@namzu/sdk 7.0.0 → 8.0.0
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/CHANGELOG.md +437 -0
- package/dist/agents/SupervisorAgent.d.ts.map +1 -1
- package/dist/agents/SupervisorAgent.js +172 -162
- package/dist/agents/SupervisorAgent.js.map +1 -1
- package/dist/agents/__tests__/supervisor-inbox-scope.test.d.ts +2 -0
- package/dist/agents/__tests__/supervisor-inbox-scope.test.d.ts.map +1 -0
- package/dist/agents/__tests__/supervisor-inbox-scope.test.js +125 -0
- package/dist/agents/__tests__/supervisor-inbox-scope.test.js.map +1 -0
- package/dist/bridge/a2a/mapper.d.ts.map +1 -1
- package/dist/bridge/a2a/mapper.js +4 -0
- package/dist/bridge/a2a/mapper.js.map +1 -1
- package/dist/bridge/sse/mapper.d.ts.map +1 -1
- package/dist/bridge/sse/mapper.js +24 -0
- package/dist/bridge/sse/mapper.js.map +1 -1
- package/dist/contracts/api.d.ts +1 -1
- package/dist/contracts/api.d.ts.map +1 -1
- package/dist/gateway/__tests__/completion-inbox.test.js +292 -2
- package/dist/gateway/__tests__/completion-inbox.test.js.map +1 -1
- package/dist/gateway/completion-inbox.d.ts +94 -6
- package/dist/gateway/completion-inbox.d.ts.map +1 -1
- package/dist/gateway/completion-inbox.js +235 -15
- package/dist/gateway/completion-inbox.js.map +1 -1
- package/dist/manager/run/persistence.d.ts +8 -0
- package/dist/manager/run/persistence.d.ts.map +1 -1
- package/dist/manager/run/persistence.js +12 -0
- package/dist/manager/run/persistence.js.map +1 -1
- package/dist/run/reporter.d.ts.map +1 -1
- package/dist/run/reporter.js +11 -0
- package/dist/run/reporter.js.map +1 -1
- package/dist/runtime/query/__tests__/completion-does-not-erase-the-answer.test.d.ts +2 -0
- package/dist/runtime/query/__tests__/completion-does-not-erase-the-answer.test.d.ts.map +1 -0
- package/dist/runtime/query/__tests__/completion-does-not-erase-the-answer.test.js +142 -0
- package/dist/runtime/query/__tests__/completion-does-not-erase-the-answer.test.js.map +1 -0
- package/dist/runtime/query/__tests__/completion-notification.test.js +414 -32
- package/dist/runtime/query/__tests__/completion-notification.test.js.map +1 -1
- package/dist/runtime/query/__tests__/context-size-on-the-wire.test.d.ts +2 -0
- package/dist/runtime/query/__tests__/context-size-on-the-wire.test.d.ts.map +1 -0
- package/dist/runtime/query/__tests__/context-size-on-the-wire.test.js +100 -0
- package/dist/runtime/query/__tests__/context-size-on-the-wire.test.js.map +1 -0
- package/dist/runtime/query/guard.d.ts +28 -0
- package/dist/runtime/query/guard.d.ts.map +1 -1
- package/dist/runtime/query/guard.js +31 -0
- package/dist/runtime/query/guard.js.map +1 -1
- package/dist/runtime/query/iteration/__tests__/settle-grace.test.d.ts +2 -0
- package/dist/runtime/query/iteration/__tests__/settle-grace.test.d.ts.map +1 -0
- package/dist/runtime/query/iteration/__tests__/settle-grace.test.js +226 -0
- package/dist/runtime/query/iteration/__tests__/settle-grace.test.js.map +1 -0
- package/dist/runtime/query/iteration/index.d.ts +92 -0
- package/dist/runtime/query/iteration/index.d.ts.map +1 -1
- package/dist/runtime/query/iteration/index.js +814 -566
- package/dist/runtime/query/iteration/index.js.map +1 -1
- package/dist/runtime/query/iteration/phases/__tests__/compaction-declined.test.d.ts +2 -0
- package/dist/runtime/query/iteration/phases/__tests__/compaction-declined.test.d.ts.map +1 -0
- package/dist/runtime/query/iteration/phases/__tests__/compaction-declined.test.js +95 -0
- package/dist/runtime/query/iteration/phases/__tests__/compaction-declined.test.js.map +1 -0
- package/dist/runtime/query/iteration/phases/compaction.d.ts +34 -0
- package/dist/runtime/query/iteration/phases/compaction.d.ts.map +1 -1
- package/dist/runtime/query/iteration/phases/compaction.js +61 -4
- package/dist/runtime/query/iteration/phases/compaction.js.map +1 -1
- package/dist/tools/__tests__/untrusted-envelope.test.js +16 -0
- package/dist/tools/__tests__/untrusted-envelope.test.js.map +1 -1
- package/dist/tools/coordinator/__tests__/completion-delivery.test.js +117 -0
- package/dist/tools/coordinator/__tests__/completion-delivery.test.js.map +1 -1
- package/dist/tools/coordinator/__tests__/task-list.test.js +57 -0
- package/dist/tools/coordinator/__tests__/task-list.test.js.map +1 -1
- package/dist/tools/coordinator/index.d.ts.map +1 -1
- package/dist/tools/coordinator/index.js +109 -11
- package/dist/tools/coordinator/index.js.map +1 -1
- package/dist/tools/untrusted-envelope.d.ts.map +1 -1
- package/dist/tools/untrusted-envelope.js +9 -1
- package/dist/tools/untrusted-envelope.js.map +1 -1
- package/dist/types/agent/gateway.d.ts +18 -0
- package/dist/types/agent/gateway.d.ts.map +1 -1
- package/dist/types/provider/interface.d.ts +35 -0
- package/dist/types/provider/interface.d.ts.map +1 -1
- package/dist/types/run/entity.d.ts +16 -0
- package/dist/types/run/entity.d.ts.map +1 -1
- package/dist/types/run/events.d.ts +75 -0
- package/dist/types/run/events.d.ts.map +1 -1
- package/dist/types/run/events.js.map +1 -1
- package/dist/types/run/prepare-step.d.ts +17 -2
- package/dist/types/run/prepare-step.d.ts.map +1 -1
- package/dist/types/verification/index.d.ts +98 -0
- package/dist/types/verification/index.d.ts.map +1 -1
- package/dist/types/verification/index.js +10 -0
- package/dist/types/verification/index.js.map +1 -1
- package/dist/verification/__tests__/argument-pattern.test.d.ts +2 -0
- package/dist/verification/__tests__/argument-pattern.test.d.ts.map +1 -0
- package/dist/verification/__tests__/argument-pattern.test.js +122 -0
- package/dist/verification/__tests__/argument-pattern.test.js.map +1 -0
- package/dist/verification/gate.d.ts.map +1 -1
- package/dist/verification/gate.js +44 -0
- package/dist/verification/gate.js.map +1 -1
- package/dist/verification/rules.d.ts.map +1 -1
- package/dist/verification/rules.js +27 -0
- package/dist/verification/rules.js.map +1 -1
- package/package.json +1 -1
- package/src/agents/SupervisorAgent.ts +175 -166
- package/src/agents/__tests__/supervisor-inbox-scope.test.ts +149 -0
- package/src/bridge/a2a/mapper.ts +4 -0
- package/src/bridge/sse/mapper.ts +25 -0
- package/src/contracts/api.ts +1 -0
- package/src/gateway/__tests__/completion-inbox.test.ts +348 -2
- package/src/gateway/completion-inbox.ts +248 -16
- package/src/manager/run/persistence.ts +12 -0
- package/src/run/reporter.ts +12 -0
- package/src/runtime/query/__tests__/completion-does-not-erase-the-answer.test.ts +163 -0
- package/src/runtime/query/__tests__/completion-notification.test.ts +486 -34
- package/src/runtime/query/__tests__/context-size-on-the-wire.test.ts +122 -0
- package/src/runtime/query/guard.ts +32 -0
- package/src/runtime/query/iteration/__tests__/settle-grace.test.ts +265 -0
- package/src/runtime/query/iteration/index.ts +902 -636
- package/src/runtime/query/iteration/phases/__tests__/compaction-declined.test.ts +124 -0
- package/src/runtime/query/iteration/phases/compaction.ts +83 -10
- package/src/tools/__tests__/untrusted-envelope.test.ts +23 -0
- package/src/tools/coordinator/__tests__/completion-delivery.test.ts +147 -0
- package/src/tools/coordinator/__tests__/task-list.test.ts +72 -0
- package/src/tools/coordinator/index.ts +122 -15
- package/src/tools/untrusted-envelope.ts +9 -1
- package/src/types/agent/gateway.ts +18 -0
- package/src/types/provider/interface.ts +39 -0
- package/src/types/run/entity.ts +17 -0
- package/src/types/run/events.ts +75 -0
- package/src/types/run/prepare-step.ts +17 -2
- package/src/types/verification/index.ts +61 -0
- package/src/verification/__tests__/argument-pattern.test.ts +158 -0
- package/src/verification/gate.ts +45 -0
- package/src/verification/rules.ts +28 -0
|
@@ -34,6 +34,34 @@ export declare class GuardCoordinator {
|
|
|
34
34
|
* one field, let the resume branch hand it over.
|
|
35
35
|
*/
|
|
36
36
|
restoreElapsed(elapsedMs: number): void;
|
|
37
|
+
/**
|
|
38
|
+
* Wall-clock left before this run is asked to start finishing.
|
|
39
|
+
*
|
|
40
|
+
* NOT the time left before the deadline, and the difference is the whole
|
|
41
|
+
* point. The checks above run BETWEEN iterations, so anything that waits
|
|
42
|
+
* inside one cannot be stopped by them, and a caller sizing such a wait
|
|
43
|
+
* needs a number the run actually owns: a fixed two-minute hold measured
|
|
44
|
+
* against a run configured for twenty seconds kept it open for 120,267 ms.
|
|
45
|
+
*
|
|
46
|
+
* Measuring to the DEADLINE was the first attempt and it was wrong. The
|
|
47
|
+
* binding constraint is `budgetWarningThreshold`, the point at which this
|
|
48
|
+
* guard stops asking for more work and asks for a closing summary — that
|
|
49
|
+
* last slice exists so the run can produce an answer, and a wait sized
|
|
50
|
+
* against the deadline eats into it. Half of the time-to-deadline, started
|
|
51
|
+
* just under the threshold, ends at 95% of the budget: half the closing
|
|
52
|
+
* reserve spent waiting for a result the closing answer was supposed to
|
|
53
|
+
* use.
|
|
54
|
+
*
|
|
55
|
+
* Zero once the threshold has passed, which is also how a caller gets the
|
|
56
|
+
* re-evaluation it needs: `forceFinalize` is sampled at the top of an
|
|
57
|
+
* iteration and this is read when the wait is about to start, so a long
|
|
58
|
+
* iteration that crossed the line in between is told to wait for nothing.
|
|
59
|
+
*
|
|
60
|
+
* Reads through the same `startTime` the limit checks use, so a run
|
|
61
|
+
* resumed from a checkpoint (see `restoreElapsed`) reports the time left
|
|
62
|
+
* on the RUN rather than on the process now hosting it.
|
|
63
|
+
*/
|
|
64
|
+
remainingBeforeFinalizeMs(): number;
|
|
37
65
|
beforeIteration(runMgr: RunPersistence, abortSignal: AbortSignal): GuardCheckResult;
|
|
38
66
|
}
|
|
39
67
|
//# sourceMappingURL=guard.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"guard.d.ts","sourceRoot":"","sources":["../../../src/runtime/query/guard.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,kCAAkC,CAAA;AAEtE,OAAO,KAAK,EAAsB,UAAU,EAAE,MAAM,0BAA0B,CAAA;AAE9E,MAAM,WAAW,WAAW;IAC3B,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB;;;;;;;OAOG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;CACxB;AAED,MAAM,WAAW,gBAAgB;IAChC,UAAU,EAAE,OAAO,CAAA;IACnB,aAAa,EAAE,OAAO,CAAA;IACtB,UAAU,CAAC,EAAE,UAAU,CAAA;IACvB,WAAW,EAAE,OAAO,CAAA;CACpB;AAED,qBAAa,gBAAgB;IAC5B,OAAO,CAAC,WAAW,CAAoB;IACvC,OAAO,CAAC,SAAS,CAAQ;gBAEb,MAAM,EAAE,WAAW;IAc/B;;;;;;;OAOG;IACH,cAAc,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI;IAIvC,eAAe,CAAC,MAAM,EAAE,cAAc,EAAE,WAAW,EAAE,WAAW,GAAG,gBAAgB;CAmCnF"}
|
|
1
|
+
{"version":3,"file":"guard.d.ts","sourceRoot":"","sources":["../../../src/runtime/query/guard.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,kCAAkC,CAAA;AAEtE,OAAO,KAAK,EAAsB,UAAU,EAAE,MAAM,0BAA0B,CAAA;AAE9E,MAAM,WAAW,WAAW;IAC3B,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB;;;;;;;OAOG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;CACxB;AAED,MAAM,WAAW,gBAAgB;IAChC,UAAU,EAAE,OAAO,CAAA;IACnB,aAAa,EAAE,OAAO,CAAA;IACtB,UAAU,CAAC,EAAE,UAAU,CAAA;IACvB,WAAW,EAAE,OAAO,CAAA;CACpB;AAED,qBAAa,gBAAgB;IAC5B,OAAO,CAAC,WAAW,CAAoB;IACvC,OAAO,CAAC,SAAS,CAAQ;gBAEb,MAAM,EAAE,WAAW;IAc/B;;;;;;;OAOG;IACH,cAAc,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI;IAIvC;;;;;;;;;;;;;;;;;;;;;;;;;;OA0BG;IACH,yBAAyB,IAAI,MAAM;IAKnC,eAAe,CAAC,MAAM,EAAE,cAAc,EAAE,WAAW,EAAE,WAAW,GAAG,gBAAgB;CAmCnF"}
|
|
@@ -21,6 +21,37 @@ export class GuardCoordinator {
|
|
|
21
21
|
restoreElapsed(elapsedMs) {
|
|
22
22
|
this.startTime = Date.now() - Math.max(0, elapsedMs);
|
|
23
23
|
}
|
|
24
|
+
/**
|
|
25
|
+
* Wall-clock left before this run is asked to start finishing.
|
|
26
|
+
*
|
|
27
|
+
* NOT the time left before the deadline, and the difference is the whole
|
|
28
|
+
* point. The checks above run BETWEEN iterations, so anything that waits
|
|
29
|
+
* inside one cannot be stopped by them, and a caller sizing such a wait
|
|
30
|
+
* needs a number the run actually owns: a fixed two-minute hold measured
|
|
31
|
+
* against a run configured for twenty seconds kept it open for 120,267 ms.
|
|
32
|
+
*
|
|
33
|
+
* Measuring to the DEADLINE was the first attempt and it was wrong. The
|
|
34
|
+
* binding constraint is `budgetWarningThreshold`, the point at which this
|
|
35
|
+
* guard stops asking for more work and asks for a closing summary — that
|
|
36
|
+
* last slice exists so the run can produce an answer, and a wait sized
|
|
37
|
+
* against the deadline eats into it. Half of the time-to-deadline, started
|
|
38
|
+
* just under the threshold, ends at 95% of the budget: half the closing
|
|
39
|
+
* reserve spent waiting for a result the closing answer was supposed to
|
|
40
|
+
* use.
|
|
41
|
+
*
|
|
42
|
+
* Zero once the threshold has passed, which is also how a caller gets the
|
|
43
|
+
* re-evaluation it needs: `forceFinalize` is sampled at the top of an
|
|
44
|
+
* iteration and this is read when the wait is about to start, so a long
|
|
45
|
+
* iteration that crossed the line in between is told to wait for nothing.
|
|
46
|
+
*
|
|
47
|
+
* Reads through the same `startTime` the limit checks use, so a run
|
|
48
|
+
* resumed from a checkpoint (see `restoreElapsed`) reports the time left
|
|
49
|
+
* on the RUN rather than on the process now hosting it.
|
|
50
|
+
*/
|
|
51
|
+
remainingBeforeFinalizeMs() {
|
|
52
|
+
const finalizeAt = this.limitConfig.timeoutMs * this.limitConfig.budgetWarningThreshold;
|
|
53
|
+
return Math.max(0, finalizeAt - (Date.now() - this.startTime));
|
|
54
|
+
}
|
|
24
55
|
beforeIteration(runMgr, abortSignal) {
|
|
25
56
|
const limitState = {
|
|
26
57
|
aborted: abortSignal.aborted,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"guard.js","sourceRoot":"","sources":["../../../src/runtime/query/guard.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAA;AA0BjF,MAAM,OAAO,gBAAgB;IACpB,WAAW,CAAoB;IAC/B,SAAS,CAAQ;IAEzB,YAAY,MAAmB;QAC9B,IAAI,CAAC,WAAW,GAAG,gBAAgB,CAClC,MAAM,CAAC,WAAW,EAClB,MAAM,CAAC,SAAS,EAChB,MAAM,CAAC,YAAY,EACnB,MAAM,CAAC,aAAa,CACpB,CAAA;QACD,oEAAoE;QACpE,gEAAgE;QAChE,kEAAkE;QAClE,SAAS;QACT,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,eAAe,IAAI,CAAC,CAAC,CAAA;IACvE,CAAC;IAED;;;;;;;OAOG;IACH,cAAc,CAAC,SAAiB;QAC/B,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,CAAA;IACrD,CAAC;IAED,eAAe,CAAC,MAAsB,EAAE,WAAwB;QAC/D,MAAM,UAAU,GAAG;YAClB,OAAO,EAAE,WAAW,CAAC,OAAO;YAC5B,WAAW,EAAE,MAAM,CAAC,UAAU,CAAC,WAAW;YAC1C,SAAS,EAAE,MAAM,CAAC,QAAQ,CAAC,SAAS;YACpC,gBAAgB,EAAE,MAAM,CAAC,gBAAgB;YACzC,SAAS,EAAE,IAAI,CAAC,SAAS;SACzB,CAAA;QAED,MAAM,WAAW,GAAG,mBAAmB,CAAC,IAAI,CAAC,WAAW,EAAE,UAAU,CAAC,CAAA;QAErE,IAAI,WAAW,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YACtC,OAAO;gBACN,UAAU,EAAE,IAAI;gBAChB,aAAa,EAAE,KAAK;gBACpB,UAAU,EAAE,WAAW,CAAC,MAAM;gBAC9B,WAAW,EAAE,WAAW,CAAC,MAAM,KAAK,WAAW;aAC/C,CAAA;QACF,CAAC;QAED,IAAI,WAAW,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YACpC,OAAO;gBACN,UAAU,EAAE,KAAK;gBACjB,aAAa,EAAE,IAAI;gBACnB,UAAU,EAAE,WAAW,CAAC,MAAM;gBAC9B,WAAW,EAAE,KAAK;aAClB,CAAA;QACF,CAAC;QAED,OAAO;YACN,UAAU,EAAE,KAAK;YACjB,aAAa,EAAE,KAAK;YACpB,WAAW,EAAE,KAAK;SAClB,CAAA;IACF,CAAC;CACD"}
|
|
1
|
+
{"version":3,"file":"guard.js","sourceRoot":"","sources":["../../../src/runtime/query/guard.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAA;AA0BjF,MAAM,OAAO,gBAAgB;IACpB,WAAW,CAAoB;IAC/B,SAAS,CAAQ;IAEzB,YAAY,MAAmB;QAC9B,IAAI,CAAC,WAAW,GAAG,gBAAgB,CAClC,MAAM,CAAC,WAAW,EAClB,MAAM,CAAC,SAAS,EAChB,MAAM,CAAC,YAAY,EACnB,MAAM,CAAC,aAAa,CACpB,CAAA;QACD,oEAAoE;QACpE,gEAAgE;QAChE,kEAAkE;QAClE,SAAS;QACT,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,eAAe,IAAI,CAAC,CAAC,CAAA;IACvE,CAAC;IAED;;;;;;;OAOG;IACH,cAAc,CAAC,SAAiB;QAC/B,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,CAAA;IACrD,CAAC;IAED;;;;;;;;;;;;;;;;;;;;;;;;;;OA0BG;IACH,yBAAyB;QACxB,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,CAAC,SAAS,GAAG,IAAI,CAAC,WAAW,CAAC,sBAAsB,CAAA;QACvF,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,CAAA;IAC/D,CAAC;IAED,eAAe,CAAC,MAAsB,EAAE,WAAwB;QAC/D,MAAM,UAAU,GAAG;YAClB,OAAO,EAAE,WAAW,CAAC,OAAO;YAC5B,WAAW,EAAE,MAAM,CAAC,UAAU,CAAC,WAAW;YAC1C,SAAS,EAAE,MAAM,CAAC,QAAQ,CAAC,SAAS;YACpC,gBAAgB,EAAE,MAAM,CAAC,gBAAgB;YACzC,SAAS,EAAE,IAAI,CAAC,SAAS;SACzB,CAAA;QAED,MAAM,WAAW,GAAG,mBAAmB,CAAC,IAAI,CAAC,WAAW,EAAE,UAAU,CAAC,CAAA;QAErE,IAAI,WAAW,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YACtC,OAAO;gBACN,UAAU,EAAE,IAAI;gBAChB,aAAa,EAAE,KAAK;gBACpB,UAAU,EAAE,WAAW,CAAC,MAAM;gBAC9B,WAAW,EAAE,WAAW,CAAC,MAAM,KAAK,WAAW;aAC/C,CAAA;QACF,CAAC;QAED,IAAI,WAAW,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YACpC,OAAO;gBACN,UAAU,EAAE,KAAK;gBACjB,aAAa,EAAE,IAAI;gBACnB,UAAU,EAAE,WAAW,CAAC,MAAM;gBAC9B,WAAW,EAAE,KAAK;aAClB,CAAA;QACF,CAAC;QAED,OAAO;YACN,UAAU,EAAE,KAAK;YACjB,aAAa,EAAE,KAAK;YACpB,WAAW,EAAE,KAAK;SAClB,CAAA;IACF,CAAC;CACD"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"settle-grace.test.d.ts","sourceRoot":"","sources":["../../../../../src/runtime/query/iteration/__tests__/settle-grace.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,226 @@
|
|
|
1
|
+
import { mkdtemp, rm } from 'node:fs/promises';
|
|
2
|
+
import { tmpdir } from 'node:os';
|
|
3
|
+
import { join } from 'node:path';
|
|
4
|
+
import { afterEach, describe, expect, it } from 'vitest';
|
|
5
|
+
import { CompletionInbox } from '../../../../gateway/completion-inbox.js';
|
|
6
|
+
import { ToolRegistry } from '../../../../registry/tool/execute.js';
|
|
7
|
+
import { DELEGATION_TIMEOUT_MS } from '../../../../tools/coordinator/index.js';
|
|
8
|
+
import { createUserMessage } from '../../../../types/message/index.js';
|
|
9
|
+
import { GuardCoordinator } from '../../guard.js';
|
|
10
|
+
import { drainQuery } from '../../index.js';
|
|
11
|
+
import { settleGraceMs } from '../index.js';
|
|
12
|
+
/**
|
|
13
|
+
* How long a finishing run waits for a worker it launched.
|
|
14
|
+
*
|
|
15
|
+
* This used to be a constant — 120 seconds, unrelated to the run holding it.
|
|
16
|
+
* Measured before the change: a run configured `timeoutMs: 20_000` was held
|
|
17
|
+
* open for 120,267 ms, because the hold sits INSIDE an iteration and the guard
|
|
18
|
+
* only checks between them, so nothing could interrupt it. The same constant
|
|
19
|
+
* abandoned workers observed at 4m21s, 5m58s and 8m04s on runs that had hours
|
|
20
|
+
* left.
|
|
21
|
+
*/
|
|
22
|
+
describe('the grace is a share of what the run has left', () => {
|
|
23
|
+
it('takes half, so the turn that reads the result still has time to happen', () => {
|
|
24
|
+
// The fraction is the whole argument. Spending everything remaining
|
|
25
|
+
// would deliver a notification into a run with no turn left to act on
|
|
26
|
+
// it — the failure this mechanism exists to prevent, in a new costume.
|
|
27
|
+
expect(settleGraceMs(60_000)).toBe(30_000);
|
|
28
|
+
});
|
|
29
|
+
it('holds for nothing when the run has nothing left', () => {
|
|
30
|
+
// A decision, not a rounding artefact: with no time left there is no
|
|
31
|
+
// turn in which to read a notification, so waiting can only delay a
|
|
32
|
+
// stop that is already due. Nothing is lost — `waitForArrival` returns
|
|
33
|
+
// before it looks at its timer when a completion is already in hand.
|
|
34
|
+
expect(settleGraceMs(0)).toBe(0);
|
|
35
|
+
});
|
|
36
|
+
it('cannot outlive the deadline it was derived from', () => {
|
|
37
|
+
// Bounded by construction, which is why the guard's inability to
|
|
38
|
+
// interrupt a hold needs no new interrupt seam.
|
|
39
|
+
for (const remaining of [1, 250, 30_000, 600_000, 3_600_000]) {
|
|
40
|
+
expect(settleGraceMs(remaining)).toBeLessThan(remaining);
|
|
41
|
+
}
|
|
42
|
+
});
|
|
43
|
+
it('stops at the longest this subsystem ever waits for a worker', () => {
|
|
44
|
+
// Only reachable for a host whose run timeout exceeds two hours.
|
|
45
|
+
expect(settleGraceMs(10 * 60 * 60 * 1000)).toBe(DELEGATION_TIMEOUT_MS);
|
|
46
|
+
});
|
|
47
|
+
});
|
|
48
|
+
describe('the guard reports time to the finalize point, not to the deadline', () => {
|
|
49
|
+
it('stops short of the closing reserve', () => {
|
|
50
|
+
// 90% of the budget, not 100%. The last tenth is what the guard keeps
|
|
51
|
+
// so a run can produce a closing answer; a wait sized against the
|
|
52
|
+
// deadline spends it on waiting instead.
|
|
53
|
+
const guard = new GuardCoordinator({ tokenBudget: 1_000, timeoutMs: 60_000 });
|
|
54
|
+
const remaining = guard.remainingBeforeFinalizeMs();
|
|
55
|
+
expect(remaining).toBeLessThanOrEqual(54_000);
|
|
56
|
+
expect(remaining).toBeGreaterThan(50_000);
|
|
57
|
+
});
|
|
58
|
+
it('subtracts the time a previous process already spent', () => {
|
|
59
|
+
// A run resumed from a checkpoint gets the remainder of ITS budget, not
|
|
60
|
+
// a fresh clock — otherwise N resumes buy N x timeoutMs, and a hold
|
|
61
|
+
// sized from the fresh clock would outlive the real deadline.
|
|
62
|
+
const guard = new GuardCoordinator({ tokenBudget: 1_000, timeoutMs: 60_000 });
|
|
63
|
+
guard.restoreElapsed(50_000);
|
|
64
|
+
expect(guard.remainingBeforeFinalizeMs()).toBeLessThanOrEqual(4_000);
|
|
65
|
+
expect(guard.remainingBeforeFinalizeMs()).toBeGreaterThan(1_000);
|
|
66
|
+
});
|
|
67
|
+
it('reports nothing once the run is already past the finalize point', () => {
|
|
68
|
+
// 95% elapsed: the guard is about to ask for a closing summary, and a
|
|
69
|
+
// hold opened here would be taken out of the answer's time.
|
|
70
|
+
const guard = new GuardCoordinator({ tokenBudget: 1_000, timeoutMs: 60_000 });
|
|
71
|
+
guard.restoreElapsed(57_000);
|
|
72
|
+
expect(guard.remainingBeforeFinalizeMs()).toBe(0);
|
|
73
|
+
});
|
|
74
|
+
it('never reports a negative remainder', () => {
|
|
75
|
+
const guard = new GuardCoordinator({ tokenBudget: 1_000, timeoutMs: 1_000 });
|
|
76
|
+
guard.restoreElapsed(500_000);
|
|
77
|
+
expect(guard.remainingBeforeFinalizeMs()).toBe(0);
|
|
78
|
+
});
|
|
79
|
+
});
|
|
80
|
+
describe('the hold cannot reach into the run closing reserve', () => {
|
|
81
|
+
it('ends before the finalize point however late it starts', () => {
|
|
82
|
+
// The failure the finalize-relative input exists for. Against the
|
|
83
|
+
// DEADLINE, a hold beginning at elapsed fraction e ends at
|
|
84
|
+
// 0.5 + 0.5e — so one starting just under the 0.9 threshold ends at
|
|
85
|
+
// 95% of the budget, with half the closing reserve gone. Against the
|
|
86
|
+
// finalize point it cannot cross 90% at all.
|
|
87
|
+
const timeoutMs = 60_000;
|
|
88
|
+
const finalizeAt = timeoutMs * 0.9;
|
|
89
|
+
for (const elapsed of [0, 30_000, 50_000, 53_000, 53_900]) {
|
|
90
|
+
const remainingBeforeFinalize = Math.max(0, finalizeAt - elapsed);
|
|
91
|
+
const endsAt = elapsed + settleGraceMs(remainingBeforeFinalize);
|
|
92
|
+
expect(endsAt, `a hold starting at ${elapsed}ms crossed the finalize point`).toBeLessThanOrEqual(finalizeAt);
|
|
93
|
+
}
|
|
94
|
+
});
|
|
95
|
+
});
|
|
96
|
+
const ZERO_USAGE = {
|
|
97
|
+
promptTokens: 0,
|
|
98
|
+
completionTokens: 0,
|
|
99
|
+
totalTokens: 0,
|
|
100
|
+
cachedTokens: 0,
|
|
101
|
+
cacheWriteTokens: 0,
|
|
102
|
+
};
|
|
103
|
+
/** Answers straight away. The hold is the only thing that can delay this run. */
|
|
104
|
+
class AnswersImmediately {
|
|
105
|
+
id = 'answers';
|
|
106
|
+
name = 'Answers Immediately';
|
|
107
|
+
async *chatStream() {
|
|
108
|
+
yield { id: 'm1', delta: { content: 'Done.' } };
|
|
109
|
+
yield { id: 'm1', delta: {}, finishReason: 'stop', usage: ZERO_USAGE };
|
|
110
|
+
}
|
|
111
|
+
}
|
|
112
|
+
const workdirs = [];
|
|
113
|
+
afterEach(async () => {
|
|
114
|
+
await Promise.all(workdirs.map((dir) => rm(dir, { recursive: true, force: true })));
|
|
115
|
+
workdirs.length = 0;
|
|
116
|
+
});
|
|
117
|
+
/**
|
|
118
|
+
* The loop, not the helper.
|
|
119
|
+
*
|
|
120
|
+
* `settleGraceMs` has its own tests above and every one of them passes with
|
|
121
|
+
* the loop still calling a constant. Only a real run can say which number the
|
|
122
|
+
* hold actually used.
|
|
123
|
+
*/
|
|
124
|
+
describe('the hold a run pays is the one its own budget allows', () => {
|
|
125
|
+
async function runHoldingFor(timeoutMs) {
|
|
126
|
+
const workingDirectory = await mkdtemp(join(tmpdir(), 'namzu-grace-'));
|
|
127
|
+
workdirs.push(workingDirectory);
|
|
128
|
+
const inbox = new CompletionInbox();
|
|
129
|
+
// `getTask` because the inbox asks the gateway about a task whose
|
|
130
|
+
// completion may have been announced before the launch was recorded.
|
|
131
|
+
// A double that omits a method the interface requires is not a smaller
|
|
132
|
+
// gateway, it is a broken one.
|
|
133
|
+
inbox.attach({ onTaskCompleted: () => () => { }, getTask: () => undefined });
|
|
134
|
+
// Outstanding and never settling: the worst case the bound exists for.
|
|
135
|
+
// `expect` records the launch as well, so the inbox owns this task.
|
|
136
|
+
inbox.expect('tsk_never');
|
|
137
|
+
const startedAt = Date.now();
|
|
138
|
+
await drainQuery({
|
|
139
|
+
provider: new AnswersImmediately(),
|
|
140
|
+
tools: new ToolRegistry(),
|
|
141
|
+
completionInbox: inbox,
|
|
142
|
+
agentId: 'agent_test',
|
|
143
|
+
agentName: 'Test Agent',
|
|
144
|
+
messages: [createUserMessage('go')],
|
|
145
|
+
workingDirectory,
|
|
146
|
+
runConfig: {
|
|
147
|
+
model: 'mock-model',
|
|
148
|
+
timeoutMs,
|
|
149
|
+
tokenBudget: 100_000,
|
|
150
|
+
maxIterations: 2,
|
|
151
|
+
maxResponseTokens: 256,
|
|
152
|
+
},
|
|
153
|
+
sessionId: 'ses_grace',
|
|
154
|
+
threadId: 'thd_grace',
|
|
155
|
+
projectId: 'prj_grace',
|
|
156
|
+
tenantId: 'tnt_grace',
|
|
157
|
+
});
|
|
158
|
+
return Date.now() - startedAt;
|
|
159
|
+
}
|
|
160
|
+
it('a two-second run does not wait two minutes for a worker', async () => {
|
|
161
|
+
const elapsed = await runHoldingFor(2_000);
|
|
162
|
+
// Half of two seconds, plus whatever the turn itself costs. The number
|
|
163
|
+
// this replaced would have parked here for 120 s regardless.
|
|
164
|
+
expect(elapsed, 'the hold outlived the run budget that bounds it').toBeLessThan(5_000);
|
|
165
|
+
}, 200_000);
|
|
166
|
+
it('a run with no time left does not hold at all', async () => {
|
|
167
|
+
// The guard stops this run at the top of its first iteration, so no
|
|
168
|
+
// hold is even reached — which is the point: the last tenth of a run
|
|
169
|
+
// already never holds, so no artificial minimum has to defend it.
|
|
170
|
+
const elapsed = await runHoldingFor(1);
|
|
171
|
+
expect(elapsed).toBeLessThan(2_000);
|
|
172
|
+
}, 200_000);
|
|
173
|
+
it('still delivers a completion that arrives inside the grace', async () => {
|
|
174
|
+
// The bound must not become an excuse to drop what was nearly ready.
|
|
175
|
+
const workingDirectory = await mkdtemp(join(tmpdir(), 'namzu-grace-hit-'));
|
|
176
|
+
workdirs.push(workingDirectory);
|
|
177
|
+
const inbox = new CompletionInbox();
|
|
178
|
+
let announce;
|
|
179
|
+
inbox.attach({
|
|
180
|
+
onTaskCompleted: (cb) => {
|
|
181
|
+
announce = cb;
|
|
182
|
+
return () => {
|
|
183
|
+
announce = undefined;
|
|
184
|
+
};
|
|
185
|
+
},
|
|
186
|
+
getTask: () => undefined,
|
|
187
|
+
});
|
|
188
|
+
inbox.expect('tsk_soon');
|
|
189
|
+
const timer = setTimeout(() => {
|
|
190
|
+
announce?.({
|
|
191
|
+
taskId: 'tsk_soon',
|
|
192
|
+
agentId: 'reviewer',
|
|
193
|
+
state: 'completed',
|
|
194
|
+
createdAt: 0,
|
|
195
|
+
completedAt: 1,
|
|
196
|
+
result: { status: 'completed', result: 'ARRIVED INSIDE THE GRACE' },
|
|
197
|
+
});
|
|
198
|
+
}, 150);
|
|
199
|
+
timer.unref?.();
|
|
200
|
+
const run = await drainQuery({
|
|
201
|
+
provider: new AnswersImmediately(),
|
|
202
|
+
tools: new ToolRegistry(),
|
|
203
|
+
completionInbox: inbox,
|
|
204
|
+
agentId: 'agent_test',
|
|
205
|
+
agentName: 'Test Agent',
|
|
206
|
+
messages: [createUserMessage('go')],
|
|
207
|
+
workingDirectory,
|
|
208
|
+
runConfig: {
|
|
209
|
+
model: 'mock-model',
|
|
210
|
+
timeoutMs: 30_000,
|
|
211
|
+
tokenBudget: 100_000,
|
|
212
|
+
maxIterations: 3,
|
|
213
|
+
maxResponseTokens: 256,
|
|
214
|
+
},
|
|
215
|
+
sessionId: 'ses_grace',
|
|
216
|
+
threadId: 'thd_grace',
|
|
217
|
+
projectId: 'prj_grace',
|
|
218
|
+
tenantId: 'tnt_grace',
|
|
219
|
+
});
|
|
220
|
+
const userText = run.messages
|
|
221
|
+
.filter((m) => m.role === 'user')
|
|
222
|
+
.map((m) => (typeof m.content === 'string' ? m.content : ''));
|
|
223
|
+
expect(userText.some((m) => m.includes('ARRIVED INSIDE THE GRACE'))).toBe(true);
|
|
224
|
+
}, 200_000);
|
|
225
|
+
});
|
|
226
|
+
//# sourceMappingURL=settle-grace.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"settle-grace.test.js","sourceRoot":"","sources":["../../../../../src/runtime/query/iteration/__tests__/settle-grace.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,kBAAkB,CAAA;AAC9C,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAA;AAChC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAA;AAChC,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAA;AAExD,OAAO,EAAE,eAAe,EAAE,MAAM,yCAAyC,CAAA;AACzE,OAAO,EAAE,YAAY,EAAE,MAAM,sCAAsC,CAAA;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,wCAAwC,CAAA;AAG9E,OAAO,EAAE,iBAAiB,EAAE,MAAM,oCAAoC,CAAA;AAGtE,OAAO,EAAE,gBAAgB,EAAE,MAAM,gBAAgB,CAAA;AACjD,OAAO,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAC3C,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAA;AAE3C;;;;;;;;;GASG;AAEH,QAAQ,CAAC,+CAA+C,EAAE,GAAG,EAAE;IAC9D,EAAE,CAAC,wEAAwE,EAAE,GAAG,EAAE;QACjF,oEAAoE;QACpE,sEAAsE;QACtE,uEAAuE;QACvE,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;IAC3C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QAC1D,qEAAqE;QACrE,oEAAoE;QACpE,uEAAuE;QACvE,qEAAqE;QACrE,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;IACjC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QAC1D,iEAAiE;QACjE,gDAAgD;QAChD,KAAK,MAAM,SAAS,IAAI,CAAC,CAAC,EAAE,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,CAAC,EAAE,CAAC;YAC9D,MAAM,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,SAAS,CAAC,CAAA;QACzD,CAAC;IACF,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,6DAA6D,EAAE,GAAG,EAAE;QACtE,iEAAiE;QACjE,MAAM,CAAC,aAAa,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAA;IACvE,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA;AAEF,QAAQ,CAAC,mEAAmE,EAAE,GAAG,EAAE;IAClF,EAAE,CAAC,oCAAoC,EAAE,GAAG,EAAE;QAC7C,sEAAsE;QACtE,kEAAkE;QAClE,yCAAyC;QACzC,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAA;QAE7E,MAAM,SAAS,GAAG,KAAK,CAAC,yBAAyB,EAAE,CAAA;QAEnD,MAAM,CAAC,SAAS,CAAC,CAAC,mBAAmB,CAAC,MAAM,CAAC,CAAA;QAC7C,MAAM,CAAC,SAAS,CAAC,CAAC,eAAe,CAAC,MAAM,CAAC,CAAA;IAC1C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,qDAAqD,EAAE,GAAG,EAAE;QAC9D,wEAAwE;QACxE,oEAAoE;QACpE,8DAA8D;QAC9D,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAA;QAC7E,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC,CAAA;QAE5B,MAAM,CAAC,KAAK,CAAC,yBAAyB,EAAE,CAAC,CAAC,mBAAmB,CAAC,KAAK,CAAC,CAAA;QACpE,MAAM,CAAC,KAAK,CAAC,yBAAyB,EAAE,CAAC,CAAC,eAAe,CAAC,KAAK,CAAC,CAAA;IACjE,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,iEAAiE,EAAE,GAAG,EAAE;QAC1E,sEAAsE;QACtE,4DAA4D;QAC5D,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAA;QAC7E,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC,CAAA;QAE5B,MAAM,CAAC,KAAK,CAAC,yBAAyB,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;IAClD,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,oCAAoC,EAAE,GAAG,EAAE;QAC7C,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,CAAA;QAC5E,KAAK,CAAC,cAAc,CAAC,OAAO,CAAC,CAAA;QAE7B,MAAM,CAAC,KAAK,CAAC,yBAAyB,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;IAClD,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA;AAEF,QAAQ,CAAC,oDAAoD,EAAE,GAAG,EAAE;IACnE,EAAE,CAAC,uDAAuD,EAAE,GAAG,EAAE;QAChE,kEAAkE;QAClE,2DAA2D;QAC3D,oEAAoE;QACpE,qEAAqE;QACrE,6CAA6C;QAC7C,MAAM,SAAS,GAAG,MAAM,CAAA;QACxB,MAAM,UAAU,GAAG,SAAS,GAAG,GAAG,CAAA;QAElC,KAAK,MAAM,OAAO,IAAI,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC;YAC3D,MAAM,uBAAuB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,OAAO,CAAC,CAAA;YACjE,MAAM,MAAM,GAAG,OAAO,GAAG,aAAa,CAAC,uBAAuB,CAAC,CAAA;YAE/D,MAAM,CACL,MAAM,EACN,sBAAsB,OAAO,+BAA+B,CAC5D,CAAC,mBAAmB,CAAC,UAAU,CAAC,CAAA;QAClC,CAAC;IACF,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA;AAEF,MAAM,UAAU,GAAG;IAClB,YAAY,EAAE,CAAC;IACf,gBAAgB,EAAE,CAAC;IACnB,WAAW,EAAE,CAAC;IACd,YAAY,EAAE,CAAC;IACf,gBAAgB,EAAE,CAAC;CACnB,CAAA;AAED,iFAAiF;AACjF,MAAM,kBAAkB;IACd,EAAE,GAAG,SAAS,CAAA;IACd,IAAI,GAAG,qBAAqB,CAAA;IACrC,KAAK,CAAC,CAAC,UAAU;QAChB,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAA;QAC/C,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,EAAE,YAAY,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,CAAA;IACvE,CAAC;CACD;AAED,MAAM,QAAQ,GAAa,EAAE,CAAA;AAC7B,SAAS,CAAC,KAAK,IAAI,EAAE;IACpB,MAAM,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAA;IACnF,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAA;AACpB,CAAC,CAAC,CAAA;AAEF;;;;;;GAMG;AACH,QAAQ,CAAC,sDAAsD,EAAE,GAAG,EAAE;IACrE,KAAK,UAAU,aAAa,CAAC,SAAiB;QAC7C,MAAM,gBAAgB,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,cAAc,CAAC,CAAC,CAAA;QACtE,QAAQ,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAA;QAE/B,MAAM,KAAK,GAAG,IAAI,eAAe,EAAE,CAAA;QACnC,kEAAkE;QAClE,qEAAqE;QACrE,uEAAuE;QACvE,+BAA+B;QAC/B,KAAK,CAAC,MAAM,CAAC,EAAE,eAAe,EAAE,GAAG,EAAE,CAAC,GAAG,EAAE,GAAE,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,SAAS,EAAW,CAAC,CAAA;QACpF,uEAAuE;QACvE,oEAAoE;QACpE,KAAK,CAAC,MAAM,CAAC,WAAqB,CAAC,CAAA;QAEnC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;QAC5B,MAAM,UAAU,CAAC;YAChB,QAAQ,EAAE,IAAI,kBAAkB,EAAE;YAClC,KAAK,EAAE,IAAI,YAAY,EAAE;YACzB,eAAe,EAAE,KAAK;YACtB,OAAO,EAAE,YAAY;YACrB,SAAS,EAAE,YAAY;YACvB,QAAQ,EAAE,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC;YACnC,gBAAgB;YAChB,SAAS,EAAE;gBACV,KAAK,EAAE,YAAY;gBACnB,SAAS;gBACT,WAAW,EAAE,OAAO;gBACpB,aAAa,EAAE,CAAC;gBAChB,iBAAiB,EAAE,GAAG;aACtB;YACD,SAAS,EAAE,WAAwB;YACnC,QAAQ,EAAE,WAAuB;YACjC,SAAS,EAAE,WAAwB;YACnC,QAAQ,EAAE,WAAuB;SACxB,CAAC,CAAA;QACX,OAAO,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAA;IAC9B,CAAC;IAED,EAAE,CAAC,yDAAyD,EAAE,KAAK,IAAI,EAAE;QACxE,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC,KAAK,CAAC,CAAA;QAE1C,uEAAuE;QACvE,6DAA6D;QAC7D,MAAM,CAAC,OAAO,EAAE,iDAAiD,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAA;IACvF,CAAC,EAAE,OAAO,CAAC,CAAA;IAEX,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC7D,oEAAoE;QACpE,qEAAqE;QACrE,kEAAkE;QAClE,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC,CAAC,CAAC,CAAA;QAEtC,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAA;IACpC,CAAC,EAAE,OAAO,CAAC,CAAA;IAEX,EAAE,CAAC,2DAA2D,EAAE,KAAK,IAAI,EAAE;QAC1E,qEAAqE;QACrE,MAAM,gBAAgB,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,kBAAkB,CAAC,CAAC,CAAA;QAC1E,QAAQ,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAA;QAE/B,MAAM,KAAK,GAAG,IAAI,eAAe,EAAE,CAAA;QACnC,IAAI,QAA+C,CAAA;QACnD,KAAK,CAAC,MAAM,CAAC;YACZ,eAAe,EAAE,CAAC,EAA2B,EAAE,EAAE;gBAChD,QAAQ,GAAG,EAAE,CAAA;gBACb,OAAO,GAAG,EAAE;oBACX,QAAQ,GAAG,SAAS,CAAA;gBACrB,CAAC,CAAA;YACF,CAAC;YACD,OAAO,EAAE,GAAG,EAAE,CAAC,SAAS;SACf,CAAC,CAAA;QACX,KAAK,CAAC,MAAM,CAAC,UAAoB,CAAC,CAAA;QAClC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC7B,QAAQ,EAAE,CAAC;gBACV,MAAM,EAAE,UAAoB;gBAC5B,OAAO,EAAE,UAAU;gBACnB,KAAK,EAAE,WAAW;gBAClB,SAAS,EAAE,CAAC;gBACZ,WAAW,EAAE,CAAC;gBACd,MAAM,EAAE,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,0BAA0B,EAAE;aACrD,CAAC,CAAA;QACjB,CAAC,EAAE,GAAG,CAAC,CAAA;QACP,KAAK,CAAC,KAAK,EAAE,EAAE,CAAA;QAEf,MAAM,GAAG,GAAG,MAAM,UAAU,CAAC;YAC5B,QAAQ,EAAE,IAAI,kBAAkB,EAAE;YAClC,KAAK,EAAE,IAAI,YAAY,EAAE;YACzB,eAAe,EAAE,KAAK;YACtB,OAAO,EAAE,YAAY;YACrB,SAAS,EAAE,YAAY;YACvB,QAAQ,EAAE,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC;YACnC,gBAAgB;YAChB,SAAS,EAAE;gBACV,KAAK,EAAE,YAAY;gBACnB,SAAS,EAAE,MAAM;gBACjB,WAAW,EAAE,OAAO;gBACpB,aAAa,EAAE,CAAC;gBAChB,iBAAiB,EAAE,GAAG;aACtB;YACD,SAAS,EAAE,WAAwB;YACnC,QAAQ,EAAE,WAAuB;YACjC,SAAS,EAAE,WAAwB;YACnC,QAAQ,EAAE,WAAuB;SACxB,CAAC,CAAA;QAEX,MAAM,QAAQ,GAAI,GAAG,CAAC,QAAiD;aACrE,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC;aAChC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;QAE9D,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IAChF,CAAC,EAAE,OAAO,CAAC,CAAA;AACZ,CAAC,CAAC,CAAA"}
|
|
@@ -4,10 +4,62 @@ import type { IterationContext } from './phases/index.js';
|
|
|
4
4
|
export type { IterationContext } from './phases/index.js';
|
|
5
5
|
export type { PhaseSignal } from './phases/index.js';
|
|
6
6
|
export type { ToolReviewOutcome } from './phases/index.js';
|
|
7
|
+
/**
|
|
8
|
+
* How long a finishing run waits for a background worker it launched.
|
|
9
|
+
*
|
|
10
|
+
* Derived from the run rather than fixed, because a constant is wrong in both
|
|
11
|
+
* directions at once. The 120 seconds this replaces held a run configured for
|
|
12
|
+
* a twenty-second timeout open for 120,267 ms — six times its own budget, and
|
|
13
|
+
* unreachable by the guard, which only checks between iterations — while on an
|
|
14
|
+
* hour-long run it abandoned workers measured at 4m21s, 5m58s and 8m04s, all
|
|
15
|
+
* of them well inside the hour the delegation tools themselves declare.
|
|
16
|
+
*
|
|
17
|
+
* **Bounded by construction, and against the right boundary.** The input is
|
|
18
|
+
* time-to-FINALIZE, not time-to-deadline (see
|
|
19
|
+
* `GuardCoordinator.remainingBeforeFinalizeMs`). Measuring to the deadline was
|
|
20
|
+
* the first attempt and it was wrong in a way that looked safe: a hold cannot
|
|
21
|
+
* outlive the deadline either way, but half of the time-to-deadline started
|
|
22
|
+
* just under the warning threshold ends at 95% of the budget — so the slice
|
|
23
|
+
* that exists for the run to produce a closing answer is half spent waiting
|
|
24
|
+
* for the result that answer was supposed to use. Against the finalize point
|
|
25
|
+
* the hold cannot reach the reserve at all, which is what makes the guard's
|
|
26
|
+
* inability to interrupt a hold a non-issue rather than a smaller issue.
|
|
27
|
+
*
|
|
28
|
+
* **The floor of zero is a decision, not a clamp artefact.** A run with no
|
|
29
|
+
* time left before it must start finishing has no turn in which to read a
|
|
30
|
+
* notification, so waiting could only delay a stop that is already due.
|
|
31
|
+
* Nothing is lost by it: `CompletionInbox.waitForArrival` returns before it
|
|
32
|
+
* looks at its timer when a completion is already in hand, so a zero grace
|
|
33
|
+
* still delivers everything that has arrived. No minimum is invented on top,
|
|
34
|
+
* because zero is exactly what a run past the threshold should wait — and
|
|
35
|
+
* reading the remainder at hold time rather than trusting `forceFinalize`,
|
|
36
|
+
* which is sampled at the top of the iteration, is what makes a long iteration
|
|
37
|
+
* that crossed the line in between compute it.
|
|
38
|
+
*
|
|
39
|
+
* **The ceiling is the longest anything in this subsystem waits for a
|
|
40
|
+
* delegated worker.** It binds only for a host whose run timeout exceeds
|
|
41
|
+
* roughly two and a quarter hours; below that the fraction is smaller.
|
|
42
|
+
*/
|
|
43
|
+
export declare function settleGraceMs(remainingBeforeFinalizeMs: number): number;
|
|
7
44
|
export declare class IterationOrchestrator {
|
|
8
45
|
private ctx;
|
|
9
46
|
/** Rejections so far. See {@link DEFAULT_ANSWER_REVIEW_LIMIT}. */
|
|
10
47
|
private answerReviewAttempts;
|
|
48
|
+
/**
|
|
49
|
+
* The previous iteration held a `stopWhen` decision open for a worker.
|
|
50
|
+
*
|
|
51
|
+
* Set when the stop predicate fired and the run took one extra turn to
|
|
52
|
+
* read a delegated result, so the turn that then ends the run can report
|
|
53
|
+
* WHY it is over. Without it the outcome was right and the record was
|
|
54
|
+
* wrong: the run stopped because the host said so and reported `end_turn`,
|
|
55
|
+
* and this repo carries thirteen `StopReason` values precisely so that a
|
|
56
|
+
* run which ends for a nameable reason names it.
|
|
57
|
+
*
|
|
58
|
+
* Lives for exactly one iteration — see the read-and-clear at the top of
|
|
59
|
+
* the loop, which is the only site that touches it besides the one that
|
|
60
|
+
* sets it.
|
|
61
|
+
*/
|
|
62
|
+
private stopDeferredForOutstandingWork;
|
|
11
63
|
constructor(ctx: IterationContext);
|
|
12
64
|
/**
|
|
13
65
|
* Adopt the run's span after construction.
|
|
@@ -18,6 +70,46 @@ export declare class IterationOrchestrator {
|
|
|
18
70
|
*/
|
|
19
71
|
setRootSpan(span: Span): void;
|
|
20
72
|
runLoop(): AsyncGenerator<RunEvent>;
|
|
73
|
+
/**
|
|
74
|
+
* Hold the run open for a worker that has not finished, and deliver it.
|
|
75
|
+
*
|
|
76
|
+
* Returns whether a completion arrived and was put in the transcript — the
|
|
77
|
+
* caller continues the loop on `true`, so the model gets a turn in which to
|
|
78
|
+
* USE the result. That turn is the entire justification for waiting, which
|
|
79
|
+
* is why only the exits that can still take one call this.
|
|
80
|
+
*
|
|
81
|
+
* Bounded by `settleGraceMs` and by `maxIterations`, so a worker that never
|
|
82
|
+
* finishes cannot keep the run open.
|
|
83
|
+
*/
|
|
84
|
+
private holdForOutstandingWork;
|
|
85
|
+
/**
|
|
86
|
+
* Account for delegated work on the way out: deliver what arrived, and say
|
|
87
|
+
* what did not.
|
|
88
|
+
*
|
|
89
|
+
* A run that ends with a worker outstanding must not leave the impression
|
|
90
|
+
* that the worker's result was delivered. There are exactly two honest
|
|
91
|
+
* outcomes and this does both:
|
|
92
|
+
*
|
|
93
|
+
* - **What has already arrived is delivered.** It makes no false claim,
|
|
94
|
+
* and dropping it is pure loss — the message rides out on
|
|
95
|
+
* `Run.messages`, so a host reads it and the next turn of a continued
|
|
96
|
+
* thread starts with it. This does NOT wait: a hold buys the model a
|
|
97
|
+
* turn in which to USE a result, and on an exit whose answer is already
|
|
98
|
+
* decided there is no such turn, so waiting would delay a settled answer
|
|
99
|
+
* to append text this run will not read. The bounded hold stays where it
|
|
100
|
+
* was, on the exits that do have a turn left.
|
|
101
|
+
* - **What is still running is NAMED, not cancelled.** Giving up on a wait
|
|
102
|
+
* is a statement about the waiter, not about the work — the rule
|
|
103
|
+
* `wait-with-idle-bound.ts` already states for the same subsystem — and
|
|
104
|
+
* "the parent answered early" is a weaker warrant for killing a child
|
|
105
|
+
* than "the clock ran out", not a stronger one. Killing a worker that
|
|
106
|
+
* may be mid-write is a policy only the host can judge, and it has
|
|
107
|
+
* `cancel_task` and the run controller to judge it with.
|
|
108
|
+
*/
|
|
109
|
+
private settleOutstandingWork;
|
|
110
|
+
/** Delegated work this run walked away from. See {@link settleOutstandingWork}. */
|
|
111
|
+
private recordAbandonedWork;
|
|
112
|
+
private deliverArrivedCompletions;
|
|
21
113
|
/**
|
|
22
114
|
* Ask the host how to shape this step.
|
|
23
115
|
*
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/runtime/query/iteration/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,IAAI,EAAkB,MAAM,oBAAoB,CAAA;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/runtime/query/iteration/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,IAAI,EAAkB,MAAM,oBAAoB,CAAA;AA+B9D,OAAO,KAAK,EAEX,QAAQ,EACR,UAAU,EAEV,MAAM,6BAA6B,CAAA;AAUpC,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AAMzD,YAAY,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AACzD,YAAY,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAA;AACpD,YAAY,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAA;AA+B1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AACH,wBAAgB,aAAa,CAAC,yBAAyB,EAAE,MAAM,GAAG,MAAM,CAKvE;AAED,qBAAa,qBAAqB;IACjC,OAAO,CAAC,GAAG,CAAkB;IAC7B,kEAAkE;IAClE,OAAO,CAAC,oBAAoB,CAAI;IAChC;;;;;;;;;;;;;OAaG;IACH,OAAO,CAAC,8BAA8B,CAAQ;gBAElC,GAAG,EAAE,gBAAgB;IAIjC;;;;;;OAMG;IACH,WAAW,CAAC,IAAI,EAAE,IAAI,GAAG,IAAI;IAItB,OAAO,IAAI,cAAc,CAAC,QAAQ,CAAC;IA4yB1C;;;;;;;;;;OAUG;YACY,sBAAsB;IA+BrC;;;;;;;;;;;;;;;;;;;;;;;OAuBG;IACH,OAAO,CAAC,qBAAqB;IAK7B,mFAAmF;IACnF,OAAO,CAAC,mBAAmB;IAW3B,OAAO,CAAC,yBAAyB;IA8BjC;;;;;;;OAOG;YACW,WAAW;IAyFzB,6DAA6D;IAC7D,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAmB;IAEzC,QAAQ,IAAI,SAAS,UAAU,EAAE;IAIjC;;;;;;;OAOG;IACH,OAAO,CAAC,UAAU;IAgDlB,wEAAwE;IACxE,OAAO,CAAC,wBAAwB,CAAI;IACpC,OAAO,CAAC,oBAAoB,CAAQ;IAEpC,OAAO,CAAC,0BAA0B;IAIlC,uEAAuE;IACvE,OAAO,CAAC,qBAAqB;IAI7B;;;;;;;OAOG;IACH;;;;;;;;;;;OAWG;IACH,OAAO,CAAC,kBAAkB;IA4B1B,OAAO,CAAC,uBAAuB;IAe/B;;;;;;;;;;;OAWG;YACW,YAAY;IAiB1B,6DAA6D;YAC/C,UAAU;IAsBxB,OAAO,CAAC,oBAAoB;YAOd,oBAAoB;CA6ElC"}
|