@evident-ai/cli 3.0.0 → 3.0.1-dev.0385cc8
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 +131 -87
- package/dist/index.js +2504 -217
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
|
|
3
3
|
// src/index.ts
|
|
4
|
+
import { createRequire } from "module";
|
|
4
5
|
import { Command } from "commander";
|
|
5
6
|
|
|
6
7
|
// src/commands/login.ts
|
|
@@ -32,10 +33,10 @@ function setTunnelUrl(url) {
|
|
|
32
33
|
tunnelOverride = url ? url.replace(/\/+$/, "") : void 0;
|
|
33
34
|
}
|
|
34
35
|
function getApiUrl() {
|
|
35
|
-
return process.env.EVIDENT_API_URL ??
|
|
36
|
+
return endpointOverride ?? process.env.EVIDENT_API_URL ?? defaults.apiUrl;
|
|
36
37
|
}
|
|
37
38
|
function getTunnelUrl() {
|
|
38
|
-
return process.env.EVIDENT_TUNNEL_URL ??
|
|
39
|
+
return tunnelOverride ?? process.env.EVIDENT_TUNNEL_URL ?? defaults.tunnelUrl;
|
|
39
40
|
}
|
|
40
41
|
var config = new Conf({
|
|
41
42
|
projectName: "evident",
|
|
@@ -54,19 +55,28 @@ function getApiUrlConfig() {
|
|
|
54
55
|
function getTunnelUrlConfig() {
|
|
55
56
|
return getTunnelUrl();
|
|
56
57
|
}
|
|
58
|
+
function credentialsKey() {
|
|
59
|
+
return getApiUrl();
|
|
60
|
+
}
|
|
57
61
|
function getCredentials() {
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
user: credentials.get("user"),
|
|
61
|
-
expiresAt: credentials.get("expiresAt")
|
|
62
|
-
};
|
|
62
|
+
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
63
|
+
return byEndpoint[credentialsKey()] ?? {};
|
|
63
64
|
}
|
|
64
65
|
function setCredentials(creds) {
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
66
|
+
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
67
|
+
byEndpoint[credentialsKey()] = {
|
|
68
|
+
token: creds.token,
|
|
69
|
+
user: creds.user,
|
|
70
|
+
expiresAt: creds.expiresAt
|
|
71
|
+
};
|
|
72
|
+
credentials.set("byEndpoint", byEndpoint);
|
|
68
73
|
}
|
|
69
74
|
function clearCredentials() {
|
|
75
|
+
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
76
|
+
delete byEndpoint[credentialsKey()];
|
|
77
|
+
credentials.set("byEndpoint", byEndpoint);
|
|
78
|
+
}
|
|
79
|
+
function clearAllCredentials() {
|
|
70
80
|
credentials.clear();
|
|
71
81
|
}
|
|
72
82
|
function getCliName() {
|
|
@@ -176,7 +186,6 @@ var api = {
|
|
|
176
186
|
|
|
177
187
|
// src/lib/keychain.ts
|
|
178
188
|
var SERVICE_NAME = "evident-cli";
|
|
179
|
-
var ACCOUNT_NAME = "default";
|
|
180
189
|
async function getKeytar() {
|
|
181
190
|
try {
|
|
182
191
|
const keytar = await import("keytar");
|
|
@@ -188,10 +197,13 @@ async function getKeytar() {
|
|
|
188
197
|
return null;
|
|
189
198
|
}
|
|
190
199
|
}
|
|
200
|
+
function keychainAccount() {
|
|
201
|
+
return getApiUrlConfig();
|
|
202
|
+
}
|
|
191
203
|
async function storeToken(credentials2) {
|
|
192
204
|
const keytar = await getKeytar();
|
|
193
205
|
if (keytar) {
|
|
194
|
-
await keytar.setPassword(SERVICE_NAME,
|
|
206
|
+
await keytar.setPassword(SERVICE_NAME, keychainAccount(), JSON.stringify(credentials2));
|
|
195
207
|
} else {
|
|
196
208
|
setCredentials({
|
|
197
209
|
token: credentials2.token,
|
|
@@ -203,12 +215,13 @@ async function storeToken(credentials2) {
|
|
|
203
215
|
async function getToken() {
|
|
204
216
|
const keytar = await getKeytar();
|
|
205
217
|
if (keytar) {
|
|
206
|
-
const
|
|
218
|
+
const account = keychainAccount();
|
|
219
|
+
const stored = await keytar.getPassword(SERVICE_NAME, account);
|
|
207
220
|
if (stored) {
|
|
208
221
|
try {
|
|
209
222
|
return JSON.parse(stored);
|
|
210
223
|
} catch {
|
|
211
|
-
await keytar.deletePassword(SERVICE_NAME,
|
|
224
|
+
await keytar.deletePassword(SERVICE_NAME, account);
|
|
212
225
|
return null;
|
|
213
226
|
}
|
|
214
227
|
}
|
|
@@ -223,12 +236,26 @@ async function getToken() {
|
|
|
223
236
|
}
|
|
224
237
|
return null;
|
|
225
238
|
}
|
|
226
|
-
async function deleteToken() {
|
|
239
|
+
async function deleteToken(options = {}) {
|
|
227
240
|
const keytar = await getKeytar();
|
|
228
241
|
if (keytar) {
|
|
229
|
-
|
|
242
|
+
if (options.all) {
|
|
243
|
+
const all = await keytar.findCredentials(SERVICE_NAME).catch(() => []);
|
|
244
|
+
await Promise.all(
|
|
245
|
+
all.map(
|
|
246
|
+
(entry) => keytar.deletePassword(SERVICE_NAME, entry.account).catch(() => {
|
|
247
|
+
})
|
|
248
|
+
)
|
|
249
|
+
);
|
|
250
|
+
} else {
|
|
251
|
+
await keytar.deletePassword(SERVICE_NAME, keychainAccount());
|
|
252
|
+
}
|
|
253
|
+
}
|
|
254
|
+
if (options.all) {
|
|
255
|
+
clearAllCredentials();
|
|
256
|
+
} else {
|
|
257
|
+
clearCredentials();
|
|
230
258
|
}
|
|
231
|
-
clearCredentials();
|
|
232
259
|
}
|
|
233
260
|
|
|
234
261
|
// src/utils/ui.ts
|
|
@@ -258,14 +285,14 @@ function blank() {
|
|
|
258
285
|
console.log();
|
|
259
286
|
}
|
|
260
287
|
function waitForEnter(prompt = "Press Enter to continue...") {
|
|
261
|
-
return new Promise((
|
|
288
|
+
return new Promise((resolve2) => {
|
|
262
289
|
process.stdout.write(chalk.dim(prompt));
|
|
263
290
|
const handler = () => {
|
|
264
291
|
process.stdin.removeListener("data", handler);
|
|
265
292
|
process.stdin.setRawMode?.(false);
|
|
266
293
|
process.stdin.pause();
|
|
267
294
|
console.log();
|
|
268
|
-
|
|
295
|
+
resolve2();
|
|
269
296
|
};
|
|
270
297
|
if (process.stdin.isTTY) {
|
|
271
298
|
process.stdin.setRawMode?.(true);
|
|
@@ -275,7 +302,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
|
|
|
275
302
|
});
|
|
276
303
|
}
|
|
277
304
|
function sleep(ms) {
|
|
278
|
-
return new Promise((
|
|
305
|
+
return new Promise((resolve2) => setTimeout(resolve2, ms));
|
|
279
306
|
}
|
|
280
307
|
|
|
281
308
|
// src/commands/login.ts
|
|
@@ -349,19 +376,19 @@ async function tokenLogin() {
|
|
|
349
376
|
console.log("Visit your Evident dashboard to generate a CLI token.");
|
|
350
377
|
blank();
|
|
351
378
|
process.stdout.write("Paste token: ");
|
|
352
|
-
const token = await new Promise((
|
|
379
|
+
const token = await new Promise((resolve2) => {
|
|
353
380
|
let data = "";
|
|
354
381
|
process.stdin.setEncoding("utf8");
|
|
355
382
|
process.stdin.on("data", (chunk) => {
|
|
356
383
|
data += chunk;
|
|
357
384
|
});
|
|
358
385
|
process.stdin.on("end", () => {
|
|
359
|
-
|
|
386
|
+
resolve2(data.trim());
|
|
360
387
|
});
|
|
361
388
|
if (process.stdin.isTTY) {
|
|
362
389
|
process.stdin.once("data", (chunk) => {
|
|
363
390
|
process.stdin.pause();
|
|
364
|
-
|
|
391
|
+
resolve2(chunk.toString().trim());
|
|
365
392
|
});
|
|
366
393
|
process.stdin.resume();
|
|
367
394
|
}
|
|
@@ -396,25 +423,32 @@ async function login(options) {
|
|
|
396
423
|
}
|
|
397
424
|
|
|
398
425
|
// src/commands/logout.ts
|
|
399
|
-
async function logout() {
|
|
426
|
+
async function logout(options = {}) {
|
|
427
|
+
if (options.all) {
|
|
428
|
+
await deleteToken({ all: true });
|
|
429
|
+
printSuccess("Logged out of all endpoints.");
|
|
430
|
+
return;
|
|
431
|
+
}
|
|
400
432
|
const credentials2 = await getToken();
|
|
401
433
|
if (!credentials2) {
|
|
402
|
-
printWarning(
|
|
434
|
+
printWarning(`You are not logged in to ${getApiUrlConfig()}.`);
|
|
403
435
|
return;
|
|
404
436
|
}
|
|
405
437
|
await deleteToken();
|
|
406
|
-
printSuccess(
|
|
438
|
+
printSuccess(`Logged out of ${getApiUrlConfig()}.`);
|
|
407
439
|
}
|
|
408
440
|
|
|
409
441
|
// src/commands/whoami.ts
|
|
410
442
|
import chalk3 from "chalk";
|
|
411
443
|
async function whoami() {
|
|
444
|
+
const apiUrl = getApiUrlConfig();
|
|
412
445
|
const credentials2 = await getToken();
|
|
413
446
|
if (!credentials2) {
|
|
414
|
-
printError(
|
|
447
|
+
printError(`Not logged in to ${apiUrl}. Run the \`login\` command to authenticate.`);
|
|
415
448
|
process.exit(1);
|
|
416
449
|
}
|
|
417
450
|
blank();
|
|
451
|
+
console.log(keyValue("Endpoint", apiUrl));
|
|
418
452
|
console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
|
|
419
453
|
console.log(keyValue("User ID", credentials2.user.id));
|
|
420
454
|
if (credentials2.expiresAt) {
|
|
@@ -449,9 +483,36 @@ var TelemetryEventTypes = {
|
|
|
449
483
|
|
|
450
484
|
// ../../packages/types/src/tunnel/index.ts
|
|
451
485
|
var MAX_FRAME_BYTES = 256 * 1024;
|
|
486
|
+
var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
|
|
487
|
+
|
|
488
|
+
// ../../packages/types/src/logging/index.ts
|
|
489
|
+
var CORRELATION_ID_HEADER = "x-evident-correlation-id";
|
|
490
|
+
function log(level, event, fields) {
|
|
491
|
+
const method = level === "debug" ? "log" : level;
|
|
492
|
+
try {
|
|
493
|
+
console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
|
|
494
|
+
} catch (err) {
|
|
495
|
+
console.error(
|
|
496
|
+
"[evident] log_serialize_failed",
|
|
497
|
+
event,
|
|
498
|
+
err instanceof Error ? err.message : String(err)
|
|
499
|
+
);
|
|
500
|
+
}
|
|
501
|
+
}
|
|
502
|
+
function stripQuery(url) {
|
|
503
|
+
try {
|
|
504
|
+
return new URL(url).pathname;
|
|
505
|
+
} catch {
|
|
506
|
+
const q = url.indexOf("?");
|
|
507
|
+
return q === -1 ? url : url.slice(0, q);
|
|
508
|
+
}
|
|
509
|
+
}
|
|
452
510
|
|
|
453
511
|
// src/lib/telemetry.ts
|
|
454
|
-
var CLI_VERSION = process.env.npm_package_version
|
|
512
|
+
var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
|
|
513
|
+
function getCliVersion() {
|
|
514
|
+
return CLI_VERSION;
|
|
515
|
+
}
|
|
455
516
|
var eventBuffer = [];
|
|
456
517
|
var flushTimeout = null;
|
|
457
518
|
var isShuttingDown = false;
|
|
@@ -645,11 +706,24 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
|
645
706
|
if (health.healthy) {
|
|
646
707
|
return health;
|
|
647
708
|
}
|
|
648
|
-
await new Promise((
|
|
709
|
+
await new Promise((resolve2) => setTimeout(resolve2, 1e3));
|
|
649
710
|
}
|
|
650
711
|
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
651
712
|
}
|
|
652
713
|
|
|
714
|
+
// src/lib/opencode/opencode-version-gate.ts
|
|
715
|
+
var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
|
|
716
|
+
function isQueueValidatedVersion(version2) {
|
|
717
|
+
if (!version2) return false;
|
|
718
|
+
return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
|
|
719
|
+
}
|
|
720
|
+
function buildOpenCodeVersionWarning(version2) {
|
|
721
|
+
if (isQueueValidatedVersion(version2)) return null;
|
|
722
|
+
const detected = version2 ? `v${version2}` : "unknown";
|
|
723
|
+
const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
|
|
724
|
+
return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
|
|
725
|
+
}
|
|
726
|
+
|
|
653
727
|
// src/lib/opencode/process.ts
|
|
654
728
|
import { execSync, spawn } from "child_process";
|
|
655
729
|
var OPENCODE_PORT_RANGE = [4096, 4097, 4098, 4099, 4100];
|
|
@@ -940,19 +1014,174 @@ async function promptOpenCodeInstall(interactive) {
|
|
|
940
1014
|
}
|
|
941
1015
|
|
|
942
1016
|
// src/lib/opencode/session.ts
|
|
943
|
-
|
|
944
|
-
|
|
1017
|
+
function opencodeBase(port) {
|
|
1018
|
+
return `http://127.0.0.1:${port}`;
|
|
1019
|
+
}
|
|
1020
|
+
async function getOpenCodeDirectory(port) {
|
|
1021
|
+
try {
|
|
1022
|
+
const res = await fetch(`${opencodeBase(port)}/path`);
|
|
1023
|
+
if (!res.ok) return null;
|
|
1024
|
+
const body = await res.json();
|
|
1025
|
+
const dir = typeof body.directory === "string" && body.directory || typeof body.worktree === "string" && body.worktree || typeof body.path?.cwd === "string" && body.path.cwd || typeof body.path?.directory === "string" && body.path.directory || null;
|
|
1026
|
+
return dir && dir.trim() ? dir.trim() : null;
|
|
1027
|
+
} catch {
|
|
1028
|
+
return null;
|
|
1029
|
+
}
|
|
1030
|
+
}
|
|
1031
|
+
function roleOf(m) {
|
|
1032
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1033
|
+
if (typeof m.role === "string") return m.role;
|
|
1034
|
+
const infoRole = m.info?.role;
|
|
1035
|
+
return typeof infoRole === "string" ? infoRole : void 0;
|
|
1036
|
+
}
|
|
1037
|
+
function completedOf(m) {
|
|
1038
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1039
|
+
return m.info?.time?.completed ?? m.time?.completed;
|
|
1040
|
+
}
|
|
1041
|
+
function createdOf(m) {
|
|
1042
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1043
|
+
return m.info?.time?.created ?? m.time?.created;
|
|
1044
|
+
}
|
|
1045
|
+
function idOf(m) {
|
|
1046
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1047
|
+
if (typeof m.id === "string") return m.id;
|
|
1048
|
+
const infoId = m.info?.id;
|
|
1049
|
+
return typeof infoId === "string" ? infoId : void 0;
|
|
1050
|
+
}
|
|
1051
|
+
function parentIdOf(m) {
|
|
1052
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1053
|
+
if (typeof m.parentID === "string") return m.parentID;
|
|
1054
|
+
const infoParent = m.info?.parentID;
|
|
1055
|
+
return typeof infoParent === "string" ? infoParent : void 0;
|
|
1056
|
+
}
|
|
1057
|
+
function finishOf(m) {
|
|
1058
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1059
|
+
if (typeof m.finish === "string") return m.finish;
|
|
1060
|
+
const infoFinish = m.info?.finish;
|
|
1061
|
+
return typeof infoFinish === "string" ? infoFinish : void 0;
|
|
1062
|
+
}
|
|
1063
|
+
function errorOf(m) {
|
|
1064
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1065
|
+
return m.info?.error ?? m.error;
|
|
1066
|
+
}
|
|
1067
|
+
function isAssistantInFlight(m) {
|
|
1068
|
+
if (completedOf(m) == null) return true;
|
|
1069
|
+
return finishOf(m) === "tool-calls";
|
|
1070
|
+
}
|
|
1071
|
+
async function getSessionMessages(port, sessionId) {
|
|
1072
|
+
try {
|
|
1073
|
+
const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
|
|
1074
|
+
if (!res.ok) return null;
|
|
1075
|
+
const body = await res.json();
|
|
1076
|
+
return Array.isArray(body) ? body : null;
|
|
1077
|
+
} catch {
|
|
1078
|
+
return null;
|
|
1079
|
+
}
|
|
1080
|
+
}
|
|
1081
|
+
function isSessionActivelyGenerating(messages) {
|
|
1082
|
+
if (!messages || messages.length === 0) return false;
|
|
1083
|
+
const last = messages[messages.length - 1];
|
|
1084
|
+
if (roleOf(last) !== "assistant") return false;
|
|
1085
|
+
return completedOf(last) == null;
|
|
1086
|
+
}
|
|
1087
|
+
function sessionLastActivityMs(session) {
|
|
1088
|
+
const candidates = [
|
|
1089
|
+
session.time?.updated,
|
|
1090
|
+
session.time?.created,
|
|
1091
|
+
session.time_updated,
|
|
1092
|
+
session.time_created,
|
|
1093
|
+
session.updated,
|
|
1094
|
+
session.created
|
|
1095
|
+
];
|
|
1096
|
+
for (const c of candidates) {
|
|
1097
|
+
if (typeof c === "number" && Number.isFinite(c)) return c;
|
|
1098
|
+
}
|
|
1099
|
+
return null;
|
|
1100
|
+
}
|
|
1101
|
+
async function listSessions(port) {
|
|
1102
|
+
try {
|
|
1103
|
+
const res = await fetch(`${opencodeBase(port)}/session`);
|
|
1104
|
+
if (!res.ok) return null;
|
|
1105
|
+
const body = await res.json();
|
|
1106
|
+
return Array.isArray(body) ? body : null;
|
|
1107
|
+
} catch {
|
|
1108
|
+
return null;
|
|
1109
|
+
}
|
|
1110
|
+
}
|
|
1111
|
+
async function deleteSession(port, id) {
|
|
1112
|
+
try {
|
|
1113
|
+
const res = await fetch(`${opencodeBase(port)}/session/${id}`, { method: "DELETE" });
|
|
1114
|
+
return res.status >= 200 && res.status < 300;
|
|
1115
|
+
} catch {
|
|
1116
|
+
return false;
|
|
1117
|
+
}
|
|
1118
|
+
}
|
|
1119
|
+
async function sessionExists(port, id) {
|
|
1120
|
+
try {
|
|
1121
|
+
const res = await fetch(`${opencodeBase(port)}/session/${id}`);
|
|
1122
|
+
if (res.status >= 200 && res.status < 300) return true;
|
|
1123
|
+
if (res.status === 404) return false;
|
|
1124
|
+
return null;
|
|
1125
|
+
} catch {
|
|
1126
|
+
return null;
|
|
1127
|
+
}
|
|
1128
|
+
}
|
|
1129
|
+
async function getSessionStatuses(port) {
|
|
1130
|
+
try {
|
|
1131
|
+
const res = await fetch(`${opencodeBase(port)}/session/status`);
|
|
1132
|
+
if (!res.ok) {
|
|
1133
|
+
console.error(
|
|
1134
|
+
`[getSessionStatuses] GET /session/status returned HTTP ${res.status} (port ${port})`
|
|
1135
|
+
);
|
|
1136
|
+
return null;
|
|
1137
|
+
}
|
|
1138
|
+
const body = await res.json();
|
|
1139
|
+
if (body == null || typeof body !== "object" || Array.isArray(body)) {
|
|
1140
|
+
console.error(
|
|
1141
|
+
`[getSessionStatuses] GET /session/status body was not a plain object (port ${port})`
|
|
1142
|
+
);
|
|
1143
|
+
return null;
|
|
1144
|
+
}
|
|
1145
|
+
return body;
|
|
1146
|
+
} catch (err) {
|
|
1147
|
+
console.error(
|
|
1148
|
+
`[getSessionStatuses] GET /session/status failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
|
|
1149
|
+
);
|
|
1150
|
+
return null;
|
|
1151
|
+
}
|
|
1152
|
+
}
|
|
1153
|
+
async function isSessionOngoing(port, id) {
|
|
1154
|
+
const map = await getSessionStatuses(port);
|
|
1155
|
+
if (map == null) return null;
|
|
1156
|
+
const entry = map[id];
|
|
1157
|
+
return entry != null && entry.type !== "idle";
|
|
1158
|
+
}
|
|
1159
|
+
async function createOpenCodeSession(port, directory) {
|
|
1160
|
+
const url = new URL(`${opencodeBase(port)}/session`);
|
|
1161
|
+
if (directory && directory.trim()) {
|
|
1162
|
+
url.searchParams.set("directory", directory.trim());
|
|
1163
|
+
}
|
|
1164
|
+
const response = await fetch(url, {
|
|
945
1165
|
method: "POST",
|
|
946
1166
|
headers: { "Content-Type": "application/json" },
|
|
947
1167
|
body: JSON.stringify({})
|
|
948
1168
|
});
|
|
949
1169
|
if (!response.ok) {
|
|
950
|
-
|
|
1170
|
+
const text = await response.text().catch(() => "");
|
|
1171
|
+
throw new Error(`Failed to create session: HTTP ${response.status}${text ? `: ${text}` : ""}`);
|
|
951
1172
|
}
|
|
952
1173
|
const data = await response.json();
|
|
953
1174
|
return data.id;
|
|
954
1175
|
}
|
|
955
|
-
|
|
1176
|
+
function messageText(m) {
|
|
1177
|
+
if (!m || !Array.isArray(m.parts)) return "";
|
|
1178
|
+
return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
|
|
1179
|
+
}
|
|
1180
|
+
async function sendPromptAsync(port, sessionId, content, options) {
|
|
1181
|
+
const before = await getSessionMessages(port, sessionId);
|
|
1182
|
+
const knownUserIds = new Set(
|
|
1183
|
+
(before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
|
|
1184
|
+
);
|
|
956
1185
|
const body = {
|
|
957
1186
|
parts: [{ type: "text", text: content }]
|
|
958
1187
|
};
|
|
@@ -968,76 +1197,218 @@ async function sendMessageToOpenCode(port, sessionId, content, options, hooks, m
|
|
|
968
1197
|
};
|
|
969
1198
|
}
|
|
970
1199
|
}
|
|
971
|
-
|
|
972
|
-
|
|
973
|
-
|
|
974
|
-
|
|
975
|
-
|
|
976
|
-
|
|
977
|
-
|
|
978
|
-
|
|
979
|
-
|
|
980
|
-
|
|
981
|
-
|
|
982
|
-
|
|
983
|
-
|
|
984
|
-
|
|
985
|
-
|
|
986
|
-
|
|
987
|
-
|
|
988
|
-
|
|
989
|
-
|
|
990
|
-
|
|
991
|
-
|
|
992
|
-
|
|
993
|
-
|
|
994
|
-
try {
|
|
995
|
-
const res = await fetch(`http://localhost:${port}/permission`);
|
|
996
|
-
if (res.ok) {
|
|
997
|
-
const permissions = await res.json();
|
|
998
|
-
for (const p of permissions) {
|
|
999
|
-
if (p.sessionID === sessionId && !reportedPermissions.has(p.id)) {
|
|
1000
|
-
reportedPermissions.add(p.id);
|
|
1001
|
-
await hooks.onPermission(p);
|
|
1002
|
-
}
|
|
1003
|
-
}
|
|
1004
|
-
}
|
|
1005
|
-
} catch {
|
|
1200
|
+
const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/prompt_async`, {
|
|
1201
|
+
method: "POST",
|
|
1202
|
+
headers: { "Content-Type": "application/json" },
|
|
1203
|
+
body: JSON.stringify(body)
|
|
1204
|
+
});
|
|
1205
|
+
if (res.status < 200 || res.status >= 300) {
|
|
1206
|
+
const text = await res.text().catch(() => "");
|
|
1207
|
+
throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
|
|
1208
|
+
}
|
|
1209
|
+
const READ_BACK_ATTEMPTS = 5;
|
|
1210
|
+
const READ_BACK_DELAY_MS = 150;
|
|
1211
|
+
for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
|
|
1212
|
+
const after = await getSessionMessages(port, sessionId);
|
|
1213
|
+
if (after) {
|
|
1214
|
+
let best = null;
|
|
1215
|
+
for (const m of after) {
|
|
1216
|
+
if (roleOf(m) !== "user") continue;
|
|
1217
|
+
const id = idOf(m);
|
|
1218
|
+
if (typeof id !== "string" || knownUserIds.has(id)) continue;
|
|
1219
|
+
if (messageText(m) !== content) continue;
|
|
1220
|
+
const created = createdOf(m) ?? 0;
|
|
1221
|
+
if (best === null || created > best.created) {
|
|
1222
|
+
best = { id, created };
|
|
1006
1223
|
}
|
|
1007
1224
|
}
|
|
1225
|
+
if (best) return best.id;
|
|
1226
|
+
}
|
|
1227
|
+
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1228
|
+
await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
|
|
1229
|
+
}
|
|
1230
|
+
}
|
|
1231
|
+
return null;
|
|
1232
|
+
}
|
|
1233
|
+
function findAssistantReplyAfter(messages, userMessageId) {
|
|
1234
|
+
if (!messages || messages.length === 0) return null;
|
|
1235
|
+
const byParent = messages.find(
|
|
1236
|
+
(m) => roleOf(m) === "assistant" && parentIdOf(m) === userMessageId
|
|
1237
|
+
);
|
|
1238
|
+
if (byParent) return byParent;
|
|
1239
|
+
const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
|
|
1240
|
+
if (userIndex === -1) return null;
|
|
1241
|
+
for (let i = userIndex + 1; i < messages.length; i++) {
|
|
1242
|
+
if (roleOf(messages[i]) === "assistant") return messages[i];
|
|
1243
|
+
}
|
|
1244
|
+
return null;
|
|
1245
|
+
}
|
|
1246
|
+
function findLastAssistantReplyFor(messages, userMessageId) {
|
|
1247
|
+
if (!messages || messages.length === 0) return null;
|
|
1248
|
+
let lastCorrelated = null;
|
|
1249
|
+
let lastNonErrored = null;
|
|
1250
|
+
for (let i = messages.length - 1; i >= 0; i--) {
|
|
1251
|
+
const m = messages[i];
|
|
1252
|
+
if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
|
|
1253
|
+
if (lastCorrelated === null) lastCorrelated = m;
|
|
1254
|
+
if (errorOf(m) == null) {
|
|
1255
|
+
lastNonErrored = m;
|
|
1256
|
+
break;
|
|
1257
|
+
}
|
|
1258
|
+
}
|
|
1259
|
+
if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
|
|
1260
|
+
const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
|
|
1261
|
+
if (userIndex === -1) return null;
|
|
1262
|
+
let last = null;
|
|
1263
|
+
let lastOk = null;
|
|
1264
|
+
for (let i = userIndex + 1; i < messages.length; i++) {
|
|
1265
|
+
const role = roleOf(messages[i]);
|
|
1266
|
+
if (role === "user") break;
|
|
1267
|
+
if (role === "assistant") {
|
|
1268
|
+
last = messages[i];
|
|
1269
|
+
if (errorOf(messages[i]) == null) lastOk = messages[i];
|
|
1008
1270
|
}
|
|
1271
|
+
}
|
|
1272
|
+
return lastOk ?? last;
|
|
1273
|
+
}
|
|
1274
|
+
function messageRunState(messages, userMessageId) {
|
|
1275
|
+
if (!messages || messages.length === 0) return "unknown";
|
|
1276
|
+
const hasUser = messages.some((m) => idOf(m) === userMessageId);
|
|
1277
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1278
|
+
if (!hasUser) {
|
|
1279
|
+
if (!reply) return "unknown";
|
|
1280
|
+
}
|
|
1281
|
+
if (!reply) return "queued";
|
|
1282
|
+
if (isAssistantInFlight(reply)) return "running";
|
|
1283
|
+
return errorOf(reply) != null ? "failed" : "done";
|
|
1284
|
+
}
|
|
1285
|
+
function isPreamblePinnedRunning(messages, userMessageId) {
|
|
1286
|
+
if (messageRunState(messages, userMessageId) !== "running") return false;
|
|
1287
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1288
|
+
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1289
|
+
}
|
|
1290
|
+
function messageError(messages, userMessageId) {
|
|
1291
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1292
|
+
const error2 = errorOf(reply);
|
|
1293
|
+
if (error2 == null) return null;
|
|
1294
|
+
if (typeof error2 === "string") return error2;
|
|
1295
|
+
if (typeof error2 === "object") {
|
|
1296
|
+
const e = error2;
|
|
1297
|
+
const dataMessage = e.data?.message;
|
|
1298
|
+
if (typeof dataMessage === "string") return dataMessage;
|
|
1299
|
+
if (typeof e.message === "string") return e.message;
|
|
1300
|
+
}
|
|
1301
|
+
return "The agent run failed.";
|
|
1302
|
+
}
|
|
1303
|
+
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1304
|
+
if (!messages || messages.length === 0) return false;
|
|
1305
|
+
return messages.some(
|
|
1306
|
+
(m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
|
|
1307
|
+
);
|
|
1308
|
+
}
|
|
1309
|
+
|
|
1310
|
+
// src/lib/opencode/session-cleanup.ts
|
|
1311
|
+
var DURATION_UNIT_MS = {
|
|
1312
|
+
s: 1e3,
|
|
1313
|
+
m: 60 * 1e3,
|
|
1314
|
+
h: 60 * 60 * 1e3,
|
|
1315
|
+
d: 24 * 60 * 60 * 1e3
|
|
1316
|
+
};
|
|
1317
|
+
function parseDurationMs(input) {
|
|
1318
|
+
const trimmed = input.trim();
|
|
1319
|
+
const match = /^(\d+)([smhd])$/.exec(trimmed);
|
|
1320
|
+
if (!match) {
|
|
1321
|
+
throw new Error(
|
|
1322
|
+
`Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
|
|
1323
|
+
);
|
|
1324
|
+
}
|
|
1325
|
+
const value = Number(match[1]);
|
|
1326
|
+
if (value <= 0) {
|
|
1327
|
+
throw new Error(`Invalid duration "${input}": must be a positive value.`);
|
|
1328
|
+
}
|
|
1329
|
+
return value * DURATION_UNIT_MS[match[2]];
|
|
1330
|
+
}
|
|
1331
|
+
function selectSessionsToDelete(sessions, opts) {
|
|
1332
|
+
const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
|
|
1333
|
+
if (maxAgeMs === void 0 && maxCount === void 0) return [];
|
|
1334
|
+
const ageEligible = (s) => {
|
|
1335
|
+
if (maxAgeMs === void 0) return false;
|
|
1336
|
+
if (s.lastActivityMs === null) return true;
|
|
1337
|
+
return nowMs - s.lastActivityMs > maxAgeMs;
|
|
1009
1338
|
};
|
|
1010
|
-
const
|
|
1011
|
-
|
|
1012
|
-
const
|
|
1339
|
+
const countEligibleIds = /* @__PURE__ */ new Set();
|
|
1340
|
+
if (maxCount !== void 0) {
|
|
1341
|
+
const byActivityDesc = [...sessions].sort(
|
|
1342
|
+
(a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
|
|
1343
|
+
);
|
|
1344
|
+
for (const s of byActivityDesc.slice(maxCount)) {
|
|
1345
|
+
countEligibleIds.add(s.id);
|
|
1346
|
+
}
|
|
1347
|
+
}
|
|
1348
|
+
const toDelete = [];
|
|
1349
|
+
for (const s of sessions) {
|
|
1350
|
+
if (protectedIds.has(s.id)) continue;
|
|
1351
|
+
if (ageEligible(s) || countEligibleIds.has(s.id)) {
|
|
1352
|
+
toDelete.push(s.id);
|
|
1353
|
+
}
|
|
1354
|
+
}
|
|
1355
|
+
return toDelete;
|
|
1356
|
+
}
|
|
1357
|
+
var DEFAULT_INTERVAL = "1h";
|
|
1358
|
+
function resolve(flag, envValue, fallback) {
|
|
1359
|
+
return flag ?? envValue ?? fallback;
|
|
1360
|
+
}
|
|
1361
|
+
function parseMaxCount(input) {
|
|
1362
|
+
const trimmed = input.trim();
|
|
1363
|
+
if (!/^\d+$/.test(trimmed)) {
|
|
1364
|
+
throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
|
|
1365
|
+
}
|
|
1366
|
+
const value = Number(trimmed);
|
|
1367
|
+
if (value <= 0) {
|
|
1368
|
+
throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
|
|
1369
|
+
}
|
|
1370
|
+
return value;
|
|
1371
|
+
}
|
|
1372
|
+
function resolveSessionCleanupConfig(flags, env = process.env) {
|
|
1373
|
+
const warnings = [];
|
|
1374
|
+
const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
|
|
1375
|
+
const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
|
|
1376
|
+
const intervalRaw = resolve(
|
|
1377
|
+
flags.interval,
|
|
1378
|
+
env.EVIDENT_SESSION_CLEANUP_INTERVAL,
|
|
1379
|
+
DEFAULT_INTERVAL
|
|
1380
|
+
);
|
|
1381
|
+
let maxAgeMs;
|
|
1382
|
+
if (maxAgeRaw !== void 0) {
|
|
1013
1383
|
try {
|
|
1014
|
-
|
|
1015
|
-
|
|
1016
|
-
|
|
1017
|
-
|
|
1018
|
-
signal: controller.signal
|
|
1019
|
-
});
|
|
1020
|
-
if (!res.ok) {
|
|
1021
|
-
const text = await res.text().catch(() => "");
|
|
1022
|
-
throw new Error(`OpenCode message failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
|
|
1023
|
-
}
|
|
1024
|
-
const sessionRes = await fetch(`http://localhost:${port}/session/${sessionId}`).catch(
|
|
1025
|
-
() => null
|
|
1384
|
+
maxAgeMs = parseDurationMs(maxAgeRaw);
|
|
1385
|
+
} catch (err) {
|
|
1386
|
+
warnings.push(
|
|
1387
|
+
`Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
|
|
1026
1388
|
);
|
|
1027
|
-
|
|
1028
|
-
|
|
1389
|
+
}
|
|
1390
|
+
}
|
|
1391
|
+
let maxCount;
|
|
1392
|
+
if (maxCountRaw !== void 0) {
|
|
1393
|
+
try {
|
|
1394
|
+
maxCount = parseMaxCount(maxCountRaw);
|
|
1029
1395
|
} catch (err) {
|
|
1030
|
-
|
|
1031
|
-
|
|
1032
|
-
|
|
1033
|
-
throw err;
|
|
1034
|
-
} finally {
|
|
1035
|
-
clearTimeout(timer);
|
|
1036
|
-
pollDone = true;
|
|
1396
|
+
warnings.push(
|
|
1397
|
+
`Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
|
|
1398
|
+
);
|
|
1037
1399
|
}
|
|
1038
|
-
}
|
|
1039
|
-
|
|
1040
|
-
|
|
1400
|
+
}
|
|
1401
|
+
let intervalMs;
|
|
1402
|
+
try {
|
|
1403
|
+
intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
|
|
1404
|
+
} catch (err) {
|
|
1405
|
+
warnings.push(
|
|
1406
|
+
`Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
|
|
1407
|
+
);
|
|
1408
|
+
intervalMs = parseDurationMs(DEFAULT_INTERVAL);
|
|
1409
|
+
}
|
|
1410
|
+
const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
|
|
1411
|
+
return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
|
|
1041
1412
|
}
|
|
1042
1413
|
|
|
1043
1414
|
// src/lib/tunnel/connection.ts
|
|
@@ -1108,18 +1479,34 @@ var StreamForwarder = class {
|
|
|
1108
1479
|
}
|
|
1109
1480
|
async handleOpen(frame) {
|
|
1110
1481
|
const { sid, method, path, headers, has_body } = frame;
|
|
1482
|
+
const correlationId = headers?.[CORRELATION_ID_HEADER];
|
|
1483
|
+
const startedAt = Date.now();
|
|
1484
|
+
if (path === TUNNEL_DRAIN_PING_PATH) {
|
|
1485
|
+
this.callbacks.onDrainPing?.();
|
|
1486
|
+
this.send({ type: "head", sid, status: 204, headers: {} });
|
|
1487
|
+
this.send({ type: "res_end", sid });
|
|
1488
|
+
return;
|
|
1489
|
+
}
|
|
1490
|
+
if (process.env.DEBUG) {
|
|
1491
|
+
log("debug", "agent_request", {
|
|
1492
|
+
correlation_id: correlationId,
|
|
1493
|
+
sid,
|
|
1494
|
+
method,
|
|
1495
|
+
path: stripQuery(path)
|
|
1496
|
+
});
|
|
1497
|
+
}
|
|
1111
1498
|
const ac = new AbortController();
|
|
1112
1499
|
let bodyPromise;
|
|
1113
1500
|
let pushBody;
|
|
1114
1501
|
let endBody;
|
|
1115
1502
|
if (has_body) {
|
|
1116
1503
|
const chunks = [];
|
|
1117
|
-
bodyPromise = new Promise((
|
|
1504
|
+
bodyPromise = new Promise((resolve2) => {
|
|
1118
1505
|
pushBody = (buf) => {
|
|
1119
1506
|
chunks.push(buf);
|
|
1120
1507
|
};
|
|
1121
1508
|
endBody = () => {
|
|
1122
|
-
|
|
1509
|
+
resolve2(Buffer.concat(chunks));
|
|
1123
1510
|
};
|
|
1124
1511
|
});
|
|
1125
1512
|
}
|
|
@@ -1154,6 +1541,14 @@ var StreamForwarder = class {
|
|
|
1154
1541
|
if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
|
|
1155
1542
|
});
|
|
1156
1543
|
this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
|
|
1544
|
+
if (process.env.DEBUG) {
|
|
1545
|
+
log("debug", "agent_response", {
|
|
1546
|
+
correlation_id: correlationId,
|
|
1547
|
+
sid,
|
|
1548
|
+
status: upstream.status,
|
|
1549
|
+
duration_ms: Date.now() - startedAt
|
|
1550
|
+
});
|
|
1551
|
+
}
|
|
1157
1552
|
this.callbacks.onHead?.(sid, upstream.status);
|
|
1158
1553
|
try {
|
|
1159
1554
|
if (upstream.body) {
|
|
@@ -1224,11 +1619,12 @@ function connectTunnel(options) {
|
|
|
1224
1619
|
onError,
|
|
1225
1620
|
onRequest,
|
|
1226
1621
|
onResponse,
|
|
1227
|
-
onInfo
|
|
1622
|
+
onInfo,
|
|
1623
|
+
onDrainPing
|
|
1228
1624
|
} = options;
|
|
1229
1625
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1230
1626
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1231
|
-
return new Promise((
|
|
1627
|
+
return new Promise((resolve2, reject) => {
|
|
1232
1628
|
const ws = new WebSocket2(url, {
|
|
1233
1629
|
headers: {
|
|
1234
1630
|
Authorization: authHeader
|
|
@@ -1237,6 +1633,7 @@ function connectTunnel(options) {
|
|
|
1237
1633
|
const streamStartTimes = /* @__PURE__ */ new Map();
|
|
1238
1634
|
const forwarder = new StreamForwarder(ws, port, {
|
|
1239
1635
|
onOpen: (sid, method, path) => {
|
|
1636
|
+
if (path === TUNNEL_DRAIN_PING_PATH) return;
|
|
1240
1637
|
streamStartTimes.set(sid, Date.now());
|
|
1241
1638
|
onRequest?.(method, path, sid);
|
|
1242
1639
|
},
|
|
@@ -1244,7 +1641,8 @@ function connectTunnel(options) {
|
|
|
1244
1641
|
const startedAt = streamStartTimes.get(sid);
|
|
1245
1642
|
streamStartTimes.delete(sid);
|
|
1246
1643
|
onResponse?.(status, startedAt ? Date.now() - startedAt : 0, sid);
|
|
1247
|
-
}
|
|
1644
|
+
},
|
|
1645
|
+
onDrainPing: () => onDrainPing?.()
|
|
1248
1646
|
});
|
|
1249
1647
|
const connectionTimeout = setTimeout(() => {
|
|
1250
1648
|
ws.close();
|
|
@@ -1291,7 +1689,7 @@ function connectTunnel(options) {
|
|
|
1291
1689
|
clearTimeout(connectionTimeout);
|
|
1292
1690
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1293
1691
|
onConnected?.(connectedAgentId);
|
|
1294
|
-
|
|
1692
|
+
resolve2({
|
|
1295
1693
|
ws,
|
|
1296
1694
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1297
1695
|
});
|
|
@@ -1386,6 +1784,7 @@ var RunnerConnection = class {
|
|
|
1386
1784
|
},
|
|
1387
1785
|
onError: (error2) => events.onError?.(error2),
|
|
1388
1786
|
onResponse: () => events.onResponse?.(),
|
|
1787
|
+
onDrainPing: () => events.onDrainPing?.(),
|
|
1389
1788
|
onInfo: (message) => events.onInfo?.(message)
|
|
1390
1789
|
});
|
|
1391
1790
|
return;
|
|
@@ -1406,17 +1805,37 @@ var RunnerConnection = class {
|
|
|
1406
1805
|
};
|
|
1407
1806
|
|
|
1408
1807
|
// src/lib/channels/driver.ts
|
|
1808
|
+
function messageIdOf(m) {
|
|
1809
|
+
if (!m || typeof m !== "object") return void 0;
|
|
1810
|
+
if (typeof m.id === "string") return m.id;
|
|
1811
|
+
const infoId = m.info?.id;
|
|
1812
|
+
return typeof infoId === "string" ? infoId : void 0;
|
|
1813
|
+
}
|
|
1409
1814
|
var DEFAULT_RETRY_POLICY = {
|
|
1410
1815
|
maxAttempts: 6,
|
|
1411
1816
|
baseDelayMs: 500,
|
|
1412
1817
|
maxDelayMs: 3e4
|
|
1413
1818
|
};
|
|
1819
|
+
var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
|
|
1820
|
+
var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
1821
|
+
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
1822
|
+
var HEARTBEAT_MS = 6e4;
|
|
1823
|
+
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
1824
|
+
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
1414
1825
|
var ChannelAuthError = class extends Error {
|
|
1415
1826
|
constructor(message) {
|
|
1416
1827
|
super(message);
|
|
1417
1828
|
this.name = "ChannelAuthError";
|
|
1418
1829
|
}
|
|
1419
1830
|
};
|
|
1831
|
+
var ChannelTerminalError = class extends Error {
|
|
1832
|
+
status;
|
|
1833
|
+
constructor(message, status) {
|
|
1834
|
+
super(message);
|
|
1835
|
+
this.name = "ChannelTerminalError";
|
|
1836
|
+
this.status = status;
|
|
1837
|
+
}
|
|
1838
|
+
};
|
|
1420
1839
|
function backoffDelay(attempt, policy) {
|
|
1421
1840
|
const exp = policy.baseDelayMs * Math.pow(2, attempt);
|
|
1422
1841
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
@@ -1435,10 +1854,127 @@ var ChannelDriver = class {
|
|
|
1435
1854
|
log;
|
|
1436
1855
|
fetchImpl;
|
|
1437
1856
|
sleep;
|
|
1857
|
+
pausedPollIntervalMs;
|
|
1858
|
+
pausedMaxWaitMs;
|
|
1859
|
+
stuckQueuedMs;
|
|
1860
|
+
now;
|
|
1438
1861
|
/** Cache of conversationId → opencode sessionId. */
|
|
1439
1862
|
sessions = /* @__PURE__ */ new Map();
|
|
1863
|
+
/**
|
|
1864
|
+
* Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
|
|
1865
|
+
* longer idempotent (no caller-supplied `messageID`), and its read-back picks
|
|
1866
|
+
* "the one new user row" — which is only unambiguous if no OTHER dispatch into
|
|
1867
|
+
* the SAME session interleaves its snapshot→POST→read-back. This map chains each
|
|
1868
|
+
* session's dispatches so they run serially; distinct sessions stay concurrent.
|
|
1869
|
+
*/
|
|
1870
|
+
sessionDispatchLocks = /* @__PURE__ */ new Map();
|
|
1871
|
+
/**
|
|
1872
|
+
* Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
|
|
1873
|
+
* session: one polling loop services all of that session's in-flight messages.
|
|
1874
|
+
* A session entry exists while it has any in-flight (dispatched-but-not-done)
|
|
1875
|
+
* message; it is removed once its in-flight set empties.
|
|
1876
|
+
*/
|
|
1877
|
+
watchers = /* @__PURE__ */ new Map();
|
|
1878
|
+
/**
|
|
1879
|
+
* AUTHORITATIVE local dedup (WI-3): Evident message ids that have been
|
|
1880
|
+
* dispatched and are still in-flight. A message in this set is never
|
|
1881
|
+
* re-`prompt_async`-ed by a subsequent poll tick while it is queued/running.
|
|
1882
|
+
* Backed by a stable minted opencode `messageID` whose duplicate re-enqueue is
|
|
1883
|
+
* idempotent on opencode (PoC fact 9) — so even if this set is lost on restart,
|
|
1884
|
+
* a steady-state-poll re-dispatch will not double-run the message.
|
|
1885
|
+
*/
|
|
1886
|
+
dispatched = /* @__PURE__ */ new Set();
|
|
1887
|
+
/**
|
|
1888
|
+
* Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
|
|
1889
|
+
* Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
|
|
1890
|
+
* so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
|
|
1891
|
+
* it is re-adopted and removed when its watcher settles or it is observed off
|
|
1892
|
+
* the processing list.
|
|
1893
|
+
*/
|
|
1894
|
+
readopted = /* @__PURE__ */ new Set();
|
|
1895
|
+
/**
|
|
1896
|
+
* "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
|
|
1897
|
+
* re-adopted running/orphan row's watcher hit its `processed_at`-anchored
|
|
1898
|
+
* deadline (or an orphan whose window already elapsed): the still-`processing`
|
|
1899
|
+
* server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
|
|
1900
|
+
* drain until the 15-min cron resets it — spamming new turns.
|
|
1901
|
+
*
|
|
1902
|
+
* CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
|
|
1903
|
+
* does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
|
|
1904
|
+
* in opencode must still be delivered via `markDone` on the next drain — so
|
|
1905
|
+
* `readoptOne` computes `state` FIRST and this set is checked only on the
|
|
1906
|
+
* non-done path. It is cleared once the row leaves the processing list (cron
|
|
1907
|
+
* reset → it drains normally as `pending`), so it can never leak.
|
|
1908
|
+
*/
|
|
1909
|
+
dontRedispatch = /* @__PURE__ */ new Set();
|
|
1910
|
+
/**
|
|
1911
|
+
* "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
|
|
1912
|
+
* re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
|
|
1913
|
+
* never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
|
|
1914
|
+
* that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
|
|
1915
|
+
* markDone failure must NOT land here (it must still retry next drain). Separate
|
|
1916
|
+
* from `dontRedispatch` because the two concerns are independent: a row can need
|
|
1917
|
+
* "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
|
|
1918
|
+
* the row leaves the processing list, exactly like `dontRedispatch`.
|
|
1919
|
+
*/
|
|
1920
|
+
doneUndeliverable = /* @__PURE__ */ new Set();
|
|
1921
|
+
/**
|
|
1922
|
+
* "Already emitted `readopt_poll_unresolved` for this row" (#229). The b1 /
|
|
1923
|
+
* unreadable-status re-evaluate leaf leaves the row UN-tracked so it is re-read
|
|
1924
|
+
* every ~2s drain until the status map becomes readable — but the server-visible
|
|
1925
|
+
* signal is an OUTCOME, so it must fire at most ONCE per row, not once per drain
|
|
1926
|
+
* (Bugbot "Re-adopt signals flood every drain"). Cleared when the row leaves the
|
|
1927
|
+
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
1928
|
+
*/
|
|
1929
|
+
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
1930
|
+
/**
|
|
1931
|
+
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
1932
|
+
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
1933
|
+
* is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
|
|
1934
|
+
* persist hasn't landed before tick N+1 re-reads the still-null
|
|
1935
|
+
* `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
|
|
1936
|
+
* A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
|
|
1937
|
+
* short-circuits while it is present, so a null-id row is re-dispatched AT MOST
|
|
1938
|
+
* ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
|
|
1939
|
+
* row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
|
|
1940
|
+
* re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
|
|
1941
|
+
* so the NEXT tick may retry exactly once more).
|
|
1942
|
+
*/
|
|
1943
|
+
awaitingReadopt = /* @__PURE__ */ new Set();
|
|
1944
|
+
/**
|
|
1945
|
+
* Cache of the opencode root directory (from `GET /path`). Resolved lazily on
|
|
1946
|
+
* first session creation so drain-created sessions are rooted at the project
|
|
1947
|
+
* directory and thus visible in `opencode web`'s session list. `undefined` =
|
|
1948
|
+
* not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
|
|
1949
|
+
*/
|
|
1950
|
+
opencodeDirectory = void 0;
|
|
1951
|
+
/**
|
|
1952
|
+
* Cache of opencode `sessionId → parentID` (its parent session, or `null` when
|
|
1953
|
+
* the session is a root with no parent). Sub-agents spawned via the `task` tool
|
|
1954
|
+
* run in CHILD sessions whose `parentID` chains up to the Evident-created
|
|
1955
|
+
* (watched) session; we resolve this once per session so a child-session
|
|
1956
|
+
* question/permission can be attributed to the watched session's subtree
|
|
1957
|
+
* (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
|
|
1958
|
+
* entry = not yet resolved; `null` = resolved root (stop walking).
|
|
1959
|
+
*/
|
|
1960
|
+
sessionParents = /* @__PURE__ */ new Map();
|
|
1440
1961
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
1441
1962
|
draining = false;
|
|
1963
|
+
/**
|
|
1964
|
+
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
1965
|
+
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
1966
|
+
* is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
|
|
1967
|
+
* drain that entered before `stop()` still registers its watcher).
|
|
1968
|
+
*/
|
|
1969
|
+
activeDrain = null;
|
|
1970
|
+
/**
|
|
1971
|
+
* Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
|
|
1972
|
+
* dispatches NEW work (it returns 0 immediately) — but the per-session watcher
|
|
1973
|
+
* loops already running keep going so in-flight turns can finish and deliver
|
|
1974
|
+
* their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
|
|
1975
|
+
* and stops opencode.
|
|
1976
|
+
*/
|
|
1977
|
+
stopped = false;
|
|
1442
1978
|
constructor(config2) {
|
|
1443
1979
|
this.agentId = config2.agentId;
|
|
1444
1980
|
this.port = config2.port;
|
|
@@ -1450,6 +1986,10 @@ var ChannelDriver = class {
|
|
|
1450
1986
|
});
|
|
1451
1987
|
this.fetchImpl = config2.fetchImpl ?? fetch;
|
|
1452
1988
|
this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
1989
|
+
this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
1990
|
+
this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
1991
|
+
this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
1992
|
+
this.now = config2.now ?? (() => Date.now());
|
|
1453
1993
|
}
|
|
1454
1994
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
1455
1995
|
get opencodeBase() {
|
|
@@ -1459,112 +1999,1452 @@ var ChannelDriver = class {
|
|
|
1459
1999
|
// Public API
|
|
1460
2000
|
// -------------------------------------------------------------------------
|
|
1461
2001
|
/**
|
|
1462
|
-
* Drain all pending channel conversations once: poll →
|
|
2002
|
+
* Drain all pending channel conversations once: poll → dispatch → register.
|
|
1463
2003
|
* Called on tunnel `connected` (WI-CHAN-4) and on each poll tick by `run.ts`.
|
|
1464
2004
|
* Re-entrant calls while a drain is in flight are skipped (return 0).
|
|
1465
2005
|
*
|
|
1466
|
-
* @returns the number of messages
|
|
2006
|
+
* @returns the number of messages NEWLY dispatched to opencode's native queue.
|
|
1467
2007
|
*/
|
|
1468
2008
|
async drainPending() {
|
|
2009
|
+
if (this.stopped) return 0;
|
|
1469
2010
|
if (this.draining) return 0;
|
|
1470
2011
|
this.draining = true;
|
|
1471
|
-
|
|
2012
|
+
const run2 = this.runDrain();
|
|
2013
|
+
this.activeDrain = run2.then(
|
|
2014
|
+
() => {
|
|
2015
|
+
this.activeDrain = null;
|
|
2016
|
+
},
|
|
2017
|
+
() => {
|
|
2018
|
+
this.activeDrain = null;
|
|
2019
|
+
}
|
|
2020
|
+
);
|
|
2021
|
+
return run2;
|
|
2022
|
+
}
|
|
2023
|
+
async runDrain() {
|
|
2024
|
+
let dispatched = 0;
|
|
1472
2025
|
try {
|
|
1473
2026
|
const conversations = await this.getPendingConversations();
|
|
2027
|
+
if (conversations.length > 0) {
|
|
2028
|
+
const total = conversations.reduce((sum, c) => sum + (c.pending_message_count ?? 0), 0);
|
|
2029
|
+
this.log({
|
|
2030
|
+
level: "info",
|
|
2031
|
+
message: `Found ${total} pending message(s) across ${conversations.length} conversation(s) \u2014 draining`
|
|
2032
|
+
});
|
|
2033
|
+
}
|
|
1474
2034
|
for (const conv of conversations) {
|
|
1475
|
-
|
|
2035
|
+
if (this.stopped) break;
|
|
2036
|
+
dispatched += await this.processConversation(conv);
|
|
1476
2037
|
}
|
|
2038
|
+
await this.readoptProcessing();
|
|
1477
2039
|
} finally {
|
|
1478
2040
|
this.draining = false;
|
|
1479
2041
|
}
|
|
1480
|
-
return
|
|
2042
|
+
return dispatched;
|
|
2043
|
+
}
|
|
2044
|
+
/**
|
|
2045
|
+
* True while any per-session watcher has a non-empty in-flight dispatched set
|
|
2046
|
+
* (Task 3.7). `run.ts` treats this as NON-idle so `--idle-timeout` cannot exit
|
|
2047
|
+
* the process while a dispatched message is still queued/running — which would
|
|
2048
|
+
* kill the turn and orphan its reply.
|
|
2049
|
+
*/
|
|
2050
|
+
hasInFlightWatchers() {
|
|
2051
|
+
for (const watcher of this.watchers.values()) {
|
|
2052
|
+
if (watcher.inFlight.size > 0) return true;
|
|
2053
|
+
}
|
|
2054
|
+
return false;
|
|
2055
|
+
}
|
|
2056
|
+
/**
|
|
2057
|
+
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2058
|
+
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
2059
|
+
* `watchers` entry whose `inFlight` set is non-empty — the same predicate
|
|
2060
|
+
* `hasInFlightWatchers()` uses, lifted to return the ids.
|
|
2061
|
+
*
|
|
2062
|
+
* Deliberately does NOT include `this.sessions` (the permanent, never-pruned
|
|
2063
|
+
* conversation→session cache). Protecting every bound-but-idle session there
|
|
2064
|
+
* would shield nearly every session and defeat cleanup — AND it is unnecessary:
|
|
2065
|
+
* `ensureSession` is self-healing (it recreates a session whose id no longer
|
|
2066
|
+
* exists), so deleting an idle bound session is harmless — the conversation's
|
|
2067
|
+
* next turn transparently rebinds a fresh one. The only thing worth protecting
|
|
2068
|
+
* is a session with a turn ACTIVELY in flight right now: tearing that down
|
|
2069
|
+
* mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
|
|
2070
|
+
*/
|
|
2071
|
+
protectedSessionIds() {
|
|
2072
|
+
const ids = /* @__PURE__ */ new Set();
|
|
2073
|
+
for (const [sessionId, watcher] of this.watchers) {
|
|
2074
|
+
if (watcher.inFlight.size > 0) ids.add(sessionId);
|
|
2075
|
+
}
|
|
2076
|
+
return ids;
|
|
2077
|
+
}
|
|
2078
|
+
/**
|
|
2079
|
+
* Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
|
|
2080
|
+
* this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
|
|
2081
|
+
* — but the watcher loops already tracking in-flight turns keep running, so a
|
|
2082
|
+
* turn that has finished (or is about to) still fires `markDone` and delivers
|
|
2083
|
+
* its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
|
|
2084
|
+
*/
|
|
2085
|
+
stop() {
|
|
2086
|
+
this.stopped = true;
|
|
2087
|
+
}
|
|
2088
|
+
/**
|
|
2089
|
+
* Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
|
|
2090
|
+
* graceful shutdown, so a turn whose reply is ready — or completes within the
|
|
2091
|
+
* window — is delivered before the process exits, instead of being cut off and
|
|
2092
|
+
* left for the ADR-0046 restart-recovery path.
|
|
2093
|
+
*
|
|
2094
|
+
* Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
|
|
2095
|
+
* far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
|
|
2096
|
+
* window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
|
|
2097
|
+
* set empties OR the timeout elapses. Anything still in flight at the timeout is
|
|
2098
|
+
* safe to abandon — it stays `processing` server-side and is re-adopted on the
|
|
2099
|
+
* next runner start (ADR-0046).
|
|
2100
|
+
*
|
|
2101
|
+
* @returns true if all in-flight work settled within the window; false if the
|
|
2102
|
+
* timeout elapsed with work still in flight.
|
|
2103
|
+
*/
|
|
2104
|
+
async waitForInFlight(timeoutMs) {
|
|
2105
|
+
const deadline = this.now() + timeoutMs;
|
|
2106
|
+
const step = Math.min(this.pausedPollIntervalMs, 250);
|
|
2107
|
+
if (this.activeDrain) {
|
|
2108
|
+
let drainSettled = false;
|
|
2109
|
+
void this.activeDrain.then(() => {
|
|
2110
|
+
drainSettled = true;
|
|
2111
|
+
});
|
|
2112
|
+
while (!drainSettled) {
|
|
2113
|
+
if (this.now() >= deadline) return false;
|
|
2114
|
+
await this.sleep(step);
|
|
2115
|
+
}
|
|
2116
|
+
}
|
|
2117
|
+
while (this.hasInFlightWatchers()) {
|
|
2118
|
+
if (this.now() >= deadline) return false;
|
|
2119
|
+
await this.sleep(step);
|
|
2120
|
+
}
|
|
2121
|
+
return true;
|
|
2122
|
+
}
|
|
2123
|
+
/**
|
|
2124
|
+
* Await all outstanding per-session watchers (WI-3).
|
|
2125
|
+
*
|
|
2126
|
+
* In production the watcher loops are deliberately started-not-awaited so the
|
|
2127
|
+
* drain loop never blocks on them and process exit is not held up (the cron
|
|
2128
|
+
* recovers any abandoned ones). This helper exists primarily for deterministic
|
|
2129
|
+
* tests that need to observe a watcher's effect (the `processing`/`done` PATCH
|
|
2130
|
+
* or its giving up) after a non-blocking `drainPending`. Watcher loops never
|
|
2131
|
+
* reject, so this resolves.
|
|
2132
|
+
*/
|
|
2133
|
+
async flushPausedWatchers() {
|
|
2134
|
+
while (true) {
|
|
2135
|
+
const loops = [...this.watchers.values()].map((w) => w.loop).filter((l) => l != null);
|
|
2136
|
+
if (loops.length === 0) return;
|
|
2137
|
+
await Promise.all(loops);
|
|
2138
|
+
const stillLive = [...this.watchers.values()].some((w) => w.loop != null);
|
|
2139
|
+
if (!stillLive) return;
|
|
2140
|
+
}
|
|
1481
2141
|
}
|
|
1482
2142
|
// -------------------------------------------------------------------------
|
|
1483
|
-
// Conversation processing
|
|
2143
|
+
// Conversation processing (WI-3 — async dispatch)
|
|
1484
2144
|
// -------------------------------------------------------------------------
|
|
2145
|
+
/**
|
|
2146
|
+
* Dispatch each pending message for a conversation to opencode's native queue
|
|
2147
|
+
* via `prompt_async` (Task 3.2) and register it with the conversation's
|
|
2148
|
+
* per-session watcher. Does NOT block on the turn and does NOT call
|
|
2149
|
+
* `markProcessing` here — that fires from the watcher on running-start.
|
|
2150
|
+
*
|
|
2151
|
+
* @returns the count of messages NEWLY dispatched (not already in-flight).
|
|
2152
|
+
*/
|
|
1485
2153
|
async processConversation(conv) {
|
|
1486
2154
|
const sessionId = await this.ensureSession(conv);
|
|
1487
2155
|
const messages = await this.getPendingMessages(conv.id);
|
|
1488
|
-
let
|
|
2156
|
+
let dispatched = 0;
|
|
2157
|
+
let skippedAlreadyDispatched = 0;
|
|
1489
2158
|
for (const message of messages) {
|
|
1490
|
-
|
|
1491
|
-
if (
|
|
1492
|
-
|
|
1493
|
-
level: "info",
|
|
1494
|
-
message: `Message ${message.id.slice(0, 8)} already claimed \u2014 skipping`,
|
|
1495
|
-
conversation_id: conv.id,
|
|
1496
|
-
message_id: message.id
|
|
1497
|
-
});
|
|
2159
|
+
if (this.stopped) break;
|
|
2160
|
+
if (this.dispatched.has(message.id)) {
|
|
2161
|
+
skippedAlreadyDispatched += 1;
|
|
1498
2162
|
continue;
|
|
1499
2163
|
}
|
|
2164
|
+
const options = {
|
|
2165
|
+
agent: message.opencode_agent ?? void 0,
|
|
2166
|
+
model: message.opencode_model ?? void 0
|
|
2167
|
+
};
|
|
2168
|
+
let opencodeMessageId;
|
|
1500
2169
|
try {
|
|
1501
|
-
await sendMessageToOpenCode(
|
|
1502
|
-
this.port,
|
|
1503
|
-
sessionId,
|
|
1504
|
-
message.content,
|
|
1505
|
-
{
|
|
1506
|
-
agent: message.opencode_agent ?? void 0,
|
|
1507
|
-
model: message.opencode_model ?? void 0
|
|
1508
|
-
},
|
|
1509
|
-
{
|
|
1510
|
-
onQuestion: (question) => this.reportInteraction(conv.id, "question", question),
|
|
1511
|
-
onPermission: (permission) => this.reportInteraction(conv.id, "permission", permission)
|
|
1512
|
-
}
|
|
1513
|
-
);
|
|
1514
|
-
await this.confirmCompletion(sessionId);
|
|
1515
|
-
await this.markDone(conv.id, message.id, sessionId);
|
|
1516
|
-
processed += 1;
|
|
1517
2170
|
this.log({
|
|
1518
2171
|
level: "info",
|
|
1519
|
-
message: `
|
|
2172
|
+
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
1520
2173
|
conversation_id: conv.id,
|
|
1521
2174
|
message_id: message.id
|
|
1522
2175
|
});
|
|
2176
|
+
opencodeMessageId = await this.dispatchLocked(
|
|
2177
|
+
sessionId,
|
|
2178
|
+
() => sendPromptAsync(this.port, sessionId, message.content, options)
|
|
2179
|
+
);
|
|
1523
2180
|
} catch (err) {
|
|
1524
2181
|
if (err instanceof ChannelAuthError) throw err;
|
|
2182
|
+
this.dispatched.delete(message.id);
|
|
2183
|
+
if (await sessionExists(this.port, sessionId) === false) {
|
|
2184
|
+
this.sessions.delete(conv.id);
|
|
2185
|
+
this.log({
|
|
2186
|
+
level: "info",
|
|
2187
|
+
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
2188
|
+
conversation_id: conv.id,
|
|
2189
|
+
message_id: message.id
|
|
2190
|
+
});
|
|
2191
|
+
break;
|
|
2192
|
+
}
|
|
1525
2193
|
await this.markFailed(conv.id, message.id).catch(() => {
|
|
1526
2194
|
});
|
|
1527
2195
|
this.log({
|
|
1528
2196
|
level: "error",
|
|
1529
|
-
message: `Message ${message.id.slice(0, 8)} failed: ${err instanceof Error ? err.message : String(err)}`,
|
|
2197
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
|
|
1530
2198
|
conversation_id: conv.id,
|
|
1531
2199
|
message_id: message.id
|
|
1532
2200
|
});
|
|
2201
|
+
continue;
|
|
1533
2202
|
}
|
|
1534
|
-
|
|
1535
|
-
|
|
2203
|
+
if (opencodeMessageId === null) {
|
|
2204
|
+
this.log({
|
|
2205
|
+
level: "error",
|
|
2206
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
2207
|
+
conversation_id: conv.id,
|
|
2208
|
+
message_id: message.id
|
|
2209
|
+
});
|
|
2210
|
+
continue;
|
|
2211
|
+
}
|
|
2212
|
+
this.dispatched.add(message.id);
|
|
2213
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
2214
|
+
dispatched += 1;
|
|
2215
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
2216
|
+
}
|
|
2217
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
2218
|
+
this.log({
|
|
2219
|
+
level: "error",
|
|
2220
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
2221
|
+
conversation_id: conv.id
|
|
2222
|
+
});
|
|
2223
|
+
}
|
|
2224
|
+
this.ensureWatcherRunning(sessionId);
|
|
2225
|
+
return dispatched;
|
|
2226
|
+
}
|
|
2227
|
+
async ensureSession(conv) {
|
|
2228
|
+
const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
|
|
2229
|
+
if (bound) {
|
|
2230
|
+
const exists = await sessionExists(this.port, bound);
|
|
2231
|
+
if (exists === false) {
|
|
2232
|
+
this.log({
|
|
2233
|
+
level: "info",
|
|
2234
|
+
message: `OpenCode session ${bound} for conversation ${conv.id.slice(0, 8)} no longer exists (deleted or DB reset) \u2014 creating a fresh session and rebinding.`,
|
|
2235
|
+
conversation_id: conv.id
|
|
2236
|
+
});
|
|
2237
|
+
this.sessions.delete(conv.id);
|
|
2238
|
+
return this.createAndBindSession(conv.id);
|
|
2239
|
+
}
|
|
2240
|
+
this.sessions.set(conv.id, bound);
|
|
2241
|
+
return bound;
|
|
2242
|
+
}
|
|
2243
|
+
return this.createAndBindSession(conv.id);
|
|
2244
|
+
}
|
|
2245
|
+
/**
|
|
2246
|
+
* Create a fresh OpenCode session for a conversation, cache the binding, and
|
|
2247
|
+
* best-effort persist it server-side. Shared by the first-ever bind and the
|
|
2248
|
+
* self-heal recreate path in `ensureSession`.
|
|
2249
|
+
*/
|
|
2250
|
+
async createAndBindSession(conversationId) {
|
|
2251
|
+
const directory = await this.resolveOpenCodeDirectory();
|
|
2252
|
+
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
2253
|
+
this.sessions.set(conversationId, sessionId);
|
|
2254
|
+
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
2255
|
+
});
|
|
2256
|
+
return sessionId;
|
|
2257
|
+
}
|
|
2258
|
+
/**
|
|
2259
|
+
* Lazily resolve (and cache) opencode's root directory via `GET /path`.
|
|
2260
|
+
* Resolved once per driver: `undefined` until first lookup, then the directory
|
|
2261
|
+
* string or `null` if unavailable (we don't keep retrying a missing `/path`).
|
|
2262
|
+
*/
|
|
2263
|
+
async resolveOpenCodeDirectory() {
|
|
2264
|
+
if (this.opencodeDirectory !== void 0) return this.opencodeDirectory;
|
|
2265
|
+
this.opencodeDirectory = await getOpenCodeDirectory(this.port);
|
|
2266
|
+
if (!this.opencodeDirectory) {
|
|
2267
|
+
this.log({
|
|
2268
|
+
level: "info",
|
|
2269
|
+
message: "Could not determine opencode directory (GET /path) \u2014 new sessions may not appear in opencode web"
|
|
2270
|
+
});
|
|
2271
|
+
}
|
|
2272
|
+
return this.opencodeDirectory;
|
|
2273
|
+
}
|
|
2274
|
+
// -------------------------------------------------------------------------
|
|
2275
|
+
// Per-session watcher (WI-3)
|
|
2276
|
+
// -------------------------------------------------------------------------
|
|
2277
|
+
/**
|
|
2278
|
+
* Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
|
|
2279
|
+
* opencode session (Task 2.1a), so two dispatches into the SAME session can
|
|
2280
|
+
* never interleave and mis-correlate their read-backs. Distinct sessions run
|
|
2281
|
+
* concurrently. The chained tail intentionally ignores the prior result/error
|
|
2282
|
+
* (each dispatch reports its own outcome to its caller).
|
|
2283
|
+
*/
|
|
2284
|
+
dispatchLocked(sessionId, fn) {
|
|
2285
|
+
const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
|
|
2286
|
+
const run2 = prior.then(fn, fn);
|
|
2287
|
+
this.sessionDispatchLocks.set(
|
|
2288
|
+
sessionId,
|
|
2289
|
+
run2.then(
|
|
2290
|
+
() => void 0,
|
|
2291
|
+
() => void 0
|
|
2292
|
+
)
|
|
2293
|
+
);
|
|
2294
|
+
return run2;
|
|
2295
|
+
}
|
|
2296
|
+
/** Register a freshly-dispatched message with its session's watcher state. */
|
|
2297
|
+
registerInFlight(conv, sessionId, message, opencodeMessageId) {
|
|
2298
|
+
let watcher = this.watchers.get(sessionId);
|
|
2299
|
+
if (!watcher) {
|
|
2300
|
+
watcher = {
|
|
2301
|
+
conv,
|
|
2302
|
+
inFlight: /* @__PURE__ */ new Map(),
|
|
2303
|
+
loop: null,
|
|
2304
|
+
reportedQuestions: /* @__PURE__ */ new Set(),
|
|
2305
|
+
reportedPermissions: /* @__PURE__ */ new Set(),
|
|
2306
|
+
lastGoodPollAt: this.now(),
|
|
2307
|
+
hadUsablePoll: false
|
|
2308
|
+
};
|
|
2309
|
+
this.watchers.set(sessionId, watcher);
|
|
2310
|
+
}
|
|
2311
|
+
const now = this.now();
|
|
2312
|
+
watcher.inFlight.set(message.id, {
|
|
2313
|
+
evidentMessageId: message.id,
|
|
2314
|
+
opencodeMessageId,
|
|
2315
|
+
message,
|
|
2316
|
+
dispatchedAt: now,
|
|
2317
|
+
processingAnchorMs: now,
|
|
2318
|
+
deadline: now + this.pausedMaxWaitMs,
|
|
2319
|
+
started: false,
|
|
2320
|
+
done: false,
|
|
2321
|
+
stuckReported: false,
|
|
2322
|
+
lastAliveAt: 0,
|
|
2323
|
+
aliveInFlight: false,
|
|
2324
|
+
awaitingHumanLatched: false,
|
|
2325
|
+
pausedOnQuestion: false,
|
|
2326
|
+
pausedOnPermission: false,
|
|
2327
|
+
pausedClearConfirmed: false,
|
|
2328
|
+
pausedInFlight: false,
|
|
2329
|
+
deliveryDeadlineAnchored: false
|
|
2330
|
+
});
|
|
2331
|
+
}
|
|
2332
|
+
/**
|
|
2333
|
+
* Register a RE-ADOPTED `processing` message with its session watcher
|
|
2334
|
+
* (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
|
|
2335
|
+
* `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
|
|
2336
|
+
* `now`, so the paused/queued/unreachable cases settle on the same wall-clock a
|
|
2337
|
+
* fresh dispatch would (10 min after `processed_at`, not 10 min from now).
|
|
2338
|
+
*
|
|
2339
|
+
* This re-attaches into the SAME watcher, so the ADR-0047 progressing-vs-paused
|
|
2340
|
+
* give-up (`serviceInFlightMessage`) applies unchanged: a re-adopted turn
|
|
2341
|
+
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
2342
|
+
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
2343
|
+
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
2344
|
+
* handed to the cron. The old "the `deadline` must settle before the ~15-min
|
|
2345
|
+
* cron or they double-drive" reasoning is superseded: liveness now settles the
|
|
2346
|
+
* actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
|
|
2347
|
+
* (only the appear-guard uses it).
|
|
2348
|
+
*
|
|
2349
|
+
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
2350
|
+
* `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
|
|
2351
|
+
* fresh-run path these differ (a fresh opencode id under the same server row).
|
|
2352
|
+
*
|
|
2353
|
+
* `started` is set true so the watcher does NOT re-`markProcessing` a row the
|
|
2354
|
+
* server already flipped to `processing`; the running/done transitions still
|
|
2355
|
+
* fire from the watcher's normal branches.
|
|
2356
|
+
*/
|
|
2357
|
+
registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
|
|
2358
|
+
let watcher = this.watchers.get(sessionId);
|
|
2359
|
+
if (!watcher) {
|
|
2360
|
+
watcher = {
|
|
2361
|
+
conv,
|
|
2362
|
+
inFlight: /* @__PURE__ */ new Map(),
|
|
2363
|
+
loop: null,
|
|
2364
|
+
reportedQuestions: /* @__PURE__ */ new Set(),
|
|
2365
|
+
reportedPermissions: /* @__PURE__ */ new Set(),
|
|
2366
|
+
lastGoodPollAt: this.now(),
|
|
2367
|
+
hadUsablePoll: false
|
|
2368
|
+
};
|
|
2369
|
+
this.watchers.set(sessionId, watcher);
|
|
2370
|
+
}
|
|
2371
|
+
watcher.inFlight.set(message.id, {
|
|
2372
|
+
evidentMessageId: message.id,
|
|
2373
|
+
opencodeMessageId,
|
|
2374
|
+
message,
|
|
2375
|
+
dispatchedAt: this.now(),
|
|
2376
|
+
// Anchor the absolute-age ceiling to the SERVER-SIDE `processed_at` (the same
|
|
2377
|
+
// value seeding `deadline`), NOT `dispatchedAt` — so a re-adopted zombie's age
|
|
2378
|
+
// reflects the real turn duration and the ceiling fires on the ORIGINAL turn.
|
|
2379
|
+
processingAnchorMs: processedAtMs,
|
|
2380
|
+
deadline: processedAtMs + this.pausedMaxWaitMs,
|
|
2381
|
+
// The server row is ALREADY `processing`; do not re-fire markProcessing.
|
|
2382
|
+
started: true,
|
|
2383
|
+
done: false,
|
|
2384
|
+
// Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
|
|
2385
|
+
// AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
|
|
2386
|
+
// on `state === 'queued'` (turn produced no reply), not on `started`, so a
|
|
2387
|
+
// re-adopted row left wedged in `queued` still emits the signal once
|
|
2388
|
+
// (#210/#220 observability).
|
|
2389
|
+
stuckReported: false,
|
|
2390
|
+
// Task 5.2: a re-adopted actively-running row re-attaches into the SAME
|
|
2391
|
+
// watcher and so hits the SAME actively-running heartbeat branch in
|
|
2392
|
+
// `serviceInFlightMessage` as a fresh dispatch — monitoring observes "runner
|
|
2393
|
+
// re-adopted and is confirming this row alive" via that `alive` heartbeat,
|
|
2394
|
+
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2395
|
+
lastAliveAt: 0,
|
|
2396
|
+
aliveInFlight: false,
|
|
2397
|
+
awaitingHumanLatched: false,
|
|
2398
|
+
pausedOnQuestion: false,
|
|
2399
|
+
pausedOnPermission: false,
|
|
2400
|
+
pausedClearConfirmed: false,
|
|
2401
|
+
pausedInFlight: false,
|
|
2402
|
+
deliveryDeadlineAnchored: false
|
|
2403
|
+
});
|
|
2404
|
+
}
|
|
2405
|
+
/**
|
|
2406
|
+
* Start (but do NOT await) the per-session watcher loop if it has in-flight
|
|
2407
|
+
* work and is not already running. Single-flight per session. The loop is
|
|
2408
|
+
* tracked on the watcher and cleared when it settles; it never rejects (fully
|
|
2409
|
+
* guarded), so a failed poll/callback can never crash the run loop — the cron
|
|
2410
|
+
* stays as the safety net.
|
|
2411
|
+
*/
|
|
2412
|
+
ensureWatcherRunning(sessionId) {
|
|
2413
|
+
const watcher = this.watchers.get(sessionId);
|
|
2414
|
+
if (!watcher) return;
|
|
2415
|
+
if (watcher.loop) return;
|
|
2416
|
+
if (watcher.inFlight.size === 0) {
|
|
2417
|
+
this.watchers.delete(sessionId);
|
|
2418
|
+
return;
|
|
2419
|
+
}
|
|
2420
|
+
const loop = this.runWatcherLoop(sessionId, watcher).finally(() => {
|
|
2421
|
+
watcher.loop = null;
|
|
2422
|
+
if (watcher.inFlight.size === 0) {
|
|
2423
|
+
this.watchers.delete(sessionId);
|
|
2424
|
+
}
|
|
2425
|
+
});
|
|
2426
|
+
watcher.loop = loop;
|
|
2427
|
+
}
|
|
2428
|
+
/**
|
|
2429
|
+
* The per-session polling loop (WI-3). Once per tick it:
|
|
2430
|
+
* 1. polls `GET /session/:id/message` once and, per in-flight message,
|
|
2431
|
+
* computes `messageRunState` and fires markProcessing (queued→running) /
|
|
2432
|
+
* markDone (done) exactly once per transition;
|
|
2433
|
+
* 2. applies the idle-path re-dispatch guard (a dispatched message that never
|
|
2434
|
+
* APPEARS → re-dispatch — D1 obligation 2);
|
|
2435
|
+
* 3. polls `/question` + `/permission` (scoped to the session) and surfaces
|
|
2436
|
+
* NEW ones via `reportInteraction`, carrying the PAUSED message's own
|
|
2437
|
+
* `source_message_id`;
|
|
2438
|
+
* 4. drops messages that completed or timed out from the in-flight set.
|
|
2439
|
+
* Exits when the in-flight set empties. Never throws.
|
|
2440
|
+
*/
|
|
2441
|
+
async runWatcherLoop(sessionId, watcher) {
|
|
2442
|
+
try {
|
|
2443
|
+
while (watcher.inFlight.size > 0) {
|
|
2444
|
+
await this.sleep(this.pausedPollIntervalMs);
|
|
2445
|
+
let messages = null;
|
|
2446
|
+
try {
|
|
2447
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
2448
|
+
if (res.ok) {
|
|
2449
|
+
const body = await res.json();
|
|
2450
|
+
messages = Array.isArray(body) ? body : null;
|
|
2451
|
+
}
|
|
2452
|
+
} catch {
|
|
2453
|
+
}
|
|
2454
|
+
if (messages != null && messages.length > 0) {
|
|
2455
|
+
watcher.lastGoodPollAt = this.now();
|
|
2456
|
+
watcher.hadUsablePoll = true;
|
|
2457
|
+
} else {
|
|
2458
|
+
const emptyButReachable = messages != null;
|
|
2459
|
+
const graceApplies = !emptyButReachable || watcher.hadUsablePoll;
|
|
2460
|
+
if (graceApplies && this.now() - watcher.lastGoodPollAt < POLL_MISS_GRACE_MS) {
|
|
2461
|
+
continue;
|
|
2462
|
+
}
|
|
2463
|
+
}
|
|
2464
|
+
const { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk } = await this.pollInteractions(sessionId, watcher, messages);
|
|
2465
|
+
for (const inFlight of [...watcher.inFlight.values()]) {
|
|
2466
|
+
await this.serviceInFlightMessage(
|
|
2467
|
+
sessionId,
|
|
2468
|
+
watcher,
|
|
2469
|
+
inFlight,
|
|
2470
|
+
messages,
|
|
2471
|
+
openQuestions,
|
|
2472
|
+
openPermissions,
|
|
2473
|
+
questionsPolledOk,
|
|
2474
|
+
permissionsPolledOk
|
|
2475
|
+
);
|
|
2476
|
+
}
|
|
2477
|
+
}
|
|
2478
|
+
} catch (err) {
|
|
2479
|
+
if (err instanceof ChannelAuthError) {
|
|
2480
|
+
this.log({
|
|
2481
|
+
level: "error",
|
|
2482
|
+
message: `Session watcher aborted on auth failure for session ${sessionId.slice(0, 8)} \u2014 clearing in-flight state for re-drive after re-auth: ${err.message}`,
|
|
2483
|
+
conversation_id: watcher.conv.id
|
|
2484
|
+
});
|
|
2485
|
+
for (const evidentMessageId of [...watcher.inFlight.keys()]) {
|
|
2486
|
+
this.readopted.delete(evidentMessageId);
|
|
2487
|
+
this.removeInFlight(watcher, evidentMessageId);
|
|
2488
|
+
}
|
|
2489
|
+
return;
|
|
2490
|
+
}
|
|
2491
|
+
this.log({
|
|
2492
|
+
level: "error",
|
|
2493
|
+
message: `Session watcher failed for session ${sessionId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
|
|
2494
|
+
conversation_id: watcher.conv.id
|
|
2495
|
+
});
|
|
2496
|
+
}
|
|
2497
|
+
}
|
|
2498
|
+
/**
|
|
2499
|
+
* On FIRST observing a terminal (done/failed) state, ensure the delivery
|
|
2500
|
+
* (markDone/markFailed) transient-retry path has a real window. A long
|
|
2501
|
+
* ACTIVELY-running turn is kept past its original `deadline`, so by completion
|
|
2502
|
+
* `now >= deadline` already holds and the retry bound below would fire on the
|
|
2503
|
+
* first transient PATCH failure — dropping the message before its reply lands
|
|
2504
|
+
* (Bugbot "Stale deadline aborts long-turn delivery"). Re-anchor once (latched)
|
|
2505
|
+
* to a fresh `pausedMaxWaitMs` window; only extend if the current deadline is at
|
|
2506
|
+
* or past now, so a still-ample window is left untouched.
|
|
2507
|
+
*/
|
|
2508
|
+
anchorDeliveryDeadline(inFlight) {
|
|
2509
|
+
if (inFlight.deliveryDeadlineAnchored) return;
|
|
2510
|
+
inFlight.deliveryDeadlineAnchored = true;
|
|
2511
|
+
if (this.now() >= inFlight.deadline) {
|
|
2512
|
+
inFlight.deadline = this.now() + this.pausedMaxWaitMs;
|
|
2513
|
+
}
|
|
2514
|
+
}
|
|
2515
|
+
/**
|
|
2516
|
+
* Drive ONE in-flight message's lifecycle from the tick's message snapshot.
|
|
2517
|
+
* Fires markProcessing on queued→running and markDone on done (each once),
|
|
2518
|
+
* applies the idle-path re-dispatch guard, and removes the message from the
|
|
2519
|
+
* in-flight set on completion or timeout.
|
|
2520
|
+
*/
|
|
2521
|
+
async serviceInFlightMessage(sessionId, watcher, inFlight, messages, openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk) {
|
|
2522
|
+
const conv = watcher.conv;
|
|
2523
|
+
const state = messageRunState(messages, inFlight.opencodeMessageId);
|
|
2524
|
+
const id = inFlight.evidentMessageId;
|
|
2525
|
+
if (openQuestions.has(id)) inFlight.pausedOnQuestion = true;
|
|
2526
|
+
else if (questionsPolledOk) inFlight.pausedOnQuestion = false;
|
|
2527
|
+
if (openPermissions.has(id)) inFlight.pausedOnPermission = true;
|
|
2528
|
+
else if (permissionsPolledOk) inFlight.pausedOnPermission = false;
|
|
2529
|
+
const observedOpen = openQuestions.has(id) || openPermissions.has(id);
|
|
2530
|
+
const latchedPaused = inFlight.pausedOnQuestion || inFlight.pausedOnPermission;
|
|
2531
|
+
const awaitingHuman = observedOpen || latchedPaused;
|
|
2532
|
+
if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
|
|
2533
|
+
let claimed;
|
|
2534
|
+
try {
|
|
2535
|
+
claimed = await this.markProcessing(
|
|
2536
|
+
conv.id,
|
|
2537
|
+
inFlight.evidentMessageId,
|
|
2538
|
+
sessionId,
|
|
2539
|
+
inFlight.opencodeMessageId
|
|
2540
|
+
);
|
|
2541
|
+
} catch (err) {
|
|
2542
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2543
|
+
this.log({
|
|
2544
|
+
level: "error",
|
|
2545
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2546
|
+
conversation_id: conv.id,
|
|
2547
|
+
message_id: inFlight.evidentMessageId
|
|
2548
|
+
});
|
|
2549
|
+
return;
|
|
2550
|
+
}
|
|
2551
|
+
inFlight.started = true;
|
|
2552
|
+
if (!claimed) {
|
|
2553
|
+
this.log({
|
|
2554
|
+
level: "info",
|
|
2555
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
|
|
2556
|
+
conversation_id: conv.id,
|
|
2557
|
+
message_id: inFlight.evidentMessageId
|
|
2558
|
+
});
|
|
2559
|
+
}
|
|
2560
|
+
}
|
|
2561
|
+
if (state === "done") {
|
|
2562
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
2563
|
+
if (!inFlight.done) {
|
|
2564
|
+
this.log({
|
|
2565
|
+
level: "info",
|
|
2566
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
2567
|
+
conversation_id: conv.id,
|
|
2568
|
+
message_id: inFlight.evidentMessageId
|
|
2569
|
+
});
|
|
2570
|
+
try {
|
|
2571
|
+
await this.markDone(
|
|
2572
|
+
conv.id,
|
|
2573
|
+
inFlight.evidentMessageId,
|
|
2574
|
+
sessionId,
|
|
2575
|
+
inFlight.opencodeMessageId
|
|
2576
|
+
);
|
|
2577
|
+
} catch (err) {
|
|
2578
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2579
|
+
if (err instanceof ChannelTerminalError) {
|
|
2580
|
+
this.log({
|
|
2581
|
+
level: "error",
|
|
2582
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2583
|
+
conversation_id: conv.id,
|
|
2584
|
+
message_id: inFlight.evidentMessageId
|
|
2585
|
+
});
|
|
2586
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2587
|
+
return;
|
|
2588
|
+
}
|
|
2589
|
+
if (this.now() >= inFlight.deadline) {
|
|
2590
|
+
this.log({
|
|
2591
|
+
level: "error",
|
|
2592
|
+
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)}`,
|
|
2593
|
+
conversation_id: conv.id,
|
|
2594
|
+
message_id: inFlight.evidentMessageId
|
|
2595
|
+
});
|
|
2596
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2597
|
+
return;
|
|
2598
|
+
}
|
|
2599
|
+
this.log({
|
|
2600
|
+
level: "error",
|
|
2601
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2602
|
+
conversation_id: conv.id,
|
|
2603
|
+
message_id: inFlight.evidentMessageId
|
|
2604
|
+
});
|
|
2605
|
+
return;
|
|
2606
|
+
}
|
|
2607
|
+
inFlight.done = true;
|
|
2608
|
+
}
|
|
2609
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2610
|
+
return;
|
|
2611
|
+
}
|
|
2612
|
+
if (state === "failed") {
|
|
2613
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
2614
|
+
if (!inFlight.done) {
|
|
2615
|
+
const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
|
|
2616
|
+
this.log({
|
|
2617
|
+
level: "error",
|
|
2618
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
2619
|
+
conversation_id: conv.id,
|
|
2620
|
+
message_id: inFlight.evidentMessageId
|
|
2621
|
+
});
|
|
2622
|
+
try {
|
|
2623
|
+
await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
|
|
2624
|
+
} catch (err) {
|
|
2625
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2626
|
+
if (err instanceof ChannelTerminalError) {
|
|
2627
|
+
this.log({
|
|
2628
|
+
level: "error",
|
|
2629
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2630
|
+
conversation_id: conv.id,
|
|
2631
|
+
message_id: inFlight.evidentMessageId
|
|
2632
|
+
});
|
|
2633
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2634
|
+
return;
|
|
2635
|
+
}
|
|
2636
|
+
if (this.now() >= inFlight.deadline) {
|
|
2637
|
+
this.log({
|
|
2638
|
+
level: "error",
|
|
2639
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
2640
|
+
conversation_id: conv.id,
|
|
2641
|
+
message_id: inFlight.evidentMessageId
|
|
2642
|
+
});
|
|
2643
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2644
|
+
return;
|
|
2645
|
+
}
|
|
2646
|
+
this.log({
|
|
2647
|
+
level: "error",
|
|
2648
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2649
|
+
conversation_id: conv.id,
|
|
2650
|
+
message_id: inFlight.evidentMessageId
|
|
2651
|
+
});
|
|
2652
|
+
return;
|
|
2653
|
+
}
|
|
2654
|
+
inFlight.done = true;
|
|
2655
|
+
}
|
|
2656
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2657
|
+
return;
|
|
2658
|
+
}
|
|
2659
|
+
const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
|
|
2660
|
+
const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
|
|
2661
|
+
if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
|
|
2662
|
+
inFlight.stuckReported = true;
|
|
2663
|
+
void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
|
|
2664
|
+
stuck_for_ms: this.now() - inFlight.dispatchedAt
|
|
2665
|
+
});
|
|
2666
|
+
}
|
|
2667
|
+
const activelyRunning = state === "running" && !awaitingHuman;
|
|
2668
|
+
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
2669
|
+
this.log({
|
|
2670
|
+
level: "error",
|
|
2671
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} exceeded the absolute processing ceiling (${Math.round((this.now() - inFlight.processingAnchorMs) / 6e4)}min, session ${sessionId}) while still actively running \u2014 releasing so the cron can reclaim it`,
|
|
2672
|
+
conversation_id: conv.id,
|
|
2673
|
+
message_id: inFlight.evidentMessageId
|
|
2674
|
+
});
|
|
2675
|
+
void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
|
|
2676
|
+
watched_for_ms: this.now() - inFlight.processingAnchorMs
|
|
2677
|
+
});
|
|
2678
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2679
|
+
return;
|
|
2680
|
+
}
|
|
2681
|
+
if (activelyRunning && !inFlight.awaitingHumanLatched && !inFlight.aliveInFlight && this.now() - inFlight.lastAliveAt >= HEARTBEAT_MS) {
|
|
2682
|
+
inFlight.aliveInFlight = true;
|
|
2683
|
+
void this.postSignal(conv.id, inFlight.evidentMessageId, "alive").then((ok) => {
|
|
2684
|
+
inFlight.aliveInFlight = false;
|
|
2685
|
+
if (ok) inFlight.lastAliveAt = this.now();
|
|
2686
|
+
});
|
|
2687
|
+
}
|
|
2688
|
+
if (awaitingHuman) {
|
|
2689
|
+
if (!inFlight.awaitingHumanLatched) {
|
|
2690
|
+
inFlight.deadline = this.now() + this.pausedMaxWaitMs;
|
|
2691
|
+
inFlight.awaitingHumanLatched = true;
|
|
2692
|
+
}
|
|
2693
|
+
if (!inFlight.pausedClearConfirmed && !inFlight.pausedInFlight) {
|
|
2694
|
+
inFlight.pausedInFlight = true;
|
|
2695
|
+
void this.postSignal(conv.id, inFlight.evidentMessageId, "paused").then((ok) => {
|
|
2696
|
+
inFlight.pausedInFlight = false;
|
|
2697
|
+
if (ok && inFlight.awaitingHumanLatched) inFlight.pausedClearConfirmed = true;
|
|
2698
|
+
});
|
|
2699
|
+
}
|
|
2700
|
+
} else if (inFlight.awaitingHumanLatched) {
|
|
2701
|
+
inFlight.awaitingHumanLatched = false;
|
|
2702
|
+
inFlight.pausedOnQuestion = false;
|
|
2703
|
+
inFlight.pausedOnPermission = false;
|
|
2704
|
+
inFlight.pausedClearConfirmed = false;
|
|
2705
|
+
}
|
|
2706
|
+
const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
|
|
2707
|
+
const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
|
|
2708
|
+
(sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
|
|
2709
|
+
);
|
|
2710
|
+
const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
|
|
2711
|
+
if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
|
|
2712
|
+
this.log({
|
|
2713
|
+
level: "info",
|
|
2714
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
|
|
2715
|
+
conversation_id: conv.id,
|
|
2716
|
+
message_id: inFlight.evidentMessageId
|
|
2717
|
+
});
|
|
2718
|
+
void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
|
|
2719
|
+
watched_for_ms: this.now() - inFlight.dispatchedAt
|
|
2720
|
+
});
|
|
2721
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2722
|
+
}
|
|
2723
|
+
}
|
|
2724
|
+
// -------------------------------------------------------------------------
|
|
2725
|
+
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
2726
|
+
// -------------------------------------------------------------------------
|
|
2727
|
+
/**
|
|
2728
|
+
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
2729
|
+
*
|
|
2730
|
+
* The pending drain only re-drives `pending` rows; a message already flipped to
|
|
2731
|
+
* `processing` before the runner died is watched by nobody until the 15-min
|
|
2732
|
+
* cron resets it. Here we fetch those rows, and per row resolve its correlated
|
|
2733
|
+
* reply against opencode's OWN session store — completing, re-attaching, or
|
|
2734
|
+
* (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
|
|
2735
|
+
* is idempotent per message (Invariant 2): a row a watcher already tracks is
|
|
2736
|
+
* skipped in `readoptOne` — one driver, no double-drive.
|
|
2737
|
+
*
|
|
2738
|
+
* Only `ChannelAuthError` propagates (to `drainPending`, like the pending
|
|
2739
|
+
* path); every other early return LOGS a reason with context — no silent drop.
|
|
2740
|
+
*/
|
|
2741
|
+
async readoptProcessing() {
|
|
2742
|
+
const rows = await this.getProcessingMessages();
|
|
2743
|
+
if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0 || this.readoptPollUnresolvedSignalled.size > 0) {
|
|
2744
|
+
const stillProcessing = new Set(rows.map((r) => r.id));
|
|
2745
|
+
for (const id of [
|
|
2746
|
+
...this.dontRedispatch,
|
|
2747
|
+
...this.doneUndeliverable,
|
|
2748
|
+
...this.readoptPollUnresolvedSignalled
|
|
2749
|
+
]) {
|
|
2750
|
+
if (!stillProcessing.has(id)) {
|
|
2751
|
+
const cleared = this.dontRedispatch.delete(id);
|
|
2752
|
+
const clearedUndeliverable = this.doneUndeliverable.delete(id);
|
|
2753
|
+
this.readoptPollUnresolvedSignalled.delete(id);
|
|
2754
|
+
if (cleared || clearedUndeliverable) {
|
|
2755
|
+
this.log({
|
|
2756
|
+
level: "info",
|
|
2757
|
+
message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
|
|
2758
|
+
message_id: id
|
|
2759
|
+
});
|
|
2760
|
+
}
|
|
2761
|
+
}
|
|
2762
|
+
}
|
|
2763
|
+
}
|
|
2764
|
+
if (rows.length === 0) return;
|
|
2765
|
+
const bySession = /* @__PURE__ */ new Map();
|
|
2766
|
+
for (const row of rows) {
|
|
2767
|
+
if (!row.opencode_session_id) {
|
|
2768
|
+
this.log({
|
|
2769
|
+
level: "error",
|
|
2770
|
+
message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
|
|
2771
|
+
conversation_id: row.conversation_id,
|
|
2772
|
+
message_id: row.id
|
|
2773
|
+
});
|
|
2774
|
+
continue;
|
|
2775
|
+
}
|
|
2776
|
+
const list = bySession.get(row.opencode_session_id) ?? [];
|
|
2777
|
+
list.push(row);
|
|
2778
|
+
bySession.set(row.opencode_session_id, list);
|
|
2779
|
+
}
|
|
2780
|
+
for (const [sessionId, sessionRows] of bySession) {
|
|
2781
|
+
let messages;
|
|
2782
|
+
try {
|
|
2783
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
2784
|
+
if (!res.ok) {
|
|
2785
|
+
this.log({
|
|
2786
|
+
level: "error",
|
|
2787
|
+
message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
|
|
2788
|
+
});
|
|
2789
|
+
continue;
|
|
2790
|
+
}
|
|
2791
|
+
const body = await res.json();
|
|
2792
|
+
if (!Array.isArray(body)) {
|
|
2793
|
+
this.log({
|
|
2794
|
+
level: "error",
|
|
2795
|
+
message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
|
|
2796
|
+
});
|
|
2797
|
+
continue;
|
|
2798
|
+
}
|
|
2799
|
+
messages = body;
|
|
2800
|
+
} catch (err) {
|
|
2801
|
+
this.log({
|
|
2802
|
+
level: "error",
|
|
2803
|
+
message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
|
|
2804
|
+
});
|
|
2805
|
+
continue;
|
|
2806
|
+
}
|
|
2807
|
+
const anyUntracked = sessionRows.some((row) => !this.isTracked(sessionId, row.id));
|
|
2808
|
+
const sessionOngoing = anyUntracked ? await isSessionOngoing(this.port, sessionId) : null;
|
|
2809
|
+
for (const row of sessionRows) {
|
|
2810
|
+
await this.readoptOne(sessionId, row, messages, sessionOngoing);
|
|
2811
|
+
}
|
|
2812
|
+
}
|
|
2813
|
+
}
|
|
2814
|
+
/**
|
|
2815
|
+
* Re-adopt ONE `processing` row against the tick's session message snapshot
|
|
2816
|
+
* (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
|
|
2817
|
+
*
|
|
2818
|
+
* Branches on `messageRunState(messages, row.opencode_message_id)` — the
|
|
2819
|
+
* opencode-assigned user-message id persisted on the first `processing` PATCH
|
|
2820
|
+
* (#218). A row with a NULL stored id (dispatched but the read-back never landed
|
|
2821
|
+
* before the restart) has no id to correlate → treated as an orphan and
|
|
2822
|
+
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
2823
|
+
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
2824
|
+
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
2825
|
+
* errored turn is reported failed on restart, NOT re-dispatched;
|
|
2826
|
+
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
2827
|
+
* tracking the stored id so the reply correlates by it;
|
|
2828
|
+
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
2829
|
+
*
|
|
2830
|
+
* Only `ChannelAuthError` propagates.
|
|
2831
|
+
*/
|
|
2832
|
+
async readoptOne(sessionId, row, messages, sessionOngoing) {
|
|
2833
|
+
if (this.isTracked(sessionId, row.id)) {
|
|
2834
|
+
this.log({
|
|
2835
|
+
level: "info",
|
|
2836
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
|
|
2837
|
+
conversation_id: row.conversation_id,
|
|
2838
|
+
message_id: row.id
|
|
2839
|
+
});
|
|
2840
|
+
return;
|
|
2841
|
+
}
|
|
2842
|
+
const ocId = row.opencode_message_id;
|
|
2843
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
2844
|
+
if (state === "done") {
|
|
2845
|
+
if (this.doneUndeliverable.has(row.id)) {
|
|
2846
|
+
this.log({
|
|
2847
|
+
level: "info",
|
|
2848
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
|
|
2849
|
+
conversation_id: row.conversation_id,
|
|
2850
|
+
message_id: row.id
|
|
2851
|
+
});
|
|
2852
|
+
return;
|
|
2853
|
+
}
|
|
2854
|
+
this.log({
|
|
2855
|
+
level: "info",
|
|
2856
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
|
|
2857
|
+
conversation_id: row.conversation_id,
|
|
2858
|
+
message_id: row.id
|
|
2859
|
+
});
|
|
2860
|
+
try {
|
|
2861
|
+
await this.markDone(row.conversation_id, row.id, sessionId, ocId);
|
|
2862
|
+
} catch (err) {
|
|
2863
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2864
|
+
if (err instanceof ChannelTerminalError) {
|
|
2865
|
+
this.doneUndeliverable.add(row.id);
|
|
2866
|
+
this.log({
|
|
2867
|
+
level: "error",
|
|
2868
|
+
message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
|
|
2869
|
+
conversation_id: row.conversation_id,
|
|
2870
|
+
message_id: row.id
|
|
2871
|
+
});
|
|
2872
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
|
|
2873
|
+
return;
|
|
2874
|
+
}
|
|
2875
|
+
this.log({
|
|
2876
|
+
level: "error",
|
|
2877
|
+
message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
2878
|
+
conversation_id: row.conversation_id,
|
|
2879
|
+
message_id: row.id
|
|
2880
|
+
});
|
|
2881
|
+
return;
|
|
2882
|
+
}
|
|
2883
|
+
this.dontRedispatch.delete(row.id);
|
|
2884
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
2885
|
+
return;
|
|
2886
|
+
}
|
|
2887
|
+
if (state === "failed") {
|
|
2888
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
2889
|
+
this.log({
|
|
2890
|
+
level: "error",
|
|
2891
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
2892
|
+
conversation_id: row.conversation_id,
|
|
2893
|
+
message_id: row.id
|
|
2894
|
+
});
|
|
2895
|
+
try {
|
|
2896
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2);
|
|
2897
|
+
} catch (err) {
|
|
2898
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2899
|
+
if (err instanceof ChannelTerminalError) {
|
|
2900
|
+
this.doneUndeliverable.add(row.id);
|
|
2901
|
+
this.log({
|
|
2902
|
+
level: "error",
|
|
2903
|
+
message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
|
|
2904
|
+
conversation_id: row.conversation_id,
|
|
2905
|
+
message_id: row.id
|
|
2906
|
+
});
|
|
2907
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
|
|
2908
|
+
return;
|
|
2909
|
+
}
|
|
2910
|
+
this.log({
|
|
2911
|
+
level: "error",
|
|
2912
|
+
message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
2913
|
+
conversation_id: row.conversation_id,
|
|
2914
|
+
message_id: row.id
|
|
2915
|
+
});
|
|
2916
|
+
return;
|
|
2917
|
+
}
|
|
2918
|
+
this.dontRedispatch.delete(row.id);
|
|
2919
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_failed");
|
|
2920
|
+
return;
|
|
2921
|
+
}
|
|
2922
|
+
if (this.dontRedispatch.has(row.id)) {
|
|
2923
|
+
this.log({
|
|
2924
|
+
level: "info",
|
|
2925
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
|
|
2926
|
+
conversation_id: row.conversation_id,
|
|
2927
|
+
message_id: row.id
|
|
2928
|
+
});
|
|
2929
|
+
return;
|
|
2930
|
+
}
|
|
2931
|
+
let statusReadableOngoing = null;
|
|
2932
|
+
if (state === "running" && ocId) {
|
|
2933
|
+
const reply = findLastAssistantReplyFor(messages, ocId);
|
|
2934
|
+
const shape = this.replyCompletionShape(reply);
|
|
2935
|
+
const ongoing = sessionOngoing;
|
|
2936
|
+
statusReadableOngoing = ongoing;
|
|
2937
|
+
if (ongoing === false) {
|
|
2938
|
+
this.log({
|
|
2939
|
+
level: "info",
|
|
2940
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status (absent/idle) \u2014 re-dispatching from scratch (status-gated recovery)`,
|
|
2941
|
+
conversation_id: row.conversation_id,
|
|
2942
|
+
message_id: row.id
|
|
2943
|
+
});
|
|
2944
|
+
await this.forceReadoptRun(sessionId, row);
|
|
2945
|
+
return;
|
|
2946
|
+
}
|
|
2947
|
+
if (ongoing === true) {
|
|
2948
|
+
this.log({
|
|
2949
|
+
level: "info",
|
|
2950
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} and session ${sessionId.slice(0, 8)} is ongoing per GET /session/status (busy/retry) \u2014 re-attaching watcher (no re-dispatch)`,
|
|
2951
|
+
conversation_id: row.conversation_id,
|
|
2952
|
+
message_id: row.id
|
|
2953
|
+
});
|
|
2954
|
+
} else {
|
|
2955
|
+
if (shape === "b1") {
|
|
2956
|
+
this.log({
|
|
2957
|
+
level: "info",
|
|
2958
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} running/b1 but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 NOT latching a b1 row on a transient status blip; leaving it un-tracked to re-evaluate on the next drain`,
|
|
2959
|
+
conversation_id: row.conversation_id,
|
|
2960
|
+
message_id: row.id
|
|
2961
|
+
});
|
|
2962
|
+
if (!this.readoptPollUnresolvedSignalled.has(row.id)) {
|
|
2963
|
+
this.readoptPollUnresolvedSignalled.add(row.id);
|
|
2964
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_poll_unresolved");
|
|
2965
|
+
}
|
|
2966
|
+
return;
|
|
2967
|
+
}
|
|
2968
|
+
this.log({
|
|
2969
|
+
level: "info",
|
|
2970
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 falling back to the #253 preamble + descendant cross-check`,
|
|
2971
|
+
conversation_id: row.conversation_id,
|
|
2972
|
+
message_id: row.id
|
|
2973
|
+
});
|
|
2974
|
+
}
|
|
2975
|
+
}
|
|
2976
|
+
if (statusReadableOngoing === null && state === "running" && ocId && isPreamblePinnedRunning(messages, ocId)) {
|
|
2977
|
+
const descendantAlive = await this.isAnyDescendantSessionAlive(sessionId);
|
|
2978
|
+
if (descendantAlive === true) {
|
|
2979
|
+
this.log({
|
|
2980
|
+
level: "info",
|
|
2981
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned running (root ${sessionId.slice(0, 8)}) but a live descendant sub-agent session was found \u2014 treating as still running, re-attaching watcher (no re-dispatch)`,
|
|
2982
|
+
conversation_id: row.conversation_id,
|
|
2983
|
+
message_id: row.id
|
|
2984
|
+
});
|
|
2985
|
+
} else {
|
|
2986
|
+
this.log({
|
|
2987
|
+
level: "info",
|
|
2988
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned on recovery (root ${sessionId.slice(0, 8)}), no live descendant runner \u2014 re-dispatching from scratch${descendantAlive === null ? " (descendant liveness indeterminate; a restart guarantees no live runner, so this does NOT block the re-dispatch)" : ""}`,
|
|
2989
|
+
conversation_id: row.conversation_id,
|
|
2990
|
+
message_id: row.id
|
|
2991
|
+
});
|
|
2992
|
+
await this.forceReadoptRun(sessionId, row);
|
|
2993
|
+
return;
|
|
2994
|
+
}
|
|
2995
|
+
}
|
|
2996
|
+
if ((state === "running" || state === "queued") && ocId) {
|
|
2997
|
+
const conv = this.convForRow(sessionId, row);
|
|
2998
|
+
const message = this.queuedMessageForRow(row);
|
|
2999
|
+
this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
|
|
3000
|
+
this.dispatched.add(row.id);
|
|
3001
|
+
this.readopted.add(row.id);
|
|
3002
|
+
this.ensureWatcherRunning(sessionId);
|
|
3003
|
+
this.log({
|
|
3004
|
+
level: "info",
|
|
3005
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
|
|
3006
|
+
conversation_id: row.conversation_id,
|
|
3007
|
+
message_id: row.id
|
|
3008
|
+
});
|
|
3009
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_reattached");
|
|
3010
|
+
return;
|
|
3011
|
+
}
|
|
3012
|
+
await this.forceReadoptRun(sessionId, row);
|
|
3013
|
+
}
|
|
3014
|
+
/**
|
|
3015
|
+
* Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
|
|
3016
|
+
*
|
|
3017
|
+
* #218/WI-5: the row's user message is absent (never kept, or a null stored id),
|
|
3018
|
+
* so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
|
|
3019
|
+
* read it back, and register the watcher under the assigned id so the reply
|
|
3020
|
+
* correlates server-side.
|
|
3021
|
+
*
|
|
3022
|
+
* ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
|
|
3023
|
+
* id). Without a guard, if this dispatches on tick N but the read-back+persist
|
|
3024
|
+
* hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
|
|
3025
|
+
* tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
|
|
3026
|
+
* latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
|
|
3027
|
+
* short-circuit while the row is latched; clear it on a successful dispatch (the
|
|
3028
|
+
* row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
|
|
3029
|
+
* re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
|
|
3030
|
+
* may retry exactly once more).
|
|
3031
|
+
*
|
|
3032
|
+
* `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
|
|
3033
|
+
* `processed_at` (Invariant 1).
|
|
3034
|
+
*/
|
|
3035
|
+
async forceReadoptRun(sessionId, row) {
|
|
3036
|
+
if (this.stopped) {
|
|
3037
|
+
this.log({
|
|
3038
|
+
level: "info",
|
|
3039
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but the runner is stopping \u2014 not starting a fresh turn; leaving for restart recovery`,
|
|
3040
|
+
conversation_id: row.conversation_id,
|
|
3041
|
+
message_id: row.id
|
|
3042
|
+
});
|
|
3043
|
+
return;
|
|
3044
|
+
}
|
|
3045
|
+
if (this.awaitingReadopt.has(row.id)) {
|
|
3046
|
+
this.log({
|
|
3047
|
+
level: "info",
|
|
3048
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
|
|
3049
|
+
conversation_id: row.conversation_id,
|
|
3050
|
+
message_id: row.id
|
|
3051
|
+
});
|
|
3052
|
+
return;
|
|
3053
|
+
}
|
|
3054
|
+
if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
|
|
3055
|
+
this.dontRedispatch.add(row.id);
|
|
3056
|
+
this.log({
|
|
3057
|
+
level: "info",
|
|
3058
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but its re-adopt window has already elapsed \u2014 not dispatching an unwatchable turn; parking until it leaves processing (cron will reset it)`,
|
|
3059
|
+
conversation_id: row.conversation_id,
|
|
3060
|
+
message_id: row.id
|
|
3061
|
+
});
|
|
3062
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_window_elapsed");
|
|
3063
|
+
return;
|
|
3064
|
+
}
|
|
3065
|
+
const options = {
|
|
3066
|
+
agent: row.opencode_agent ?? void 0,
|
|
3067
|
+
model: row.opencode_model ?? void 0
|
|
3068
|
+
};
|
|
3069
|
+
this.log({
|
|
3070
|
+
level: "info",
|
|
3071
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
|
|
3072
|
+
conversation_id: row.conversation_id,
|
|
3073
|
+
message_id: row.id
|
|
3074
|
+
});
|
|
3075
|
+
this.awaitingReadopt.add(row.id);
|
|
3076
|
+
let ocId;
|
|
3077
|
+
try {
|
|
3078
|
+
ocId = await this.dispatchLocked(
|
|
3079
|
+
sessionId,
|
|
3080
|
+
() => sendPromptAsync(this.port, sessionId, row.content, options)
|
|
3081
|
+
);
|
|
3082
|
+
} catch (err) {
|
|
3083
|
+
this.awaitingReadopt.delete(row.id);
|
|
3084
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3085
|
+
this.log({
|
|
3086
|
+
level: "error",
|
|
3087
|
+
message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3088
|
+
conversation_id: row.conversation_id,
|
|
3089
|
+
message_id: row.id
|
|
3090
|
+
});
|
|
3091
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
3092
|
+
return;
|
|
3093
|
+
}
|
|
3094
|
+
if (ocId === null) {
|
|
3095
|
+
this.awaitingReadopt.delete(row.id);
|
|
3096
|
+
this.log({
|
|
3097
|
+
level: "error",
|
|
3098
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
|
|
3099
|
+
conversation_id: row.conversation_id,
|
|
3100
|
+
message_id: row.id
|
|
3101
|
+
});
|
|
3102
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
3103
|
+
return;
|
|
3104
|
+
}
|
|
3105
|
+
const conv = this.convForRow(sessionId, row);
|
|
3106
|
+
const message = this.queuedMessageForRow(row);
|
|
3107
|
+
this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
|
|
3108
|
+
this.dispatched.add(row.id);
|
|
3109
|
+
this.readopted.add(row.id);
|
|
3110
|
+
this.awaitingReadopt.delete(row.id);
|
|
3111
|
+
this.ensureWatcherRunning(sessionId);
|
|
3112
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_redispatched");
|
|
1536
3113
|
}
|
|
1537
|
-
|
|
1538
|
-
|
|
1539
|
-
|
|
1540
|
-
|
|
1541
|
-
|
|
1542
|
-
|
|
1543
|
-
|
|
1544
|
-
const
|
|
1545
|
-
|
|
1546
|
-
|
|
3114
|
+
/**
|
|
3115
|
+
* True if `evidentMessageId` is already being driven — either in the
|
|
3116
|
+
* authoritative `dispatched` set or a live watcher's in-flight set for this
|
|
3117
|
+
* session (Invariant 2, WI-5). Either signal means a watcher owns the row.
|
|
3118
|
+
*/
|
|
3119
|
+
isTracked(sessionId, evidentMessageId) {
|
|
3120
|
+
if (this.dispatched.has(evidentMessageId)) return true;
|
|
3121
|
+
const watcher = this.watchers.get(sessionId);
|
|
3122
|
+
return watcher?.inFlight.has(evidentMessageId) ?? false;
|
|
3123
|
+
}
|
|
3124
|
+
/**
|
|
3125
|
+
* Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
|
|
3126
|
+
* deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
|
|
3127
|
+
* for `processing` rows, but if it is somehow null/unparseable fall back to
|
|
3128
|
+
* `now` (defensive) AND log — a fallback means the anchor is weaker than
|
|
3129
|
+
* intended, which is worth surfacing.
|
|
3130
|
+
*/
|
|
3131
|
+
processedAtMs(row) {
|
|
3132
|
+
const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
|
|
3133
|
+
if (!Number.isNaN(parsed)) return parsed;
|
|
3134
|
+
this.log({
|
|
3135
|
+
level: "error",
|
|
3136
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
|
|
3137
|
+
conversation_id: row.conversation_id,
|
|
3138
|
+
message_id: row.id
|
|
1547
3139
|
});
|
|
1548
|
-
return
|
|
3140
|
+
return this.now();
|
|
3141
|
+
}
|
|
3142
|
+
/** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
|
|
3143
|
+
convForRow(sessionId, row) {
|
|
3144
|
+
return {
|
|
3145
|
+
id: row.conversation_id,
|
|
3146
|
+
agent_id: this.agentId,
|
|
3147
|
+
opencode_session_id: sessionId,
|
|
3148
|
+
pending_message_count: 0,
|
|
3149
|
+
oldest_pending_at: row.processed_at
|
|
3150
|
+
};
|
|
3151
|
+
}
|
|
3152
|
+
/** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
|
|
3153
|
+
queuedMessageForRow(row) {
|
|
3154
|
+
return {
|
|
3155
|
+
id: row.id,
|
|
3156
|
+
content: row.content,
|
|
3157
|
+
status: "processing",
|
|
3158
|
+
opencode_agent: row.opencode_agent,
|
|
3159
|
+
opencode_model: row.opencode_model,
|
|
3160
|
+
source_message_id: row.source_message_id,
|
|
3161
|
+
slack_user_id: row.slack_user_id
|
|
3162
|
+
};
|
|
3163
|
+
}
|
|
3164
|
+
/**
|
|
3165
|
+
* Remove a message from the in-flight set AND the authoritative dispatched
|
|
3166
|
+
* set. Once the in-flight set empties, the watcher loop's `while` guard exits
|
|
3167
|
+
* and its `.finally` removes the session entry from `this.watchers`.
|
|
3168
|
+
*
|
|
3169
|
+
* Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
|
|
3170
|
+
* (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
|
|
3171
|
+
* cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
|
|
3172
|
+
* re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
|
|
3173
|
+
* re-adopted message that completed (`done`) needs no marker — it's leaving
|
|
3174
|
+
* `processing`. This suppresses only re-dispatch: if its reply later completes,
|
|
3175
|
+
* the done branch still delivers it (Bugbot #202).
|
|
3176
|
+
*/
|
|
3177
|
+
removeInFlight(watcher, evidentMessageId) {
|
|
3178
|
+
const inFlight = watcher.inFlight.get(evidentMessageId);
|
|
3179
|
+
if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
|
|
3180
|
+
this.dontRedispatch.add(evidentMessageId);
|
|
3181
|
+
this.log({
|
|
3182
|
+
level: "info",
|
|
3183
|
+
message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
|
|
3184
|
+
conversation_id: watcher.conv.id,
|
|
3185
|
+
message_id: evidentMessageId
|
|
3186
|
+
});
|
|
3187
|
+
}
|
|
3188
|
+
watcher.inFlight.delete(evidentMessageId);
|
|
3189
|
+
this.dispatched.delete(evidentMessageId);
|
|
3190
|
+
}
|
|
3191
|
+
/**
|
|
3192
|
+
* Poll `/question` + `/permission` (scoped to the session) and surface NEW ones
|
|
3193
|
+
* via `reportInteraction` (Task 3.5), carrying the PAUSED message's own
|
|
3194
|
+
* `source_message_id` so the server @mentions the correct person under
|
|
3195
|
+
* concurrency. Dedups by interaction id across ticks (reused per-session sets).
|
|
3196
|
+
*
|
|
3197
|
+
* The interaction is attributed to the in-flight message it paused on. opencode
|
|
3198
|
+
* stamps a `messageID` on a permission (and `tool.messageID` on a question) =
|
|
3199
|
+
* the assistant message id, whose `parentID` is the user message id — but the
|
|
3200
|
+
* simplest robust attribution here is: the single in-flight message that is
|
|
3201
|
+
* RUNNING (not done) is the one that paused. With one running message that is
|
|
3202
|
+
* unambiguous; with several we prefer an explicit messageID match, else the
|
|
3203
|
+
* oldest running message.
|
|
3204
|
+
*
|
|
3205
|
+
* Returns the set of in-flight Evident message ids that are paused awaiting a
|
|
3206
|
+
* human — an outstanding (still-open) question/permission is attributed to them.
|
|
3207
|
+
* `serviceInFlightMessage` uses this to keep an actively-running turn watched
|
|
3208
|
+
* forever (ADR-0047) while still bounding a turn merely blocked on a person who
|
|
3209
|
+
* may never answer. Attribution here covers ALL open interactions, not just
|
|
3210
|
+
* NEW (un-deduped) ones — a question stays "awaiting a human" until answered,
|
|
3211
|
+
* even after it was already surfaced to the channel.
|
|
3212
|
+
*/
|
|
3213
|
+
async pollInteractions(sessionId, watcher, messages) {
|
|
3214
|
+
const openQuestions = /* @__PURE__ */ new Set();
|
|
3215
|
+
const openPermissions = /* @__PURE__ */ new Set();
|
|
3216
|
+
let questionsPolledOk = true;
|
|
3217
|
+
let permissionsPolledOk = true;
|
|
3218
|
+
let questions = [];
|
|
3219
|
+
try {
|
|
3220
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/question`);
|
|
3221
|
+
if (res.ok) {
|
|
3222
|
+
const body = await res.json();
|
|
3223
|
+
if (Array.isArray(body)) {
|
|
3224
|
+
questions = body;
|
|
3225
|
+
} else {
|
|
3226
|
+
questionsPolledOk = false;
|
|
3227
|
+
}
|
|
3228
|
+
} else {
|
|
3229
|
+
questionsPolledOk = false;
|
|
3230
|
+
}
|
|
3231
|
+
} catch {
|
|
3232
|
+
questionsPolledOk = false;
|
|
3233
|
+
}
|
|
3234
|
+
for (const q of questions) {
|
|
3235
|
+
if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
|
|
3236
|
+
const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
|
|
3237
|
+
if (paused) openQuestions.add(paused.evidentMessageId);
|
|
3238
|
+
if (watcher.reportedQuestions.has(q.id)) continue;
|
|
3239
|
+
const reported = await this.reportInteraction(
|
|
3240
|
+
watcher.conv.id,
|
|
3241
|
+
"question",
|
|
3242
|
+
q,
|
|
3243
|
+
paused?.message.source_message_id ?? void 0
|
|
3244
|
+
);
|
|
3245
|
+
if (reported) watcher.reportedQuestions.add(q.id);
|
|
3246
|
+
}
|
|
3247
|
+
let permissions = [];
|
|
3248
|
+
try {
|
|
3249
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
|
|
3250
|
+
if (res.ok) {
|
|
3251
|
+
const body = await res.json();
|
|
3252
|
+
if (Array.isArray(body)) {
|
|
3253
|
+
permissions = body;
|
|
3254
|
+
} else {
|
|
3255
|
+
permissionsPolledOk = false;
|
|
3256
|
+
}
|
|
3257
|
+
} else {
|
|
3258
|
+
permissionsPolledOk = false;
|
|
3259
|
+
}
|
|
3260
|
+
} catch {
|
|
3261
|
+
permissionsPolledOk = false;
|
|
3262
|
+
}
|
|
3263
|
+
for (const p of permissions) {
|
|
3264
|
+
if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
|
|
3265
|
+
const paused = this.attributeInteraction(watcher, p.messageID, messages);
|
|
3266
|
+
if (paused) openPermissions.add(paused.evidentMessageId);
|
|
3267
|
+
if (watcher.reportedPermissions.has(p.id)) continue;
|
|
3268
|
+
const reported = await this.reportInteraction(
|
|
3269
|
+
watcher.conv.id,
|
|
3270
|
+
"permission",
|
|
3271
|
+
p,
|
|
3272
|
+
paused?.message.source_message_id ?? void 0
|
|
3273
|
+
);
|
|
3274
|
+
if (reported) watcher.reportedPermissions.add(p.id);
|
|
3275
|
+
}
|
|
3276
|
+
return { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk };
|
|
3277
|
+
}
|
|
3278
|
+
/**
|
|
3279
|
+
* True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
|
|
3280
|
+
* i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
|
|
3281
|
+
* watched root. Sub-agents spawned via the `task` tool run in child sessions,
|
|
3282
|
+
* so their questions/permissions live under a different `sessionID` that must
|
|
3283
|
+
* still be attributed to the root conversation the watcher owns.
|
|
3284
|
+
*
|
|
3285
|
+
* Parents are cached in `sessionParents` so we walk each session at most once;
|
|
3286
|
+
* a bounded depth cap guards against a cycle or a pathological chain, and any
|
|
3287
|
+
* fetch failure is treated as "not a descendant" (best-effort — the interaction
|
|
3288
|
+
* simply isn't surfaced this tick and is retried next tick once resolvable).
|
|
3289
|
+
*/
|
|
3290
|
+
async sessionBelongsTo(sessionId, rootSessionId) {
|
|
3291
|
+
let current = sessionId;
|
|
3292
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
3293
|
+
if (current === rootSessionId) return true;
|
|
3294
|
+
const parent = await this.resolveSessionParent(current);
|
|
3295
|
+
if (parent === null || parent === void 0) return false;
|
|
3296
|
+
current = parent;
|
|
3297
|
+
}
|
|
3298
|
+
return false;
|
|
1549
3299
|
}
|
|
1550
3300
|
/**
|
|
1551
|
-
*
|
|
1552
|
-
*
|
|
1553
|
-
*
|
|
3301
|
+
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3302
|
+
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
3303
|
+
* unreachable / the session can't be read (so the caller stops walking without
|
|
3304
|
+
* caching a wrong answer — the next tick retries).
|
|
1554
3305
|
*/
|
|
1555
|
-
async
|
|
3306
|
+
async resolveSessionParent(sessionId) {
|
|
3307
|
+
const cached = this.sessionParents.get(sessionId);
|
|
3308
|
+
if (cached !== void 0) return cached;
|
|
3309
|
+
let parent = void 0;
|
|
1556
3310
|
try {
|
|
1557
3311
|
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
|
|
1558
|
-
if (
|
|
1559
|
-
|
|
1560
|
-
|
|
1561
|
-
this.log({
|
|
1562
|
-
level: "info",
|
|
1563
|
-
message: `Session ${sessionId.slice(0, 8)} not marked completed on reconcile \u2014 delivering anyway`
|
|
1564
|
-
});
|
|
3312
|
+
if (res.ok) {
|
|
3313
|
+
const body = await res.json();
|
|
3314
|
+
parent = body && typeof body.parentID === "string" ? body.parentID : null;
|
|
1565
3315
|
}
|
|
1566
3316
|
} catch {
|
|
3317
|
+
parent = void 0;
|
|
3318
|
+
}
|
|
3319
|
+
if (parent !== void 0) this.sessionParents.set(sessionId, parent);
|
|
3320
|
+
return parent;
|
|
3321
|
+
}
|
|
3322
|
+
/**
|
|
3323
|
+
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3324
|
+
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
3325
|
+
*
|
|
3326
|
+
* The PRIMARY recovery trigger is "preamble-pinned on recovery ⇒ idle" — a
|
|
3327
|
+
* runner restart wipes OpenCode's in-memory `SessionStatus`/`Runner`, so a
|
|
3328
|
+
* completed `finish: "tool-calls"` root reply encountered during re-adoption is
|
|
3329
|
+
* idle by OpenCode's own definition and is re-dispatched. This method exists only
|
|
3330
|
+
* so the WI-3 caller can VETO that re-dispatch in the rare case a descendant is
|
|
3331
|
+
* provably in flight at the exact moment of recovery.
|
|
3332
|
+
*
|
|
3333
|
+
* "Alive" criterion (TIGHTENED): a descendant is alive only when it is PROVABLY,
|
|
3334
|
+
* ACTIVELY generating — its LAST message is an assistant still mid-generation
|
|
3335
|
+
* (`completed == null`, via `isSessionActivelyGenerating`). An
|
|
3336
|
+
* INCOMPLETE-BUT-NOT-GENERATING child — last message a user message, or a
|
|
3337
|
+
* completed `finish: "tool-calls"` step — is NOT alive after a restart (nothing
|
|
3338
|
+
* is generating once the runner is gone), so it does NOT veto. (This is
|
|
3339
|
+
* deliberately NOT `!isTurnComplete`, which also matches those dead-but-non-terminal
|
|
3340
|
+
* shapes and would falsely veto — re-hanging the very turn this path recovers.)
|
|
3341
|
+
*
|
|
3342
|
+
* Return contract (encoded so WI-3 need not re-derive it):
|
|
3343
|
+
* - `true` → a descendant is provably, actively generating (veto re-dispatch).
|
|
3344
|
+
* - `false` → descendants exist but none is actively generating (the restart
|
|
3345
|
+
* case), OR no descendant is found at all.
|
|
3346
|
+
* - `null` → liveness is INDETERMINATE (enumeration via `listSessions` failed).
|
|
3347
|
+
*
|
|
3348
|
+
* ⚠️ `null` (UNKNOWN) MUST NOT be treated as "alive": WI-3 treats `null` the same
|
|
3349
|
+
* as `false` and does NOT veto — a restart guarantees no live runner, so an
|
|
3350
|
+
* indeterminate cross-check almost always means "couldn't reach a child that no
|
|
3351
|
+
* longer exists". The inversion lives in the caller; this method just reports
|
|
3352
|
+
* true/false/null faithfully.
|
|
3353
|
+
*
|
|
3354
|
+
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
3355
|
+
* (already proven by the existing child-session interaction tests, via
|
|
3356
|
+
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
3357
|
+
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
3358
|
+
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
3359
|
+
* `SessionStatus` only.
|
|
3360
|
+
*/
|
|
3361
|
+
async isAnyDescendantSessionAlive(rootSessionId) {
|
|
3362
|
+
const sessions = await listSessions(this.port);
|
|
3363
|
+
if (!sessions) {
|
|
3364
|
+
this.log({
|
|
3365
|
+
level: "error",
|
|
3366
|
+
message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
|
|
3367
|
+
});
|
|
3368
|
+
return null;
|
|
3369
|
+
}
|
|
3370
|
+
for (const candidate of sessions) {
|
|
3371
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
3372
|
+
if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
|
|
3373
|
+
const childMsgs = await getSessionMessages(this.port, candidate.id);
|
|
3374
|
+
if (isSessionActivelyGenerating(childMsgs)) {
|
|
3375
|
+
return true;
|
|
3376
|
+
}
|
|
3377
|
+
}
|
|
3378
|
+
return false;
|
|
3379
|
+
}
|
|
3380
|
+
/**
|
|
3381
|
+
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
3382
|
+
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
3383
|
+
* - `b1` — the reply itself is still in flight (`time.completed == null`) —
|
|
3384
|
+
* the aborted-in-flight production bug after a restart.
|
|
3385
|
+
* - `b2` — a COMPLETED reply pinned running only by `finish === "tool-calls"`
|
|
3386
|
+
* (the sub-agent preamble — #253's shape).
|
|
3387
|
+
* - `other` — any other shape (defensive; a running row is normally b1 or b2).
|
|
3388
|
+
* Reads `info.time.completed` / `info.finish` (tolerating the legacy top-level
|
|
3389
|
+
* shape) directly rather than re-importing the module-private `completedOf`/
|
|
3390
|
+
* `finishOf` — this is a display label only, not a correctness predicate.
|
|
3391
|
+
*/
|
|
3392
|
+
replyCompletionShape(reply) {
|
|
3393
|
+
if (!reply) return "other";
|
|
3394
|
+
const completed = reply.info?.time?.completed ?? reply.time?.completed;
|
|
3395
|
+
if (completed == null) return "b1";
|
|
3396
|
+
const finish = reply.info?.finish ?? reply.finish;
|
|
3397
|
+
return finish === "tool-calls" ? "b2" : "other";
|
|
3398
|
+
}
|
|
3399
|
+
/**
|
|
3400
|
+
* Attribute a surfaced interaction to the in-flight message it paused on (M-1).
|
|
3401
|
+
*
|
|
3402
|
+
* The interaction carries `interactionMessageId` — the ASSISTANT message id
|
|
3403
|
+
* that raised it (a question's `tool.messageID` / a permission's `messageID`).
|
|
3404
|
+
* That assistant message is the reply to ONE of our minted user messages
|
|
3405
|
+
* (correlated by `parentID`, GATE-B). So when we have the tick's message
|
|
3406
|
+
* snapshot, we resolve each running in-flight message's correlated assistant
|
|
3407
|
+
* reply (`findAssistantReplyAfter`) and match its id against
|
|
3408
|
+
* `interactionMessageId` — giving an EXACT attribution even with several
|
|
3409
|
+
* messages in flight concurrently in one session.
|
|
3410
|
+
*
|
|
3411
|
+
* We fall back to the oldest running message ONLY when no exact match is
|
|
3412
|
+
* possible (the id is absent, the snapshot is missing, or the reply has not yet
|
|
3413
|
+
* been correlated). With a single running message either path is exact. Never
|
|
3414
|
+
* throws.
|
|
3415
|
+
*
|
|
3416
|
+
* Attribution must NOT depend on our own `started` PATCH flag: opencode can
|
|
3417
|
+
* START a turn AND raise a question/permission BEFORE our next tick fires
|
|
3418
|
+
* `markProcessing` (which sets `started`). Relying on `started` would leave the
|
|
3419
|
+
* running set empty in that window and let the server fall back to "newest
|
|
3420
|
+
* processing/pending" — possibly @mentioning a FOLLOW-UP author rather than the
|
|
3421
|
+
* person whose active turn actually paused. So we derive "running" from the
|
|
3422
|
+
* tick's `messages` snapshot via `messageRunState` instead.
|
|
3423
|
+
*/
|
|
3424
|
+
attributeInteraction(watcher, interactionMessageId, messages) {
|
|
3425
|
+
const inFlight = [...watcher.inFlight.values()].filter((m) => !m.done);
|
|
3426
|
+
if (inFlight.length === 0) return void 0;
|
|
3427
|
+
if (interactionMessageId && messages) {
|
|
3428
|
+
const exact = inFlight.find((m) => {
|
|
3429
|
+
const reply = findAssistantReplyAfter(messages, m.opencodeMessageId);
|
|
3430
|
+
return reply != null && messageIdOf(reply) === interactionMessageId;
|
|
3431
|
+
});
|
|
3432
|
+
if (exact) return exact;
|
|
3433
|
+
}
|
|
3434
|
+
const byOldest = (a, b) => a.dispatchedAt - b.dispatchedAt;
|
|
3435
|
+
if (messages) {
|
|
3436
|
+
const runningPerSnapshot = inFlight.filter(
|
|
3437
|
+
(m) => messageRunState(messages, m.opencodeMessageId) === "running"
|
|
3438
|
+
);
|
|
3439
|
+
if (runningPerSnapshot.length > 0) {
|
|
3440
|
+
return runningPerSnapshot.sort(byOldest)[0];
|
|
3441
|
+
}
|
|
1567
3442
|
}
|
|
3443
|
+
const startedRunning = inFlight.filter((m) => m.started);
|
|
3444
|
+
if (startedRunning.length > 0) {
|
|
3445
|
+
return startedRunning.sort(byOldest)[0];
|
|
3446
|
+
}
|
|
3447
|
+
return inFlight.sort(byOldest)[0];
|
|
1568
3448
|
}
|
|
1569
3449
|
// -------------------------------------------------------------------------
|
|
1570
3450
|
// Evident API calls (combinedAuth thread routes)
|
|
@@ -1598,49 +3478,191 @@ var ChannelDriver = class {
|
|
|
1598
3478
|
}
|
|
1599
3479
|
return await res.json();
|
|
1600
3480
|
}
|
|
1601
|
-
|
|
3481
|
+
/**
|
|
3482
|
+
* Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
|
|
3483
|
+
* The pending path (`getPendingConversations`/`getPendingMessages`) only
|
|
3484
|
+
* surfaces `pending` rows, so a message already `processing` when the runner
|
|
3485
|
+
* died is invisible to it — this dedicated endpoint returns exactly those rows
|
|
3486
|
+
* with the fields the re-adopt path needs (`processed_at`,
|
|
3487
|
+
* `opencode_session_id`, routing).
|
|
3488
|
+
*
|
|
3489
|
+
* Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
|
|
3490
|
+
* array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
|
|
3491
|
+
* other non-ok so `drainPending`'s try/finally leaves `draining` false and the
|
|
3492
|
+
* next tick retries.
|
|
3493
|
+
*/
|
|
3494
|
+
async getProcessingMessages() {
|
|
3495
|
+
const res = await this.fetchImpl(
|
|
3496
|
+
`${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
|
|
3497
|
+
{ headers: { Authorization: this.getAuthHeader() } }
|
|
3498
|
+
);
|
|
3499
|
+
this.assertAuth(res, "fetching processing messages");
|
|
3500
|
+
if (!res.ok) {
|
|
3501
|
+
throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
|
|
3502
|
+
}
|
|
3503
|
+
const data = await res.json();
|
|
3504
|
+
let messages = data.messages ?? [];
|
|
3505
|
+
if (this.conversationFilter) {
|
|
3506
|
+
messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
|
|
3507
|
+
}
|
|
3508
|
+
return messages;
|
|
3509
|
+
}
|
|
3510
|
+
/**
|
|
3511
|
+
* EXISTING combinedAuth route — now fired by the watcher on queued→running
|
|
3512
|
+
* (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
|
|
3513
|
+
* opencode_session_id}` → `notifyMessageStarted` (hourglass→runner swap +
|
|
3514
|
+
* deep-linked "View in Evident" notice).
|
|
3515
|
+
*
|
|
3516
|
+
* Return/throw contract (consumed by the watcher's swap-to-running guard):
|
|
3517
|
+
* - returns `true` → the server transitioned the row to processing;
|
|
3518
|
+
* - returns `false` → the server gave a DEFINITIVE "already-processing"
|
|
3519
|
+
* answer (a non-retryable, non-auth status — e.g. a
|
|
3520
|
+
* conflict because a duplicate already transitioned it),
|
|
3521
|
+
* so the caller treats it as already-started and does NOT
|
|
3522
|
+
* retry;
|
|
3523
|
+
* - throws `ChannelAuthError` on 401/403 (terminal auth failure);
|
|
3524
|
+
* - throws on a TRANSIENT failure (retryable 5xx/429 status, or a
|
|
3525
|
+
* network-level error from `fetch`) — i.e. NO definitive server response —
|
|
3526
|
+
* so the caller leaves the message un-started and retries the swap on the
|
|
3527
|
+
* next tick.
|
|
3528
|
+
* A single attempt (no internal retry): the watcher's per-tick loop is the
|
|
3529
|
+
* retry vehicle for the swap-to-running.
|
|
3530
|
+
*/
|
|
3531
|
+
async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
|
|
1602
3532
|
const res = await this.fetchImpl(
|
|
1603
3533
|
`${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
1604
3534
|
{
|
|
1605
3535
|
method: "PATCH",
|
|
1606
3536
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
1607
|
-
body: JSON.stringify({
|
|
3537
|
+
body: JSON.stringify({
|
|
3538
|
+
status: "processing",
|
|
3539
|
+
opencode_session_id: sessionId,
|
|
3540
|
+
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
|
|
3541
|
+
})
|
|
1608
3542
|
}
|
|
1609
3543
|
);
|
|
1610
3544
|
this.assertAuth(res, "marking message as processing");
|
|
1611
|
-
|
|
3545
|
+
if (res.ok) return true;
|
|
3546
|
+
if (isRetryableStatus(res.status)) {
|
|
3547
|
+
throw new Error(`marking message as processing: HTTP ${res.status}`);
|
|
3548
|
+
}
|
|
3549
|
+
return false;
|
|
1612
3550
|
}
|
|
1613
3551
|
/**
|
|
1614
|
-
* EXISTING combinedAuth completion route — idempotent
|
|
1615
|
-
*
|
|
3552
|
+
* EXISTING combinedAuth completion route — idempotent (WI-CHAN-2). `PATCH
|
|
3553
|
+
* .../messages/:id {status:'done', opencode_session_id}`. The server's
|
|
1616
3554
|
* `queued_conversation_messages.status`/`processed_at` gate makes a re-call
|
|
1617
|
-
* for an already-`done` message a no-op (no double Slack post).
|
|
3555
|
+
* for an already-`done` message a no-op (no double Slack post). Fired by the
|
|
3556
|
+
* watcher on per-message completion (Task 3.4) — no `confirmCompletion`
|
|
3557
|
+
* round-trip (we already observed completion via the message list).
|
|
3558
|
+
*
|
|
3559
|
+
* SINGLE ATTEMPT (no in-call `callWithRetry` backoff). The per-session watcher
|
|
3560
|
+
* services its in-flight messages SEQUENTIALLY within a tick
|
|
3561
|
+
* (`runWatcherLoop` → `serviceInFlightMessage`), so a long multi-attempt
|
|
3562
|
+
* backoff here would BLOCK sibling messages in the SAME session/tick: while
|
|
3563
|
+
* message A's done PATCH burned its internal retries, message B could not be
|
|
3564
|
+
* swapped to running even though opencode had already started it. Instead this
|
|
3565
|
+
* does ONE PATCH and surfaces the SAME outcome contract the watcher's markDone
|
|
3566
|
+
* handler already relies on, leaning on the per-tick retry across ticks
|
|
3567
|
+
* (bounded by `inFlight.deadline`) rather than an in-call retry:
|
|
3568
|
+
* - resolves (`void`) → the server transitioned the row to done
|
|
3569
|
+
* (or idempotently confirmed already-done);
|
|
3570
|
+
* - throws `ChannelAuthError` → 401/403 (terminal auth failure → loop
|
|
3571
|
+
* cleanup, Finding 1);
|
|
3572
|
+
* - throws `ChannelTerminalError`→ non-retryable, non-auth 4xx (will never
|
|
3573
|
+
* succeed → straight to the cron, Finding 4);
|
|
3574
|
+
* - throws a plain `Error` → TRANSIENT 5xx/429 or a network-level error
|
|
3575
|
+
* (no definitive server response → the
|
|
3576
|
+
* watcher retries next tick within the
|
|
3577
|
+
* deadline, Finding 4).
|
|
1618
3578
|
*/
|
|
1619
|
-
async markDone(conversationId, messageId, sessionId) {
|
|
3579
|
+
async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
|
|
3580
|
+
const res = await this.fetchImpl(
|
|
3581
|
+
`${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
3582
|
+
{
|
|
3583
|
+
method: "PATCH",
|
|
3584
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3585
|
+
body: JSON.stringify({
|
|
3586
|
+
status: "done",
|
|
3587
|
+
opencode_session_id: sessionId,
|
|
3588
|
+
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
|
|
3589
|
+
})
|
|
3590
|
+
}
|
|
3591
|
+
);
|
|
3592
|
+
this.assertAuth(res, "marking message as done");
|
|
3593
|
+
if (res.ok) return;
|
|
3594
|
+
if (isRetryableStatus(res.status)) {
|
|
3595
|
+
throw new Error(`marking message as done: HTTP ${res.status}`);
|
|
3596
|
+
}
|
|
3597
|
+
throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
|
|
3598
|
+
}
|
|
3599
|
+
/**
|
|
3600
|
+
* Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
|
|
3601
|
+
* when provided (issue #182): a bare `markFailed(conv, msg)` sends
|
|
3602
|
+
* `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
|
|
3603
|
+
* OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
|
|
3604
|
+
* failure reason reaches the channel.
|
|
3605
|
+
*/
|
|
3606
|
+
async markFailed(conversationId, messageId, sessionId, error2) {
|
|
3607
|
+
const body = { status: "failed" };
|
|
3608
|
+
if (sessionId !== void 0) body.opencode_session_id = sessionId;
|
|
3609
|
+
if (error2 !== void 0) body.error = error2;
|
|
1620
3610
|
await this.callWithRetry(
|
|
1621
|
-
"marking message as
|
|
3611
|
+
"marking message as failed",
|
|
1622
3612
|
() => this.fetchImpl(
|
|
1623
3613
|
`${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
1624
3614
|
{
|
|
1625
3615
|
method: "PATCH",
|
|
1626
3616
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
1627
|
-
body: JSON.stringify(
|
|
3617
|
+
body: JSON.stringify(body)
|
|
1628
3618
|
}
|
|
1629
3619
|
)
|
|
1630
3620
|
);
|
|
1631
3621
|
}
|
|
1632
|
-
|
|
1633
|
-
|
|
1634
|
-
|
|
1635
|
-
|
|
1636
|
-
|
|
3622
|
+
/**
|
|
3623
|
+
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
3624
|
+
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
3625
|
+
* — the server records it via `log()` (no DB write, no notification). This is
|
|
3626
|
+
* fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
|
|
3627
|
+
* MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
|
|
3628
|
+
* watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
|
|
3629
|
+
* context (no silent catch, per development-workflow).
|
|
3630
|
+
*
|
|
3631
|
+
* Returns whether the POST SUCCEEDED (2xx). Most callers ignore this (pure
|
|
3632
|
+
* telemetry), but the `paused` liveness-clear uses it to know whether to
|
|
3633
|
+
* RE-ASSERT on a later tick — a single dropped `paused` POST must not leave a
|
|
3634
|
+
* stale `last_seen_alive_at` on a still-paused row (Bugbot "Failed paused signal
|
|
3635
|
+
* leaves liveness").
|
|
3636
|
+
*/
|
|
3637
|
+
async postSignal(conversationId, messageId, signal, extra) {
|
|
3638
|
+
try {
|
|
3639
|
+
const res = await this.fetchImpl(
|
|
3640
|
+
`${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
|
|
1637
3641
|
{
|
|
1638
|
-
method: "
|
|
3642
|
+
method: "POST",
|
|
1639
3643
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
1640
|
-
body: JSON.stringify({
|
|
3644
|
+
body: JSON.stringify({ signal, ...extra })
|
|
1641
3645
|
}
|
|
1642
|
-
)
|
|
1643
|
-
|
|
3646
|
+
);
|
|
3647
|
+
if (!res.ok) {
|
|
3648
|
+
this.log({
|
|
3649
|
+
level: "error",
|
|
3650
|
+
message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
|
|
3651
|
+
conversation_id: conversationId,
|
|
3652
|
+
message_id: messageId
|
|
3653
|
+
});
|
|
3654
|
+
return false;
|
|
3655
|
+
}
|
|
3656
|
+
return true;
|
|
3657
|
+
} catch (err) {
|
|
3658
|
+
this.log({
|
|
3659
|
+
level: "error",
|
|
3660
|
+
message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
|
|
3661
|
+
conversation_id: conversationId,
|
|
3662
|
+
message_id: messageId
|
|
3663
|
+
});
|
|
3664
|
+
return false;
|
|
3665
|
+
}
|
|
1644
3666
|
}
|
|
1645
3667
|
async persistSession(conversationId, sessionId) {
|
|
1646
3668
|
const res = await this.fetchImpl(
|
|
@@ -1655,10 +3677,17 @@ var ChannelDriver = class {
|
|
|
1655
3677
|
}
|
|
1656
3678
|
/**
|
|
1657
3679
|
* EXISTING combinedAuth interaction route (WI-CHAN-3) — idempotent + retried.
|
|
1658
|
-
* `POST .../interactive-event {type, data}`. The server
|
|
1659
|
-
* interaction and posts a link to the proxied opencode-web
|
|
3680
|
+
* `POST .../interactive-event {type, data, source_message_id?}`. The server
|
|
3681
|
+
* persists the interaction and posts a link to the proxied opencode-web
|
|
3682
|
+
* conversation, @mentioning the user who triggered THIS message's turn.
|
|
3683
|
+
*
|
|
3684
|
+
* WI-3 / WI-4 contract: `source_message_id` is the PAUSED message's own Slack
|
|
3685
|
+
* ts (`message.source_message_id`). The server resolves the @mention from that
|
|
3686
|
+
* message's user FIRST (falling back to the old "newest processing" precedence
|
|
3687
|
+
* only when absent), so the correct person is mentioned under concurrency. It
|
|
3688
|
+
* is OPTIONAL for back-compat with older clients / legacy rows.
|
|
1660
3689
|
*/
|
|
1661
|
-
async reportInteraction(conversationId, type, data) {
|
|
3690
|
+
async reportInteraction(conversationId, type, data, sourceMessageId) {
|
|
1662
3691
|
try {
|
|
1663
3692
|
await this.callWithRetry(
|
|
1664
3693
|
"reporting interactive event",
|
|
@@ -1667,7 +3696,9 @@ var ChannelDriver = class {
|
|
|
1667
3696
|
{
|
|
1668
3697
|
method: "POST",
|
|
1669
3698
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
1670
|
-
body: JSON.stringify(
|
|
3699
|
+
body: JSON.stringify(
|
|
3700
|
+
sourceMessageId ? { type, data, source_message_id: sourceMessageId } : { type, data }
|
|
3701
|
+
)
|
|
1671
3702
|
}
|
|
1672
3703
|
)
|
|
1673
3704
|
);
|
|
@@ -1676,6 +3707,7 @@ var ChannelDriver = class {
|
|
|
1676
3707
|
message: `${type} surfaced to channel (id: ${data.id.slice(0, 8)})`,
|
|
1677
3708
|
conversation_id: conversationId
|
|
1678
3709
|
});
|
|
3710
|
+
return true;
|
|
1679
3711
|
} catch (err) {
|
|
1680
3712
|
if (err instanceof ChannelAuthError) throw err;
|
|
1681
3713
|
this.log({
|
|
@@ -1683,6 +3715,7 @@ var ChannelDriver = class {
|
|
|
1683
3715
|
message: `Failed to surface ${type}: ${err instanceof Error ? err.message : String(err)}`,
|
|
1684
3716
|
conversation_id: conversationId
|
|
1685
3717
|
});
|
|
3718
|
+
return false;
|
|
1686
3719
|
}
|
|
1687
3720
|
}
|
|
1688
3721
|
// -------------------------------------------------------------------------
|
|
@@ -1721,8 +3754,9 @@ var ChannelDriver = class {
|
|
|
1721
3754
|
await this.sleep(backoffDelay(attempt, this.retry));
|
|
1722
3755
|
continue;
|
|
1723
3756
|
}
|
|
3757
|
+
break;
|
|
1724
3758
|
}
|
|
1725
|
-
throw new
|
|
3759
|
+
throw new ChannelTerminalError(`${context}: HTTP ${res.status}`, res.status);
|
|
1726
3760
|
}
|
|
1727
3761
|
throw lastError instanceof Error ? lastError : new Error(`${context}: exhausted retries`);
|
|
1728
3762
|
}
|
|
@@ -1848,23 +3882,45 @@ Port ${port} is already in use.`));
|
|
|
1848
3882
|
spinner.fail("Failed to start OpenCode");
|
|
1849
3883
|
throw new Error("OpenCode failed to start");
|
|
1850
3884
|
}
|
|
1851
|
-
spinner.
|
|
1852
|
-
`OpenCode running on port ${port}${health.version ? ` (v${health.version})` : ""}`
|
|
1853
|
-
);
|
|
3885
|
+
spinner.stop();
|
|
1854
3886
|
return { port, process: proc, version: health.version ?? null };
|
|
1855
3887
|
}
|
|
1856
3888
|
return { port, process: null, version: null };
|
|
1857
3889
|
}
|
|
1858
3890
|
|
|
1859
3891
|
// src/commands/agent-lookup.ts
|
|
3892
|
+
async function readErrorMessage(response) {
|
|
3893
|
+
const text = await response.text().catch(() => "");
|
|
3894
|
+
if (!text) return response.statusText || void 0;
|
|
3895
|
+
try {
|
|
3896
|
+
const data = JSON.parse(text);
|
|
3897
|
+
const message = data.message ?? data.error;
|
|
3898
|
+
if (typeof message === "string" && message.trim()) {
|
|
3899
|
+
return message;
|
|
3900
|
+
}
|
|
3901
|
+
} catch {
|
|
3902
|
+
}
|
|
3903
|
+
return text.trim() || response.statusText || void 0;
|
|
3904
|
+
}
|
|
3905
|
+
function authFailureHint(apiUrl, serverMessage) {
|
|
3906
|
+
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
3907
|
+
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
3908
|
+
}
|
|
1860
3909
|
async function resolveAgentIdFromKey(authHeader) {
|
|
1861
3910
|
const apiUrl = getApiUrlConfig();
|
|
1862
3911
|
try {
|
|
1863
3912
|
const response = await fetch(`${apiUrl}/me`, {
|
|
1864
3913
|
headers: { Authorization: authHeader }
|
|
1865
3914
|
});
|
|
3915
|
+
if (response.status === 401) {
|
|
3916
|
+
const serverMessage = await readErrorMessage(response);
|
|
3917
|
+
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
3918
|
+
}
|
|
1866
3919
|
if (!response.ok) {
|
|
1867
|
-
|
|
3920
|
+
const serverMessage = await readErrorMessage(response);
|
|
3921
|
+
return {
|
|
3922
|
+
error: `Failed to resolve agent from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
3923
|
+
};
|
|
1868
3924
|
}
|
|
1869
3925
|
const data = await response.json();
|
|
1870
3926
|
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
@@ -1878,20 +3934,52 @@ async function resolveAgentIdFromKey(authHeader) {
|
|
|
1878
3934
|
return { error: `Failed to resolve agent from key: ${message}` };
|
|
1879
3935
|
}
|
|
1880
3936
|
}
|
|
3937
|
+
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
3938
|
+
const apiUrl = getApiUrlConfig();
|
|
3939
|
+
try {
|
|
3940
|
+
const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
|
|
3941
|
+
method: "POST",
|
|
3942
|
+
headers: { Authorization: authHeader }
|
|
3943
|
+
});
|
|
3944
|
+
if (!response.ok) {
|
|
3945
|
+
const serverMessage = await readErrorMessage(response);
|
|
3946
|
+
return {
|
|
3947
|
+
ok: false,
|
|
3948
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
3949
|
+
};
|
|
3950
|
+
}
|
|
3951
|
+
return { ok: true };
|
|
3952
|
+
} catch (error2) {
|
|
3953
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
3954
|
+
}
|
|
3955
|
+
}
|
|
1881
3956
|
async function getAgentInfo(agentId, authHeader) {
|
|
1882
3957
|
const apiUrl = getApiUrlConfig();
|
|
1883
3958
|
try {
|
|
1884
3959
|
const response = await fetch(`${apiUrl}/agents/${agentId}`, {
|
|
1885
3960
|
headers: { Authorization: authHeader }
|
|
1886
3961
|
});
|
|
1887
|
-
if (response.status === 404) {
|
|
1888
|
-
return { valid: false, error: "Agent not found" };
|
|
1889
|
-
}
|
|
1890
3962
|
if (response.status === 401) {
|
|
1891
|
-
|
|
3963
|
+
const serverMessage = await readErrorMessage(response);
|
|
3964
|
+
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
3965
|
+
}
|
|
3966
|
+
if (response.status === 403) {
|
|
3967
|
+
const serverMessage = await readErrorMessage(response);
|
|
3968
|
+
return {
|
|
3969
|
+
valid: false,
|
|
3970
|
+
error: serverMessage ?? "You do not have access to this agent (it may belong to a different team or organization)."
|
|
3971
|
+
};
|
|
3972
|
+
}
|
|
3973
|
+
if (response.status === 404) {
|
|
3974
|
+
const serverMessage = await readErrorMessage(response);
|
|
3975
|
+
return { valid: false, error: serverMessage ?? `Agent ${agentId} not found` };
|
|
1892
3976
|
}
|
|
1893
3977
|
if (!response.ok) {
|
|
1894
|
-
|
|
3978
|
+
const serverMessage = await readErrorMessage(response);
|
|
3979
|
+
return {
|
|
3980
|
+
valid: false,
|
|
3981
|
+
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
3982
|
+
};
|
|
1895
3983
|
}
|
|
1896
3984
|
const agent = await response.json();
|
|
1897
3985
|
if (agent.agent_type !== "local") {
|
|
@@ -1910,7 +3998,9 @@ async function getAgentInfo(agentId, authHeader) {
|
|
|
1910
3998
|
// src/commands/run.ts
|
|
1911
3999
|
var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
1912
4000
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
1913
|
-
|
|
4001
|
+
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4002
|
+
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
4003
|
+
function log2(state, message, isError = false) {
|
|
1914
4004
|
if (state.json) {
|
|
1915
4005
|
console.log(
|
|
1916
4006
|
JSON.stringify({
|
|
@@ -1935,9 +4025,9 @@ function logActivity(state, entry) {
|
|
|
1935
4025
|
}
|
|
1936
4026
|
if (!state.interactive) {
|
|
1937
4027
|
if (entry.type === "error") {
|
|
1938
|
-
|
|
4028
|
+
log2(state, entry.error ?? "Unknown error", true);
|
|
1939
4029
|
} else if (entry.type === "info" && entry.message) {
|
|
1940
|
-
|
|
4030
|
+
log2(state, entry.message);
|
|
1941
4031
|
}
|
|
1942
4032
|
}
|
|
1943
4033
|
}
|
|
@@ -2023,6 +4113,7 @@ async function handleAuthError(state, error2) {
|
|
|
2023
4113
|
}
|
|
2024
4114
|
async function driveChannels(state, driver) {
|
|
2025
4115
|
let idlePolls = 0;
|
|
4116
|
+
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
2026
4117
|
while (state.running) {
|
|
2027
4118
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
2028
4119
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
@@ -2032,9 +4123,11 @@ async function driveChannels(state, driver) {
|
|
|
2032
4123
|
try {
|
|
2033
4124
|
const processed = await driver.drainPending();
|
|
2034
4125
|
state.messageCount += processed;
|
|
2035
|
-
|
|
4126
|
+
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4127
|
+
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4128
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
|
|
2036
4129
|
idlePolls = 0;
|
|
2037
|
-
if (state.interactive) displayStatus(state);
|
|
4130
|
+
if (processed > 0 && state.interactive) displayStatus(state);
|
|
2038
4131
|
} else if (state.idleTimeout !== null) {
|
|
2039
4132
|
idlePolls++;
|
|
2040
4133
|
if (idlePolls === 1) {
|
|
@@ -2061,7 +4154,7 @@ async function driveChannels(state, driver) {
|
|
|
2061
4154
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
2062
4155
|
if (state.interactive) displayStatus(state);
|
|
2063
4156
|
}
|
|
2064
|
-
await new Promise((
|
|
4157
|
+
await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
|
|
2065
4158
|
if (state.idleTimeout !== null && idlePolls >= 2) {
|
|
2066
4159
|
const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
|
|
2067
4160
|
if (idleMs > state.idleTimeout * 1e3) {
|
|
@@ -2072,8 +4165,122 @@ async function driveChannels(state, driver) {
|
|
|
2072
4165
|
}
|
|
2073
4166
|
}
|
|
2074
4167
|
}
|
|
2075
|
-
|
|
4168
|
+
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
4169
|
+
async function runSweep(state, driver, config2) {
|
|
4170
|
+
const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
|
|
4171
|
+
try {
|
|
4172
|
+
const sessions = await listSessions(state.port);
|
|
4173
|
+
if (sessions === null) {
|
|
4174
|
+
logActivity(state, {
|
|
4175
|
+
type: "info",
|
|
4176
|
+
message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
|
|
4177
|
+
});
|
|
4178
|
+
return;
|
|
4179
|
+
}
|
|
4180
|
+
const toDelete = selectSessionsToDelete(
|
|
4181
|
+
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
4182
|
+
{
|
|
4183
|
+
maxAgeMs: config2.maxAgeMs,
|
|
4184
|
+
maxCount: config2.maxCount,
|
|
4185
|
+
nowMs: Date.now(),
|
|
4186
|
+
protectedIds: driver.protectedSessionIds()
|
|
4187
|
+
}
|
|
4188
|
+
);
|
|
4189
|
+
const protectedNow = driver.protectedSessionIds();
|
|
4190
|
+
let deleted = 0;
|
|
4191
|
+
let failed = 0;
|
|
4192
|
+
let skippedNewlyActive = 0;
|
|
4193
|
+
for (const id of toDelete) {
|
|
4194
|
+
if (protectedNow.has(id)) {
|
|
4195
|
+
skippedNewlyActive++;
|
|
4196
|
+
logActivity(state, {
|
|
4197
|
+
type: "info",
|
|
4198
|
+
message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
|
|
4199
|
+
});
|
|
4200
|
+
continue;
|
|
4201
|
+
}
|
|
4202
|
+
if (await deleteSession(state.port, id)) deleted++;
|
|
4203
|
+
else failed++;
|
|
4204
|
+
}
|
|
4205
|
+
const failedNote = failed > 0 ? `, failed ${failed}` : "";
|
|
4206
|
+
const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
|
|
4207
|
+
logActivity(state, {
|
|
4208
|
+
type: "info",
|
|
4209
|
+
message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
|
|
4210
|
+
});
|
|
4211
|
+
} catch (error2) {
|
|
4212
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4213
|
+
logActivity(state, {
|
|
4214
|
+
type: "error",
|
|
4215
|
+
error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
|
|
4216
|
+
});
|
|
4217
|
+
}
|
|
4218
|
+
}
|
|
4219
|
+
function scheduleSessionCleanup(state, driver, options) {
|
|
4220
|
+
const config2 = resolveSessionCleanupConfig(
|
|
4221
|
+
{
|
|
4222
|
+
maxAge: options.sessionCleanupMaxAge,
|
|
4223
|
+
maxCount: options.sessionCleanupMaxCount,
|
|
4224
|
+
interval: options.sessionCleanupInterval
|
|
4225
|
+
},
|
|
4226
|
+
process.env
|
|
4227
|
+
);
|
|
4228
|
+
for (const warning2 of config2.warnings) {
|
|
4229
|
+
logActivity(state, { type: "info", message: `Session cleanup: ${warning2}` });
|
|
4230
|
+
}
|
|
4231
|
+
if (!config2.enabled) return;
|
|
4232
|
+
logActivity(state, {
|
|
4233
|
+
type: "info",
|
|
4234
|
+
message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
|
|
4235
|
+
});
|
|
4236
|
+
const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
|
|
4237
|
+
const firstSweep = setTimeout(
|
|
4238
|
+
() => void runSweep(state, driver, config2),
|
|
4239
|
+
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
4240
|
+
);
|
|
4241
|
+
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
4242
|
+
}
|
|
4243
|
+
async function notifyOffline(state) {
|
|
4244
|
+
if (!state.agentId || !state.authHeader) return;
|
|
4245
|
+
if (!state.connected) {
|
|
4246
|
+
log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
|
|
4247
|
+
return;
|
|
4248
|
+
}
|
|
4249
|
+
const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
|
|
4250
|
+
if (result.ok) {
|
|
4251
|
+
log2(state, "Notified Evident the agent is going offline");
|
|
4252
|
+
} else {
|
|
4253
|
+
logActivity(state, {
|
|
4254
|
+
type: "error",
|
|
4255
|
+
error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
|
|
4256
|
+
});
|
|
4257
|
+
if (state.interactive) displayStatus(state);
|
|
4258
|
+
}
|
|
4259
|
+
}
|
|
4260
|
+
async function cleanup(state, opts = {}) {
|
|
2076
4261
|
state.running = false;
|
|
4262
|
+
for (const timer of state.sessionCleanupTimers) {
|
|
4263
|
+
clearInterval(timer);
|
|
4264
|
+
clearTimeout(timer);
|
|
4265
|
+
}
|
|
4266
|
+
state.sessionCleanupTimers = [];
|
|
4267
|
+
if (opts.graceful && state.channelDriver) {
|
|
4268
|
+
state.channelDriver.stop();
|
|
4269
|
+
log2(state, "Draining in-flight channel work before shutdown...");
|
|
4270
|
+
if (state.interactive) {
|
|
4271
|
+
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4272
|
+
displayStatus(state);
|
|
4273
|
+
}
|
|
4274
|
+
const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
|
|
4275
|
+
if (!settled) {
|
|
4276
|
+
logActivity(state, {
|
|
4277
|
+
type: "info",
|
|
4278
|
+
message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
|
|
4279
|
+
});
|
|
4280
|
+
if (state.interactive) displayStatus(state);
|
|
4281
|
+
}
|
|
4282
|
+
}
|
|
4283
|
+
await notifyOffline(state);
|
|
2077
4284
|
if (state.connection) {
|
|
2078
4285
|
state.connection.close();
|
|
2079
4286
|
state.connection = null;
|
|
@@ -2084,7 +4291,7 @@ async function cleanup(state) {
|
|
|
2084
4291
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
2085
4292
|
displayStatus(state);
|
|
2086
4293
|
} else {
|
|
2087
|
-
|
|
4294
|
+
log2(state, "Stopped OpenCode process");
|
|
2088
4295
|
}
|
|
2089
4296
|
state.opencodeProcess = null;
|
|
2090
4297
|
}
|
|
@@ -2104,26 +4311,32 @@ async function run(options) {
|
|
|
2104
4311
|
opencodeVersion: null,
|
|
2105
4312
|
opencodeProcess: null,
|
|
2106
4313
|
connection: null,
|
|
4314
|
+
channelDriver: null,
|
|
2107
4315
|
running: true,
|
|
4316
|
+
shuttingDown: false,
|
|
2108
4317
|
activityLog: [],
|
|
2109
4318
|
messageCount: 0,
|
|
4319
|
+
lastProxiedActivityAt: null,
|
|
4320
|
+
sessionCleanupTimers: [],
|
|
2110
4321
|
authHeader: ""
|
|
2111
4322
|
};
|
|
2112
4323
|
if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
|
|
2113
|
-
|
|
4324
|
+
log2(
|
|
2114
4325
|
state,
|
|
2115
4326
|
"Warning: No --idle-timeout set in CI environment. The runner will poll indefinitely until the job times out. Consider adding --idle-timeout 30 to avoid wasting runner minutes.",
|
|
2116
4327
|
false
|
|
2117
4328
|
);
|
|
2118
4329
|
}
|
|
2119
4330
|
const handleSignal = async () => {
|
|
4331
|
+
if (state.shuttingDown) return;
|
|
4332
|
+
state.shuttingDown = true;
|
|
2120
4333
|
if (state.interactive) {
|
|
2121
4334
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
2122
4335
|
displayStatus(state);
|
|
2123
4336
|
} else {
|
|
2124
|
-
|
|
4337
|
+
log2(state, "Shutting down...");
|
|
2125
4338
|
}
|
|
2126
|
-
await cleanup(state);
|
|
4339
|
+
await cleanup(state, { graceful: true });
|
|
2127
4340
|
await shutdownTelemetry();
|
|
2128
4341
|
process.exit(0);
|
|
2129
4342
|
};
|
|
@@ -2154,7 +4367,7 @@ async function run(options) {
|
|
|
2154
4367
|
const resolved = await resolveAgentIdFromKey(state.authHeader);
|
|
2155
4368
|
if (resolved.agent_id) {
|
|
2156
4369
|
state.agentId = resolved.agent_id;
|
|
2157
|
-
|
|
4370
|
+
log2(state, `Resolved agent ID from key: ${state.agentId}`);
|
|
2158
4371
|
if (state.interactive && !state.json) {
|
|
2159
4372
|
logActivity(state, {
|
|
2160
4373
|
type: "info",
|
|
@@ -2217,14 +4430,21 @@ async function run(options) {
|
|
|
2217
4430
|
port: state.port,
|
|
2218
4431
|
interactive: state.interactive,
|
|
2219
4432
|
agentId: state.agentId,
|
|
2220
|
-
log: (message) =>
|
|
4433
|
+
log: (message) => log2(state, message)
|
|
2221
4434
|
});
|
|
2222
4435
|
state.port = oc.port;
|
|
2223
4436
|
state.opencodeProcess = oc.process;
|
|
2224
4437
|
state.opencodeVersion = oc.version;
|
|
2225
4438
|
state.opencodeConnected = oc.process !== null || oc.version !== null;
|
|
2226
|
-
const
|
|
2227
|
-
ocSpinner?.succeed(`OpenCode running on port ${state.port}${
|
|
4439
|
+
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
4440
|
+
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
4441
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
4442
|
+
if (versionWarning) {
|
|
4443
|
+
log2(state, versionWarning, false);
|
|
4444
|
+
if (state.interactive && !state.json) {
|
|
4445
|
+
logActivity(state, { type: "info", message: versionWarning });
|
|
4446
|
+
}
|
|
4447
|
+
}
|
|
2228
4448
|
} catch (error2) {
|
|
2229
4449
|
ocSpinner?.fail(error2.message);
|
|
2230
4450
|
throw error2;
|
|
@@ -2236,12 +4456,14 @@ async function run(options) {
|
|
|
2236
4456
|
apiUrl: getApiUrlConfig(),
|
|
2237
4457
|
getAuthHeader: () => state.authHeader,
|
|
2238
4458
|
conversationFilter: state.conversationFilter,
|
|
4459
|
+
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
2239
4460
|
log: (entry) => logActivity(state, {
|
|
2240
4461
|
type: entry.level === "error" ? "error" : "info",
|
|
2241
4462
|
message: entry.message,
|
|
2242
4463
|
error: entry.level === "error" ? entry.message : void 0
|
|
2243
4464
|
})
|
|
2244
4465
|
});
|
|
4466
|
+
state.channelDriver = channelDriver;
|
|
2245
4467
|
const connection = new RunnerConnection({
|
|
2246
4468
|
agentId: state.agentId,
|
|
2247
4469
|
getAuthHeader: () => state.authHeader,
|
|
@@ -2255,10 +4477,29 @@ async function run(options) {
|
|
|
2255
4477
|
type: "info",
|
|
2256
4478
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
|
|
2257
4479
|
});
|
|
2258
|
-
emitAgentConnected(state.agentId, {
|
|
4480
|
+
emitAgentConnected(state.agentId, {
|
|
4481
|
+
port: state.port,
|
|
4482
|
+
cli_version: getCliVersion(),
|
|
4483
|
+
opencode_version: state.opencodeVersion
|
|
4484
|
+
});
|
|
2259
4485
|
if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
|
|
2260
4486
|
if (state.interactive) displayStatus(state);
|
|
2261
|
-
channelDriver.drainPending().
|
|
4487
|
+
channelDriver.drainPending().then((processed) => {
|
|
4488
|
+
if (processed > 0) {
|
|
4489
|
+
state.messageCount += processed;
|
|
4490
|
+
logActivity(state, {
|
|
4491
|
+
type: "info",
|
|
4492
|
+
message: `Drained ${processed} queued message(s) on connect`
|
|
4493
|
+
});
|
|
4494
|
+
if (state.interactive) displayStatus(state);
|
|
4495
|
+
}
|
|
4496
|
+
}).catch((error2) => {
|
|
4497
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4498
|
+
logActivity(state, {
|
|
4499
|
+
type: "error",
|
|
4500
|
+
error: `Failed to drain queued messages on connect: ${message}`
|
|
4501
|
+
});
|
|
4502
|
+
if (state.interactive) displayStatus(state);
|
|
2262
4503
|
});
|
|
2263
4504
|
},
|
|
2264
4505
|
onDisconnected: (code, reason) => {
|
|
@@ -2274,9 +4515,40 @@ async function run(options) {
|
|
|
2274
4515
|
logActivity(state, { type: "error", error: error2 });
|
|
2275
4516
|
if (state.interactive) displayStatus(state);
|
|
2276
4517
|
},
|
|
2277
|
-
// Web traffic is proxied transparently;
|
|
4518
|
+
// Web traffic is proxied transparently; note opencode is live and stamp
|
|
4519
|
+
// proxied activity so the idle loop treats interactive proxy use as work.
|
|
4520
|
+
// Fires per forwarded response head (incl. every SSE open) and excludes
|
|
4521
|
+
// the internal drain-ping, so an actively-used proxy keeps the timer
|
|
4522
|
+
// fresh while a lone idle SSE with no follow-up requests still ages out.
|
|
2278
4523
|
onResponse: () => {
|
|
2279
4524
|
state.opencodeConnected = true;
|
|
4525
|
+
state.lastProxiedActivityAt = Date.now();
|
|
4526
|
+
},
|
|
4527
|
+
// A channel message was queued and the api-worker pinged us over the
|
|
4528
|
+
// tunnel to drain immediately instead of waiting for the next poll tick.
|
|
4529
|
+
// Best-effort + non-fatal: mirror the on-connect drain block. A failed
|
|
4530
|
+
// drain here is logged and swallowed — the steady-state poll retries, so
|
|
4531
|
+
// a lost/failed ping can never orphan a message (§2 invariant).
|
|
4532
|
+
onDrainPing: () => {
|
|
4533
|
+
if (!state.running) return;
|
|
4534
|
+
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
4535
|
+
channelDriver.drainPending().then((processed) => {
|
|
4536
|
+
if (processed > 0) {
|
|
4537
|
+
state.messageCount += processed;
|
|
4538
|
+
logActivity(state, {
|
|
4539
|
+
type: "info",
|
|
4540
|
+
message: `Drained ${processed} queued message(s) on ping`
|
|
4541
|
+
});
|
|
4542
|
+
if (state.interactive) displayStatus(state);
|
|
4543
|
+
}
|
|
4544
|
+
}).catch((error2) => {
|
|
4545
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4546
|
+
logActivity(state, {
|
|
4547
|
+
type: "error",
|
|
4548
|
+
error: `Failed to drain queued messages on ping: ${message}`
|
|
4549
|
+
});
|
|
4550
|
+
if (state.interactive) displayStatus(state);
|
|
4551
|
+
});
|
|
2280
4552
|
},
|
|
2281
4553
|
onInfo: (message) => logActivity(state, { type: "info", message })
|
|
2282
4554
|
}
|
|
@@ -2288,12 +4560,12 @@ async function run(options) {
|
|
|
2288
4560
|
if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
|
|
2289
4561
|
throw error2;
|
|
2290
4562
|
}
|
|
2291
|
-
|
|
2292
|
-
|
|
2293
|
-
|
|
2294
|
-
log(state, "Driving channel messages...");
|
|
4563
|
+
scheduleSessionCleanup(state, channelDriver, options);
|
|
4564
|
+
if (!interactive || state.json) {
|
|
4565
|
+
log2(state, "Driving channel messages...");
|
|
2295
4566
|
}
|
|
2296
4567
|
await driveChannels(state, channelDriver);
|
|
4568
|
+
if (state.shuttingDown) return;
|
|
2297
4569
|
await cleanup(state);
|
|
2298
4570
|
if (state.json) {
|
|
2299
4571
|
console.log(
|
|
@@ -2303,11 +4575,12 @@ async function run(options) {
|
|
|
2303
4575
|
})
|
|
2304
4576
|
);
|
|
2305
4577
|
} else if (!interactive) {
|
|
2306
|
-
|
|
4578
|
+
log2(state, `Completed. Processed ${state.messageCount} message(s).`);
|
|
2307
4579
|
}
|
|
2308
4580
|
await shutdownTelemetry();
|
|
2309
4581
|
process.exit(0);
|
|
2310
4582
|
} catch (error2) {
|
|
4583
|
+
if (state.shuttingDown) return;
|
|
2311
4584
|
await cleanup(state);
|
|
2312
4585
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
2313
4586
|
if (state.json) {
|
|
@@ -2325,8 +4598,9 @@ async function run(options) {
|
|
|
2325
4598
|
}
|
|
2326
4599
|
|
|
2327
4600
|
// src/index.ts
|
|
4601
|
+
var { version } = createRequire(import.meta.url)("../package.json");
|
|
2328
4602
|
var program = new Command();
|
|
2329
|
-
program.name("evident").description("Run OpenCode locally and connect it to Evident").version(
|
|
4603
|
+
program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
|
|
2330
4604
|
"--endpoint <url>",
|
|
2331
4605
|
"Evident API base URL (default: production; e.g. http://localhost:3001)"
|
|
2332
4606
|
).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {
|
|
@@ -2339,9 +4613,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
2339
4613
|
}
|
|
2340
4614
|
});
|
|
2341
4615
|
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);
|
|
2342
|
-
program.command("logout").description("Remove stored credentials").action(logout);
|
|
4616
|
+
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 }));
|
|
2343
4617
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
2344
|
-
program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").
|
|
4618
|
+
program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").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(
|
|
4619
|
+
"--session-cleanup-max-age <duration>",
|
|
4620
|
+
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
4621
|
+
).option(
|
|
4622
|
+
"--session-cleanup-max-count <n>",
|
|
4623
|
+
"Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
|
|
4624
|
+
).option(
|
|
4625
|
+
"--session-cleanup-interval <duration>",
|
|
4626
|
+
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
4627
|
+
).action(
|
|
2345
4628
|
(options) => {
|
|
2346
4629
|
run({
|
|
2347
4630
|
agent: options.agent,
|
|
@@ -2349,7 +4632,11 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
2349
4632
|
verbose: options.verbose,
|
|
2350
4633
|
conversation: options.conversation,
|
|
2351
4634
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
2352
|
-
json: options.json
|
|
4635
|
+
json: options.json,
|
|
4636
|
+
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
4637
|
+
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
4638
|
+
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
4639
|
+
sessionCleanupInterval: options.sessionCleanupInterval
|
|
2353
4640
|
});
|
|
2354
4641
|
}
|
|
2355
4642
|
);
|