Skip to content

An open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作

14.6 KB
Raw Blame
// SiYuan - https://github.com/siyuan-note/siyuan
// Copyright (c) 2020-present, b3log.org
// SPDX-License-Identifier: AGPL-3.0-or-later

package agent

import (
	"context"
	"errors"
	"fmt"
	"io"
	"net/http"
	"net/http/httptest"
	"strings"
	"sync/atomic"
	"testing"
	"time"

	openai "github.com/sashabaranov/go-openai"
	kernelConf "github.com/siyuan-note/siyuan/kernel/conf"
	kernelModel "github.com/siyuan-note/siyuan/kernel/model"
	"github.com/siyuan-note/siyuan/kernel/util"
)

func TestStreamIdleTimeoutResetsAfterEachChunk(t *testing.T) {
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		flusher := prepareTestStream(t, w)
		for i := range 6 {
			writeTestStreamChunk(t, w, flusher, fmt.Sprintf("%d", i))
			if i < 5 {
				time.Sleep(60 * time.Millisecond)
			}
		}
		writeTestStreamDone(t, w, flusher)
	}))
	defer server.Close()

	client := newTestOpenAIClient(server.URL)
	stream, _, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 0, time.Second, 250*time.Millisecond, noRetryDelay, make(chan AgentEvent, 1))
	if err != nil {
		t.Fatalf("create stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()

	chunks := 1
	for {
		_, recvErr := recvStreamWithIdleTimeout(stream, 250*time.Millisecond, cancel)
		if errors.Is(recvErr, io.EOF) {
			break
		}
		if recvErr != nil {
			t.Fatalf("receive stream failed after %d chunks: %v", chunks, recvErr)
		}
		chunks++
	}
	if chunks != 6 {
		t.Fatalf("received %d chunks, want 6", chunks)
	}
}

func TestResponsesProgressEventsResetIdleTimeout(t *testing.T) {
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		flusher := prepareTestStream(t, w)
		writeResponsesTestEvent(t, w, flusher, "response.created", map[string]any{
			"type": "response.created", "response": map[string]any{"id": "resp-progress"},
		})
		time.Sleep(60 * time.Millisecond)
		writeResponsesTestEvent(t, w, flusher, "response.in_progress", map[string]any{
			"type": "response.in_progress", "response": map[string]any{"id": "resp-progress"},
		})
		time.Sleep(60 * time.Millisecond)
		writeResponsesTestEvent(t, w, flusher, "response.output_item.added", map[string]any{
			"type": "response.output_item.added", "output_index": 0,
			"item": map[string]any{"id": "rs-progress", "type": "reasoning", "status": "in_progress"},
		})
		time.Sleep(60 * time.Millisecond)
		writeResponsesTestEvent(t, w, flusher, "response.output_text.delta", map[string]any{
			"type": "response.output_text.delta", "output_index": 1, "delta": "done",
		})
		writeResponsesTestCompleted(t, w, flusher, "resp-progress", []any{map[string]any{
			"id": "msg-progress", "type": "message", "role": "assistant", "status": "completed",
			"content": []any{map[string]any{"type": "output_text", "text": "done"}},
		}}, 5, 1)
	}))
	defer server.Close()

	stream, firstResponse, cancel, err := createProtocolStreamWithRetry(
		context.Background(), newTestOpenAIClient(server.URL), "openai-responses", testChatRequest(), nil,
		0, time.Second, 100*time.Millisecond, noRetryDelay, make(chan AgentEvent, 1))
	if err != nil {
		t.Fatalf("create Responses stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()

	var content strings.Builder
	response := firstResponse
	for {
		for _, choice := range response.Choices {
			content.WriteString(choice.Delta.Content)
		}
		response, err = recvStreamWithIdleTimeout(stream, 100*time.Millisecond, cancel)
		if errors.Is(err, io.EOF) {
			break
		}
		if err != nil {
			t.Fatalf("receive Responses progress stream failed: %v", err)
		}
	}
	if content.String() != "done" {
		t.Fatalf("unexpected Responses content: %q", content.String())
	}
}

func TestClassifyRetryResponsesErrors(t *testing.T) {
	tests := []struct {
		code string
		want string
	}{
		{code: "invalid_request_error", want: "fatal"},
		{code: "rate_limit_exceeded", want: "rate_limit"},
		{code: "subscription_sharing_usage_limit_exceeded", want: "fatal"},
		{code: "server_error", want: "server_error"},
	}
	for _, test := range tests {
		t.Run(test.code, func(t *testing.T) {
			got := classifyRetry(&openai.APIError{Code: test.code, Message: "test"})
			if got != test.want {
				t.Fatalf("retry category = %q, want %q", got, test.want)
			}
		})
	}
}

func TestCreateResponsesStreamRetriesErrorAfterProgress(t *testing.T) {
	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		attempt := requests.Add(1)
		flusher := prepareTestStream(t, w)
		writeResponsesTestEvent(t, w, flusher, "response.created", map[string]any{
			"type": "response.created", "response": map[string]any{"id": "resp-retry"},
		})
		if attempt == 1 {
			writeResponsesTestEvent(t, w, flusher, "error", map[string]any{
				"type": "error", "code": "server_error", "message": "try again",
			})
			return
		}
		writeResponsesTestEvent(t, w, flusher, "response.output_text.delta", map[string]any{
			"type": "response.output_text.delta", "output_index": 0, "delta": "success",
		})
		writeResponsesTestCompleted(t, w, flusher, "resp-retry", []any{map[string]any{
			"id": "msg-retry", "type": "message", "role": "assistant", "status": "completed",
			"content": []any{map[string]any{"type": "output_text", "text": "success"}},
		}}, 2, 1)
	}))
	defer server.Close()

	events := make(chan AgentEvent, 2)
	stream, firstResponse, cancel, err := createProtocolStreamWithRetry(
		context.Background(), newTestOpenAIClient(server.URL), "openai-responses", testChatRequest(), nil,
		1, time.Second, time.Second, noRetryDelay, events)
	if err != nil {
		t.Fatalf("create Responses stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()
	if requests.Load() != 2 {
		t.Fatalf("request count = %d, want 2", requests.Load())
	}
	if len(firstResponse.Choices) != 1 || firstResponse.Choices[0].Delta.Content != "success" {
		t.Fatalf("unexpected first Responses output: %#v", firstResponse)
	}
	select {
	case event := <-events:
		if event.Type != "retry" || event.RetryAttempt != 1 || event.RetryMax != 1 {
			t.Fatalf("unexpected retry event: %#v", event)
		}
	default:
		t.Fatal("missing Responses retry event")
	}
}

func TestStreamIdleTimeoutAfterPartialResponse(t *testing.T) {
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		flusher := prepareTestStream(t, w)
		writeTestStreamChunk(t, w, flusher, "partial")
		<-r.Context().Done()
	}))
	defer server.Close()

	client := newTestOpenAIClient(server.URL)
	stream, _, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 0, time.Second, 50*time.Millisecond, noRetryDelay, make(chan AgentEvent, 1))
	if err != nil {
		t.Fatalf("create stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()

	_, recvErr := recvStreamWithIdleTimeout(stream, 50*time.Millisecond, cancel)
	if !errors.Is(recvErr, errModelStreamIdleTimeout) {
		t.Fatalf("receive error = %v, want stream idle timeout", recvErr)
	}
}

func TestCreateStreamRetriesFirstResponseTimeoutWithFreshContext(t *testing.T) {
	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		attempt := requests.Add(1)
		flusher := prepareTestStream(t, w)
		if attempt == 1 {
			<-r.Context().Done()
			return
		}
		writeTestStreamChunk(t, w, flusher, "success")
		writeTestStreamDone(t, w, flusher)
	}))
	defer server.Close()

	events := make(chan AgentEvent, 2)
	client := newTestOpenAIClient(server.URL)
	stream, first, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 1, time.Second, 50*time.Millisecond, noRetryDelay, events)
	if err != nil {
		t.Fatalf("create stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()

	if requests.Load() != 2 {
		t.Fatalf("request count = %d, want 2", requests.Load())
	}
	if len(first.Choices) != 1 || first.Choices[0].Delta.Content != "success" {
		t.Fatalf("unexpected first response: %#v", first)
	}
	select {
	case event := <-events:
		if event.Type != "retry" || event.RetryAttempt != 1 || event.RetryMax != 1 {
			t.Fatalf("unexpected retry event: %#v", event)
		}
	default:
		t.Fatal("missing retry event")
	}
}

func TestCreateStreamRequestTimeoutAndZeroRetries(t *testing.T) {
	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(_ http.ResponseWriter, r *http.Request) {
		requests.Add(1)
		select {
		case <-r.Context().Done():
		case <-time.After(200 * time.Millisecond):
		}
	}))
	defer server.Close()

	client := newTestOpenAIClient(server.URL)
	_, _, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 0, 50*time.Millisecond, time.Second, noRetryDelay, make(chan AgentEvent, 1))
	if cancel != nil {
		cancel()
	}
	if !errors.Is(err, errModelRequestTimeout) {
		t.Fatalf("create error = %v, want request timeout", err)
	}
	if requests.Load() != 1 {
		t.Fatalf("request count = %d, want 1", requests.Load())
	}
}

func TestCreateStreamRetriesRequestTimeoutWithFreshContext(t *testing.T) {
	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		attempt := requests.Add(1)
		if attempt == 1 {
			select {
			case <-r.Context().Done():
			case <-time.After(200 * time.Millisecond):
			}
			return
		}
		flusher := prepareTestStream(t, w)
		writeTestStreamChunk(t, w, flusher, "success")
		writeTestStreamDone(t, w, flusher)
	}))
	defer server.Close()

	client := newTestOpenAIClient(server.URL)
	stream, first, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 1, 50*time.Millisecond, time.Second, noRetryDelay, make(chan AgentEvent, 2))
	if err != nil {
		t.Fatalf("create stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()
	if requests.Load() != 2 {
		t.Fatalf("request count = %d, want 2", requests.Load())
	}
	if len(first.Choices) != 1 || first.Choices[0].Delta.Content != "success" {
		t.Fatalf("unexpected first response: %#v", first)
	}
}

func TestAgentChatPartialStreamTimeoutSavesInterruptedWithoutRetry(t *testing.T) {
	useTestDataDir(t)
	originalConf := kernelModel.Conf
	kernelModel.Conf = kernelModel.NewAppConf()
	kernelModel.Conf.AI.Store(kernelConf.NewAI())
	kernelModel.Conf.AI.Load().MCP = nil
	kernelModel.Conf.AI.Load().Agent.MaxToolCallRounds = 1
	kernelModel.Conf.Variables.Store(kernelConf.NewVariables())
	t.Cleanup(func() {
		kernelModel.Conf = originalConf
	})

	session := map[string]any{
		"id":        testSessionID,
		"title":     "timeout test",
		"createdAt": int64(1),
		"updatedAt": int64(1),
		"entries":   []any{map[string]any{"id": "user-1", "type": "user", "content": "hello"}},
	}
	if revision, _, err := SaveSessionState(marshalSession(t, session)); err != nil || revision != 1 {
		t.Fatalf("save initial session failed: revision=%d, err=%v", revision, err)
	}

	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		requests.Add(1)
		flusher := prepareTestStream(t, w)
		writeTestStreamChunk(t, w, flusher, "partial")
		<-r.Context().Done()
	}))
	defer server.Close()

	events := AgentChat(context.Background(), newTestOpenAIClient(server.URL), "openai", "test-model", "", 0, testSessionID, "user-1", 1, "hello", nil, "English", nil, EditorContext{}, nil, false, time.Second, 3, "", time.Second, 50*time.Millisecond)
	contentSeen := false
	errorSeen := false
	for event := range events {
		if event.Type == "content" && event.Token == "partial" {
			contentSeen = true
		}
		if event.Type == "error" {
			errorSeen = true
		}
	}
	if !contentSeen || !errorSeen {
		t.Fatalf("unexpected events: contentSeen=%v, errorSeen=%v", contentSeen, errorSeen)
	}
	if requests.Load() != 1 {
		t.Fatalf("request count = %d, want 1", requests.Load())
	}

	runtime, err := loadRuntimeState(testSessionID)
	if err != nil {
		t.Fatal(err)
	}
	if runtime.ActiveTurn == nil || runtime.ActiveTurn.State != "interrupted" {
		t.Fatalf("runtime turn was not interrupted: %#v", runtime.ActiveTurn)
	}
	if len(runtime.ActiveTurn.Delta) != 1 || runtime.ActiveTurn.Delta[0].Role != "assistant" || runtime.ActiveTurn.Delta[0].Content != "partial" {
		t.Fatalf("partial response was not checkpointed: %#v", runtime.ActiveTurn.Delta)
	}
}

func TestCreateStreamAcceptsEmptySuccessfulResponse(t *testing.T) {
	var requests atomic.Int32
	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		requests.Add(1)
		flusher := prepareTestStream(t, w)
		writeTestStreamDone(t, w, flusher)
	}))
	defer server.Close()

	client := newTestOpenAIClient(server.URL)
	stream, _, cancel, err := createProtocolStreamWithRetry(context.Background(), client, util.OpenAIProtocolChatCompletions, testChatRequest(), nil, 1, time.Second, time.Second, noRetryDelay, make(chan AgentEvent, 1))
	if err != nil {
		t.Fatalf("create stream failed: %v", err)
	}
	defer cancel()
	defer stream.Close()
	if requests.Load() != 1 {
		t.Fatalf("request count = %d, want 1", requests.Load())
	}
	if _, recvErr := stream.Recv(); !errors.Is(recvErr, io.EOF) {
		t.Fatalf("receive error = %v, want EOF", recvErr)
	}
}

func newTestOpenAIClient(serverURL string) *util.AIClient {
	config := openai.DefaultConfig("test-key")
	config.BaseURL = serverURL + "/v1"
	return &util.AIClient{Client: openai.NewClientWithConfig(config)}
}

func testChatRequest() openai.ChatCompletionRequest {
	return openai.ChatCompletionRequest{
		Model:    "test-model",
		Messages: []openai.ChatCompletionMessage{{Role: openai.ChatMessageRoleUser, Content: "test"}},
		Stream:   true,
	}
}

func noRetryDelay(string, int) time.Duration {
	return 0
}

func prepareTestStream(t *testing.T, w http.ResponseWriter) http.Flusher {
	t.Helper()
	flusher, ok := w.(http.Flusher)
	if !ok {
		t.Fatal("response writer does not support streaming")
	}
	w.Header().Set("Content-Type", "text/event-stream")
	w.WriteHeader(http.StatusOK)
	flusher.Flush()
	return flusher
}

func writeTestStreamChunk(t *testing.T, w http.ResponseWriter, flusher http.Flusher, content string) {
	t.Helper()
	if _, err := fmt.Fprintf(w, "data: {\"id\":\"chatcmpl-test\",\"object\":\"chat.completion.chunk\",\"created\":1,\"model\":\"test-model\",\"choices\":[{\"index\":0,\"delta\":{\"content\":%q}}]}\n\n", content); err != nil {
		t.Fatalf("write stream chunk failed: %v", err)
	}
	flusher.Flush()
}

func writeTestStreamDone(t *testing.T, w http.ResponseWriter, flusher http.Flusher) {
	t.Helper()
	if _, err := io.WriteString(w, "data: [DONE]\n\n"); err != nil {
		t.Fatalf("write stream terminator failed: %v", err)
	}
	flusher.Flush()
}