|
+
).option(
|
|
6337
|
+
"--enable-file-sync-to <dir>",
|
|
6338
|
+
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
6339
|
+
(value, previous) => previous.concat([value]),
|
|
6340
|
+
[]
|
|
6341
|
+
).option(
|
|
6342
|
+
"--tunnel-ready-file <path>",
|
|
6343
|
+
"Path to write once the tunnel is connected (set by the MicroVM hooks; unused on a developer machine)"
|
|
5211
6344
|
).action(
|
|
5212
6345
|
(options) => {
|
|
5213
6346
|
run({
|
|
@@ -5224,7 +6357,11 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5224
6357
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5225
6358
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5226
6359
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5227
|
-
sessionCleanupInterval: options.sessionCleanupInterval
|
|
6360
|
+
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
6361
|
+
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
6362
|
+
// resolveFileSyncDirectories.
|
|
6363
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
6364
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5228
6365
|
});
|
|
5229
6366
|
}
|
|
5230
6367
|
);
|