|
| 1 | +// Coverage for handing replay-safe plugin-harness prompt timeouts to model fallback. |
| 2 | +import { beforeAll, beforeEach, describe, expect, it } from "vitest"; |
| 3 | +import { makeModelFallbackCfg } from "../test-helpers/model-fallback-config-fixture.js"; |
| 4 | +import { makeAttemptResult } from "./run.overflow-compaction.fixture.js"; |
| 5 | +import { |
| 6 | + loadRunOverflowCompactionHarness, |
| 7 | + MockedFailoverError, |
| 8 | + mockedClassifyFailoverReason, |
| 9 | + mockedRunEmbeddedAttempt, |
| 10 | + overflowBaseRunParams, |
| 11 | + resetRunOverflowCompactionHarnessMocks, |
| 12 | +} from "./run.overflow-compaction.harness.js"; |
| 13 | + |
| 14 | +let runEmbeddedAgent: typeof import("./run.js").runEmbeddedAgent; |
| 15 | + |
| 16 | +describe("runEmbeddedAgent prompt timeout fallback handoff", () => { |
| 17 | + beforeAll(async () => { |
| 18 | + ({ runEmbeddedAgent } = await loadRunOverflowCompactionHarness()); |
| 19 | + }); |
| 20 | + |
| 21 | + beforeEach(() => { |
| 22 | + resetRunOverflowCompactionHarnessMocks(); |
| 23 | + }); |
| 24 | + |
| 25 | + it("throws FailoverError for replay-safe harness-owned prompt timeouts when model fallbacks are configured", async () => { |
| 26 | + mockedClassifyFailoverReason.mockReturnValue("timeout"); |
| 27 | + mockedRunEmbeddedAttempt.mockResolvedValueOnce( |
| 28 | + makeAttemptResult({ |
| 29 | + assistantTexts: [], |
| 30 | + promptError: new Error("LLM request timed out."), |
| 31 | + promptErrorSource: "prompt", |
| 32 | + }), |
| 33 | + ); |
| 34 | + |
| 35 | + const promise = runEmbeddedAgent({ |
| 36 | + ...overflowBaseRunParams, |
| 37 | + provider: "openai", |
| 38 | + model: "gpt-5.4", |
| 39 | + runId: "run-prompt-timeout-fallback", |
| 40 | + config: makeModelFallbackCfg({ |
| 41 | + agents: { |
| 42 | + defaults: { |
| 43 | + model: { |
| 44 | + primary: "openai/gpt-5.4", |
| 45 | + fallbacks: ["anthropic/claude-opus-4-6"], |
| 46 | + }, |
| 47 | + }, |
| 48 | + }, |
| 49 | + }), |
| 50 | + }); |
| 51 | + |
| 52 | + await expect(promise).rejects.toBeInstanceOf(MockedFailoverError); |
| 53 | + await expect(promise).rejects.toThrow("LLM request timed out."); |
| 54 | + expect(mockedRunEmbeddedAttempt).toHaveBeenCalledTimes(1); |
| 55 | + }); |
| 56 | + |
| 57 | + it("surfaces replay-invalid prompt timeouts instead of handing them to model fallback", async () => { |
| 58 | + mockedClassifyFailoverReason.mockReturnValue("timeout"); |
| 59 | + mockedRunEmbeddedAttempt.mockResolvedValueOnce( |
| 60 | + makeAttemptResult({ |
| 61 | + assistantTexts: [], |
| 62 | + promptError: new Error("LLM request timed out."), |
| 63 | + promptErrorSource: "prompt", |
| 64 | + promptTimeoutOutcome: { |
| 65 | + message: "Harness abandoned the timed-out turn after provider activity.", |
| 66 | + replayInvalid: true, |
| 67 | + livenessState: "abandoned", |
| 68 | + }, |
| 69 | + }), |
| 70 | + ); |
| 71 | + |
| 72 | + let thrown: unknown; |
| 73 | + try { |
| 74 | + await runEmbeddedAgent({ |
| 75 | + ...overflowBaseRunParams, |
| 76 | + provider: "openai", |
| 77 | + model: "gpt-5.4", |
| 78 | + runId: "run-prompt-timeout-replay-invalid", |
| 79 | + config: makeModelFallbackCfg({ |
| 80 | + agents: { |
| 81 | + defaults: { |
| 82 | + model: { |
| 83 | + primary: "openai/gpt-5.4", |
| 84 | + fallbacks: ["anthropic/claude-opus-4-6"], |
| 85 | + }, |
| 86 | + }, |
| 87 | + }, |
| 88 | + }), |
| 89 | + }); |
| 90 | + } catch (err) { |
| 91 | + thrown = err; |
| 92 | + } |
| 93 | + |
| 94 | + expect(thrown).toBeInstanceOf(Error); |
| 95 | + expect(thrown).not.toBeInstanceOf(MockedFailoverError); |
| 96 | + expect(String((thrown as Error | undefined)?.message)).toContain("LLM request timed out."); |
| 97 | + expect(mockedRunEmbeddedAttempt).toHaveBeenCalledTimes(1); |
| 98 | + }); |
| 99 | +}); |
0 commit comments