feat(browser-fetch): add lightpanda-backed fetch service and knowledge URL fallback
Adds a new browser-fetch microservice that wraps Lightpanda for rendering JS-heavy pages, and wires it into the knowledge URL parser as a fallback when the upstream link reader returns 403/429/5xx for an allow-listed domain (e.g. zhuanlan.zhihu.com). Includes config/env plumbing, hot-reload diff support, and full deploy assets (Dockerfile target, docker-compose, k3s manifests, image-build/package scripts).
This commit is contained in:
@@ -0,0 +1,236 @@
|
||||
package app
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"context"
|
||||
"crypto/sha256"
|
||||
"encoding/hex"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"strings"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"github.com/geo-platform/tenant-api/internal/shared/config"
|
||||
"go.uber.org/zap"
|
||||
)
|
||||
|
||||
var errBrowserFetchQueueFull = errors.New("browser fetch queue is full")
|
||||
|
||||
type browserFetchResult struct {
|
||||
URL string `json:"url,omitempty"`
|
||||
Title string `json:"title,omitempty"`
|
||||
Text string `json:"text,omitempty"`
|
||||
Markdown string `json:"markdown,omitempty"`
|
||||
}
|
||||
|
||||
type browserFetchRequest struct {
|
||||
URL string `json:"url"`
|
||||
TimeoutMS int64 `json:"timeout_ms,omitempty"`
|
||||
Options map[string]interface{} `json:"options,omitempty"`
|
||||
ctx context.Context `json:"-"`
|
||||
cfg config.BrowserFetchConfig `json:"-"`
|
||||
resp chan browserFetchRequestResult `json:"-"`
|
||||
}
|
||||
|
||||
type browserFetchRequestResult struct {
|
||||
result browserFetchResult
|
||||
err error
|
||||
}
|
||||
|
||||
type browserFetchQueue struct {
|
||||
logger *zap.Logger
|
||||
client *http.Client
|
||||
jobs chan browserFetchRequest
|
||||
cache *browserFetchMemoryCache
|
||||
}
|
||||
|
||||
type browserFetchCacheEntry struct {
|
||||
result browserFetchResult
|
||||
expiresAt time.Time
|
||||
}
|
||||
|
||||
type browserFetchMemoryCache struct {
|
||||
mu sync.Mutex
|
||||
entries map[string]browserFetchCacheEntry
|
||||
}
|
||||
|
||||
func (s *KnowledgeService) browserFetchQueue(cfg config.BrowserFetchConfig) *browserFetchQueue {
|
||||
s.browserFetchMu.Lock()
|
||||
defer s.browserFetchMu.Unlock()
|
||||
if s.browserFetchQ != nil {
|
||||
return s.browserFetchQ
|
||||
}
|
||||
|
||||
config.NormalizeBrowserFetchConfig(&cfg)
|
||||
queue := &browserFetchQueue{
|
||||
logger: s.logger,
|
||||
client: &http.Client{},
|
||||
jobs: make(chan browserFetchRequest, cfg.QueueSize),
|
||||
cache: &browserFetchMemoryCache{
|
||||
entries: make(map[string]browserFetchCacheEntry),
|
||||
},
|
||||
}
|
||||
|
||||
for i := 0; i < cfg.WorkerConcurrency; i++ {
|
||||
go queue.runWorker(i)
|
||||
}
|
||||
s.browserFetchQ = queue
|
||||
return queue
|
||||
}
|
||||
|
||||
func (q *browserFetchQueue) Fetch(ctx context.Context, rawURL string, cfg config.BrowserFetchConfig) (browserFetchResult, error) {
|
||||
if q == nil {
|
||||
return browserFetchResult{}, errBrowserFetchNotAvailable
|
||||
}
|
||||
config.NormalizeBrowserFetchConfig(&cfg)
|
||||
cacheKey := browserFetchCacheKey(rawURL)
|
||||
if cfg.CacheTTL > 0 {
|
||||
if result, ok := q.cache.Get(cacheKey); ok {
|
||||
return result, nil
|
||||
}
|
||||
}
|
||||
|
||||
request := browserFetchRequest{
|
||||
URL: rawURL,
|
||||
TimeoutMS: int64(cfg.RequestTimeout / time.Millisecond),
|
||||
Options: map[string]interface{}{
|
||||
"provider": cfg.Provider,
|
||||
},
|
||||
ctx: ctx,
|
||||
cfg: cfg,
|
||||
resp: make(chan browserFetchRequestResult, 1),
|
||||
}
|
||||
|
||||
select {
|
||||
case q.jobs <- request:
|
||||
case <-ctx.Done():
|
||||
return browserFetchResult{}, ctx.Err()
|
||||
default:
|
||||
return browserFetchResult{}, errBrowserFetchQueueFull
|
||||
}
|
||||
|
||||
select {
|
||||
case result := <-request.resp:
|
||||
if result.err != nil {
|
||||
return browserFetchResult{}, result.err
|
||||
}
|
||||
if cfg.CacheTTL > 0 {
|
||||
q.cache.Set(cacheKey, result.result, cfg.CacheTTL)
|
||||
}
|
||||
return result.result, nil
|
||||
case <-ctx.Done():
|
||||
return browserFetchResult{}, ctx.Err()
|
||||
}
|
||||
}
|
||||
|
||||
func (q *browserFetchQueue) runWorker(workerID int) {
|
||||
for request := range q.jobs {
|
||||
result, err := q.execute(context.Background(), request)
|
||||
if err != nil && q.logger != nil {
|
||||
q.logger.Warn("browser fetch request failed",
|
||||
zap.Int("worker_id", workerID),
|
||||
zap.String("url", request.URL),
|
||||
zap.Error(err),
|
||||
)
|
||||
}
|
||||
request.resp <- browserFetchRequestResult{result: result, err: err}
|
||||
}
|
||||
}
|
||||
|
||||
func (q *browserFetchQueue) execute(parent context.Context, request browserFetchRequest) (browserFetchResult, error) {
|
||||
if request.ctx != nil {
|
||||
parent = request.ctx
|
||||
}
|
||||
timeout := time.Duration(request.TimeoutMS) * time.Millisecond
|
||||
if timeout <= 0 {
|
||||
timeout = 45 * time.Second
|
||||
}
|
||||
ctx, cancel := context.WithTimeout(parent, timeout)
|
||||
defer cancel()
|
||||
|
||||
return q.callService(ctx, request.URL, request.cfg)
|
||||
}
|
||||
|
||||
func (q *browserFetchQueue) callService(ctx context.Context, rawURL string, cfg config.BrowserFetchConfig) (browserFetchResult, error) {
|
||||
config.NormalizeBrowserFetchConfig(&cfg)
|
||||
endpoint := strings.TrimRight(cfg.ServiceURL, "/") + "/fetch"
|
||||
payload := browserFetchRequest{
|
||||
URL: rawURL,
|
||||
TimeoutMS: int64(cfg.RequestTimeout / time.Millisecond),
|
||||
}
|
||||
body, err := json.Marshal(payload)
|
||||
if err != nil {
|
||||
return browserFetchResult{}, fmt.Errorf("marshal browser fetch request: %w", err)
|
||||
}
|
||||
|
||||
req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(body))
|
||||
if err != nil {
|
||||
return browserFetchResult{}, fmt.Errorf("create browser fetch request: %w", err)
|
||||
}
|
||||
req.Header.Set("Content-Type", "application/json")
|
||||
if cfg.Token != "" {
|
||||
req.Header.Set("Authorization", "Bearer "+cfg.Token)
|
||||
}
|
||||
|
||||
resp, err := q.client.Do(req)
|
||||
if err != nil {
|
||||
return browserFetchResult{}, fmt.Errorf("execute browser fetch request: %w", err)
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
|
||||
if resp.StatusCode >= http.StatusBadRequest {
|
||||
data, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
|
||||
return browserFetchResult{}, fmt.Errorf("browser fetch service failed: status=%d body=%s", resp.StatusCode, strings.TrimSpace(string(data)))
|
||||
}
|
||||
|
||||
var result browserFetchResult
|
||||
if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
|
||||
return browserFetchResult{}, fmt.Errorf("decode browser fetch response: %w", err)
|
||||
}
|
||||
return result, nil
|
||||
}
|
||||
|
||||
func (c *browserFetchMemoryCache) Get(key string) (browserFetchResult, bool) {
|
||||
if c == nil || key == "" {
|
||||
return browserFetchResult{}, false
|
||||
}
|
||||
now := time.Now()
|
||||
c.mu.Lock()
|
||||
defer c.mu.Unlock()
|
||||
entry, ok := c.entries[key]
|
||||
if !ok {
|
||||
return browserFetchResult{}, false
|
||||
}
|
||||
if now.After(entry.expiresAt) {
|
||||
delete(c.entries, key)
|
||||
return browserFetchResult{}, false
|
||||
}
|
||||
return entry.result, true
|
||||
}
|
||||
|
||||
func (c *browserFetchMemoryCache) Set(key string, result browserFetchResult, ttl time.Duration) {
|
||||
if c == nil || key == "" || ttl <= 0 {
|
||||
return
|
||||
}
|
||||
c.mu.Lock()
|
||||
defer c.mu.Unlock()
|
||||
c.entries[key] = browserFetchCacheEntry{
|
||||
result: result,
|
||||
expiresAt: time.Now().Add(ttl),
|
||||
}
|
||||
}
|
||||
|
||||
func browserFetchCacheKey(rawURL string) string {
|
||||
normalized := strings.TrimSpace(rawURL)
|
||||
if parsed, err := url.Parse(normalized); err == nil && parsed.Host != "" {
|
||||
parsed.Fragment = ""
|
||||
normalized = parsed.String()
|
||||
}
|
||||
sum := sha256.Sum256([]byte(normalized))
|
||||
return hex.EncodeToString(sum[:])
|
||||
}
|
||||
@@ -49,6 +49,8 @@ type KnowledgeService struct {
|
||||
jobs chan knowledgeParseJob
|
||||
cleanupMu sync.Mutex
|
||||
cleanupInFlight map[string]struct{}
|
||||
browserFetchMu sync.Mutex
|
||||
browserFetchQ *browserFetchQueue
|
||||
}
|
||||
|
||||
type knowledgeParseJob struct {
|
||||
@@ -199,7 +201,7 @@ func (s *KnowledgeService) WithConfigProvider(provider runtimeConfigProvider) *K
|
||||
func (s *KnowledgeService) runtimeConfig() knowledgeRuntimeConfig {
|
||||
if s != nil && s.configProvider != nil {
|
||||
if cfg := s.configProvider.Current(); cfg != nil {
|
||||
return knowledgeRuntimeFromConfig(cfg.Retrieval, cfg.LLM)
|
||||
return knowledgeRuntimeFromFullConfig(cfg)
|
||||
}
|
||||
}
|
||||
if s == nil {
|
||||
|
||||
@@ -4,12 +4,16 @@ import (
|
||||
"bytes"
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/geo-platform/tenant-api/internal/shared/config"
|
||||
"github.com/geo-platform/tenant-api/internal/shared/llm"
|
||||
"github.com/geo-platform/tenant-api/internal/tenant/prompts"
|
||||
"go.uber.org/zap"
|
||||
@@ -23,6 +27,37 @@ const (
|
||||
defaultKnowledgeURLMarkdownOutputToken = 12000
|
||||
)
|
||||
|
||||
var errBrowserFetchNotAvailable = fmt.Errorf("browser fetch fallback is not available")
|
||||
|
||||
type sourceFetchRestrictedError struct {
|
||||
URL string
|
||||
Status int
|
||||
Cause error
|
||||
}
|
||||
|
||||
func (e sourceFetchRestrictedError) Error() string {
|
||||
cause := ""
|
||||
if e.Cause != nil {
|
||||
cause = strings.TrimSpace(e.Cause.Error())
|
||||
}
|
||||
if cause == "" {
|
||||
cause = "source website requires browser rendering or blocks server-side fetch"
|
||||
}
|
||||
if e.Status > 0 {
|
||||
return fmt.Sprintf("source website fetch restricted status=%d: %s", e.Status, cause)
|
||||
}
|
||||
return "source website fetch restricted: " + cause
|
||||
}
|
||||
|
||||
func (e sourceFetchRestrictedError) Unwrap() error {
|
||||
return e.Cause
|
||||
}
|
||||
|
||||
func isSourceFetchRestrictedError(err error) bool {
|
||||
var restricted sourceFetchRestrictedError
|
||||
return errors.As(err, &restricted)
|
||||
}
|
||||
|
||||
type arkToolExecuteRequest struct {
|
||||
ActionName string `json:"action_name"`
|
||||
ToolName string `json:"tool_name"`
|
||||
@@ -50,7 +85,20 @@ type arkWebDataItem struct {
|
||||
func (s *KnowledgeService) parseWebsiteKnowledge(ctx context.Context, rawURL string) (*knowledgeParsedContent, error) {
|
||||
item, err := s.executeLinkReader(ctx, rawURL)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
if !s.shouldTryBrowserFetchFallback(rawURL, err) {
|
||||
return nil, classifySourceFetchError(rawURL, err)
|
||||
}
|
||||
fallback, fallbackErr := s.parseWebsiteKnowledgeWithBrowserFetch(ctx, rawURL)
|
||||
if fallbackErr == nil {
|
||||
return fallback, nil
|
||||
}
|
||||
if s.logger != nil {
|
||||
s.logger.Warn("knowledge browser fetch fallback failed",
|
||||
zap.String("url", rawURL),
|
||||
zap.Error(fallbackErr),
|
||||
)
|
||||
}
|
||||
return nil, classifySourceFetchError(rawURL, err)
|
||||
}
|
||||
|
||||
text := normalizeKnowledgeText(item.Content)
|
||||
@@ -76,6 +124,65 @@ func (s *KnowledgeService) parseWebsiteKnowledge(ctx context.Context, rawURL str
|
||||
}, nil
|
||||
}
|
||||
|
||||
func (s *KnowledgeService) parseWebsiteKnowledgeWithBrowserFetch(ctx context.Context, rawURL string) (*knowledgeParsedContent, error) {
|
||||
runtimeCfg := s.runtimeConfig()
|
||||
browserCfg := runtimeCfg.BrowserFetch
|
||||
config.NormalizeBrowserFetchConfig(&browserCfg)
|
||||
if !browserCfg.Enabled {
|
||||
return nil, errBrowserFetchNotAvailable
|
||||
}
|
||||
|
||||
provider := strings.ToLower(strings.TrimSpace(browserCfg.Provider))
|
||||
if provider != "lightpanda" && provider != "lightpanda_service" {
|
||||
return nil, fmt.Errorf("unsupported browser fetch provider %q", browserCfg.Provider)
|
||||
}
|
||||
|
||||
if strings.TrimSpace(browserCfg.ServiceURL) == "" {
|
||||
return nil, fmt.Errorf("%w: browser_fetch.service_url is empty", errBrowserFetchNotAvailable)
|
||||
}
|
||||
|
||||
callCtx := ctx
|
||||
if browserCfg.RequestTimeout > 0 {
|
||||
var cancel context.CancelFunc
|
||||
callCtx, cancel = context.WithTimeout(ctx, browserCfg.RequestTimeout)
|
||||
defer cancel()
|
||||
}
|
||||
|
||||
result, err := s.browserFetchQueue(browserCfg).Fetch(callCtx, rawURL, browserCfg)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
text := normalizeKnowledgeText(result.Text)
|
||||
if text == "" {
|
||||
text = normalizeKnowledgeText(result.Markdown)
|
||||
}
|
||||
if text == "" {
|
||||
return nil, fmt.Errorf("browser fetch returned empty content")
|
||||
}
|
||||
|
||||
title := strings.TrimSpace(result.Title)
|
||||
markdown := strings.TrimSpace(result.Markdown)
|
||||
if markdown == "" {
|
||||
var err error
|
||||
markdown, err = s.formatWebsiteMarkdown(ctx, title, text, rawURL)
|
||||
if err != nil {
|
||||
if s.logger != nil {
|
||||
s.logger.Warn("knowledge browser fetch markdown formatting failed, fallback to local markdown",
|
||||
zap.String("url", rawURL),
|
||||
zap.Error(err),
|
||||
)
|
||||
}
|
||||
markdown = buildKnowledgeMarkdownFromText(title, text)
|
||||
}
|
||||
}
|
||||
|
||||
return &knowledgeParsedContent{
|
||||
Text: text,
|
||||
Markdown: markdown,
|
||||
}, nil
|
||||
}
|
||||
|
||||
func (s *KnowledgeService) executeLinkReader(ctx context.Context, rawURL string) (*arkWebDataItem, error) {
|
||||
runtimeCfg := s.runtimeConfig()
|
||||
baseURL := strings.TrimRight(strings.TrimSpace(runtimeCfg.ArkBaseURL), "/")
|
||||
@@ -135,6 +242,90 @@ func (s *KnowledgeService) executeLinkReader(ctx context.Context, rawURL string)
|
||||
return &item, nil
|
||||
}
|
||||
|
||||
func classifySourceFetchError(rawURL string, err error) error {
|
||||
if err == nil {
|
||||
return nil
|
||||
}
|
||||
status := extractHTTPStatusFromError(err)
|
||||
if status == http.StatusForbidden || status == http.StatusTooManyRequests || status >= http.StatusInternalServerError {
|
||||
return sourceFetchRestrictedError{URL: rawURL, Status: status, Cause: err}
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
func extractHTTPStatusFromError(err error) int {
|
||||
if err == nil {
|
||||
return 0
|
||||
}
|
||||
message := err.Error()
|
||||
idx := strings.Index(message, "status=")
|
||||
if idx < 0 {
|
||||
return 0
|
||||
}
|
||||
value := message[idx+len("status="):]
|
||||
end := 0
|
||||
for end < len(value) && value[end] >= '0' && value[end] <= '9' {
|
||||
end++
|
||||
}
|
||||
if end == 0 {
|
||||
return 0
|
||||
}
|
||||
status, parseErr := strconv.Atoi(value[:end])
|
||||
if parseErr != nil {
|
||||
return 0
|
||||
}
|
||||
return status
|
||||
}
|
||||
|
||||
func (s *KnowledgeService) shouldTryBrowserFetchFallback(rawURL string, err error) bool {
|
||||
if s == nil {
|
||||
return false
|
||||
}
|
||||
runtimeCfg := s.runtimeConfig()
|
||||
browserCfg := runtimeCfg.BrowserFetch
|
||||
config.NormalizeBrowserFetchConfig(&browserCfg)
|
||||
if !browserCfg.Enabled {
|
||||
return false
|
||||
}
|
||||
if !browserFetchDomainAllowed(rawURL, browserCfg.AllowedDomains) {
|
||||
return false
|
||||
}
|
||||
status := extractHTTPStatusFromError(err)
|
||||
if status <= 0 {
|
||||
return false
|
||||
}
|
||||
for _, trigger := range browserCfg.TriggerStatus {
|
||||
if status == trigger {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func browserFetchDomainAllowed(rawURL string, allowedDomains []string) bool {
|
||||
if len(allowedDomains) == 0 {
|
||||
return false
|
||||
}
|
||||
parsed, err := url.Parse(rawURL)
|
||||
if err != nil {
|
||||
return false
|
||||
}
|
||||
host := strings.ToLower(strings.TrimSpace(parsed.Hostname()))
|
||||
if host == "" {
|
||||
return false
|
||||
}
|
||||
for _, domain := range allowedDomains {
|
||||
domain = strings.ToLower(strings.TrimPrefix(strings.TrimSpace(domain), "."))
|
||||
if domain == "" {
|
||||
continue
|
||||
}
|
||||
if host == domain || strings.HasSuffix(host, "."+domain) {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func (s *KnowledgeService) formatWebsiteMarkdown(ctx context.Context, title, content, rawURL string) (string, error) {
|
||||
content = strings.TrimSpace(content)
|
||||
if content == "" {
|
||||
|
||||
@@ -26,6 +26,7 @@ type knowledgeRuntimeConfig struct {
|
||||
ArkBaseURL string
|
||||
ArkAPIKey string
|
||||
URLMarkdownModel string
|
||||
BrowserFetch config.BrowserFetchConfig
|
||||
}
|
||||
|
||||
type generationStreamDisabledConfigProvider struct {
|
||||
@@ -103,3 +104,14 @@ func knowledgeRuntimeFromConfig(retrievalCfg config.RetrievalConfig, llmCfg conf
|
||||
URLMarkdownModel: model,
|
||||
}
|
||||
}
|
||||
|
||||
func knowledgeRuntimeFromFullConfig(cfg *config.Config) knowledgeRuntimeConfig {
|
||||
if cfg == nil {
|
||||
return knowledgeRuntimeFromConfig(config.RetrievalConfig{}, config.LLMConfig{})
|
||||
}
|
||||
runtime := knowledgeRuntimeFromConfig(cfg.Retrieval, cfg.LLM)
|
||||
browserFetch := cfg.BrowserFetch
|
||||
config.NormalizeBrowserFetchConfig(&browserFetch)
|
||||
runtime.BrowserFetch = browserFetch
|
||||
return runtime
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user