t3-code-android-nightly/.repos/alchemy-effect/packages/alchemy/test/AWS/Bedrock/Bindings.test.ts

import * as AWS from "@/AWS";
import * as Core from "@/Test/Core";
import * as Test from "@/Test/Alchemy";
import { describe, expect } from "alchemy-test";
import * as Data from "effect/Data";
import * as Effect from "effect/Effect";
import * as Schedule from "effect/Schedule";
import * as HttpClient from "effect/http/HttpClient";
import * as HttpClientRequest from "effect/http/HttpClientRequest";
import BedrockTestFunctionLive, { BedrockTestFunction } from "./handler";

const testOptions = { providers: AWS.providers() };
const { test, beforeAll, afterAll } = Test.make(testOptions);
const sharedStack = Core.scratchStack(testOptions, "BedrockBindings");

// Lambda function URL cold-start (DNS, IAM propagation, init) can take well
// over 60s on a fresh deploy under parallel-suite load. Budget ~150s of
// readiness polling.
const readinessPolicy = Schedule.max([
  Schedule.fixed("2 seconds"),
  Schedule.recurs(75),
]);

let baseUrl: string;

class TransientUpstream extends Data.TaggedError("TransientUpstream")<{
  readonly status: number;
  readonly body: string;
}> {}

// The shared Lambda fixture occasionally answers a transient 5xx under
// parallel load (cold re-init, IAM propagation, Bedrock throttling surfaced
// as a defect). Retry 5xx only; a genuine 4xx/assertion failure surfaces
// immediately.
const send = (request: HttpClientRequest.HttpClientRequest) =>
  HttpClient.execute(request).pipe(
    Effect.flatMap((response) =>
      response.status >= 500
        ? response.text.pipe(
            Effect.flatMap((body) =>
              Effect.fail(
                new TransientUpstream({ status: response.status, body }),
              ),
            ),
          )
        : Effect.succeed(response),
    ),
    Effect.retry({
      while: (e) => e._tag === "TransientUpstream",
      schedule: Schedule.max([
        Schedule.exponential("1 second"),
        Schedule.recurs(5),
      ]),
    }),
  );

describe(
  "Bedrock Bindings",
  {
    tags: [
      "provider:aws",
      "provider:aws:bedrock",
      "provider:aws:lambda",
      "live",
    ],
  },
  () => {
    beforeAll(
      Effect.gen(function* () {
        yield* Effect.logInfo(
          "Bedrock test setup: destroying previous resources",
        );
        yield* sharedStack.destroy();

        yield* Effect.logInfo("Bedrock test setup: deploying fixture");
        const { functionUrl } = yield* sharedStack.deploy(
          Effect.gen(function* () {
            return yield* BedrockTestFunction;
          }).pipe(Effect.provide(BedrockTestFunctionLive)),
        );

        expect(functionUrl).toBeTruthy();
        baseUrl = functionUrl!.replace(/\/+$/, "");
        const readinessUrl = `${baseUrl}/ping`;

        yield* Effect.logInfo(
          `Bedrock test setup: probing readiness at ${readinessUrl}`,
        );
        yield* HttpClient.get(readinessUrl).pipe(
          Effect.flatMap((response) =>
            response.status === 200
              ? Effect.succeed(response)
              : Effect.fail(
                  new Error(`Function not ready: ${response.status}`),
                ),
          ),
          Effect.tapError((error) =>
            Effect.logWarning(
              `Bedrock test setup: fixture not ready yet (${String(error)})`,
            ),
          ),
          Effect.retry({ schedule: readinessPolicy }),
        );
      }),
      { timeout: 240_000 },
    );

    afterAll.skipIf(!!process.env.NO_DESTROY)(sharedStack.destroy(), {
      timeout: 120_000,
    });

    describe("CountTokens", () => {
      test.provider(
        "counts input tokens without invoking the model",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/count-tokens`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              inputTokens: number;
            };

            expect(response.inputTokens).toBeGreaterThan(0);
          }),
        { timeout: 120_000 },
      );
    });

    describe("Rerank", () => {
      test.provider(
        "reranks inline documents by relevance to the query",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/rerank`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              results: Array<{ index: number; relevanceScore: number }>;
            };

            expect(response.results).toHaveLength(2);
            // The alchemy document must outrank the banana document.
            expect(response.results[0]!.index).toBe(0);
            expect(response.results[0]!.relevanceScore).toBeGreaterThan(
              response.results[1]!.relevanceScore,
            );
          }),
        { timeout: 120_000 },
      );
    });

    describe("InvokeAgent", () => {
      test.provider(
        "invokes the bound agent alias and streams a completion",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/invoke-agent`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              sessionId: string;
              chunkEvents: number;
              text: string;
            };

            expect(response.sessionId.length).toBeGreaterThan(0);
            expect(response.chunkEvents).toBeGreaterThan(0);
            // Assert a non-empty completion — never exact model output.
            expect(response.text.trim().length).toBeGreaterThan(0);
          }),
        { timeout: 120_000 },
      );
    });

    describe("GetAgentMemory + DeleteAgentMemory", () => {
      test.provider(
        "reads and clears the agent's long-term memory for a memory id",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/agent-memory`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              memoryContents: number;
              deleted: boolean;
            };

            // Session summaries generate asynchronously long after a session
            // ends, so a fresh agent has none — the route proves the IAM
            // grants and the read+delete round-trip, not summary generation.
            expect(response.memoryContents).toBeGreaterThanOrEqual(0);
            expect(response.deleted).toBe(true);
          }),
        { timeout: 120_000 },
      );
    });

    describe("Converse", () => {
      test.provider(
        "converses with the bound model and returns a completion",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/converse`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              text: string;
              stopReason: string;
              outputTokens: number;
            };

            // Assert a non-empty completion — never exact model output (LLMs
            // are nondeterministic even at temperature 0).
            expect(response.text.trim().length).toBeGreaterThan(0);
            expect(["end_turn", "max_tokens"]).toContain(response.stopReason);
            expect(response.outputTokens).toBeGreaterThan(0);
          }),
        { timeout: 120_000 },
      );
    });

    describe("ConverseStream", () => {
      test.provider(
        "streams a conversation as typed events",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/converse-stream`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              text: string;
              deltaEvents: number;
              totalEvents: number;
              stopReason: string | undefined;
            };

            // Assert a non-empty streamed completion — never exact model output.
            expect(response.text.trim().length).toBeGreaterThan(0);
            expect(response.deltaEvents).toBeGreaterThan(0);
            expect(response.totalEvents).toBeGreaterThan(response.deltaEvents);
            expect(["end_turn", "max_tokens"]).toContain(response.stopReason);
          }),
        { timeout: 120_000 },
      );
    });

    describe("InvokeModelWithResponseStream", () => {
      test.provider(
        "streams raw payload chunks from the bound model",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/invoke-model-stream`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              contentType: string;
              chunkEvents: number;
              text: string;
            };

            expect(response.contentType).toBe("application/json");
            expect(response.chunkEvents).toBeGreaterThan(0);
            // Assert a non-empty streamed completion — never exact model output.
            expect(response.text.trim().length).toBeGreaterThan(0);
          }),
        { timeout: 120_000 },
      );
    });

    describe("InvokeModel", () => {
      test.provider(
        "invokes the bound model with a raw payload and streams the response",
        (_stack) =>
          Effect.gen(function* () {
            const response = (yield* send(
              HttpClientRequest.get(`${baseUrl}/invoke-model`),
            ).pipe(Effect.flatMap((r) => r.json))) as {
              contentType: string;
              text: string;
              stopReason: string;
            };

            expect(response.contentType).toBe("application/json");
            // Assert a non-empty completion — never exact model output.
            expect(response.text.trim().length).toBeGreaterThan(0);
            expect(["end_turn", "max_tokens"]).toContain(response.stopReason);
          }),
        { timeout: 120_000 },
      );
    });
  },
);