diff --git a/CHANGELOG.md b/CHANGELOG.md index 77c5a26..395bd73 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,5 +1,12 @@ # 更新日志 +4.3.5-rc.0 - 2025-09-14 +--- +- PRE-RELEASE: v4.3.5-rc.0是v4.3.5的预发布版本,请勿在生产环境中使用; +- CHANGE: 改进`nest`实现, 减少内存分配`10371 B/op -> 1852 B/op` `43 allocs/op -> 14 allocs/op` +- CHANGE: 为`nest`加入`dispatcher`实现, 为不同情况分配适合的处理器以保证性能与兼容性 +- CHANGE: 改进路径匹配热点的内存分配 + 4.3.4 - 2025-09-14 --- - CHANGE: 改进嵌套加速实现, 增强稳定性 @@ -25,7 +32,7 @@ 4.3.0-rc.0 - 2025-08-11 --- -- PRE-RELEASE: v4.3.0-rc.0是v4.3.0发布版本,请勿在生产环境中使用; +- PRE-RELEASE: v4.3.0-rc.0是v4.3.0的预布版本,请勿在生产环境中使用; - CHANGE: 为OCI镜像(Docker)代理带来自动library附加功能 - CHANGE(refactor): 改进OCI镜像(Docker)代理路径组成流程 - ADD: 新增[WANF](https://github.com/WJQSERVER/wanf)配置文件格式支持 diff --git a/DEV-VERSION b/DEV-VERSION index 51ff7c7..332d6f7 100644 --- a/DEV-VERSION +++ b/DEV-VERSION @@ -1 +1 @@ -4.3.0-rc.0 \ No newline at end of file +4.3.5-rc.0 \ No newline at end of file diff --git a/proxy/chunkreq.go b/proxy/chunkreq.go index 7e3725e..544a312 100644 --- a/proxy/chunkreq.go +++ b/proxy/chunkreq.go @@ -134,7 +134,7 @@ func ChunkedProxyRequest(ctx context.Context, c *touka.Context, u string, cfg *c var reader io.Reader - reader, _, err = processLinks(bodyReader, c.Request.Host, cfg, c) + reader, _, err = processLinks(bodyReader, c.Request.Host, cfg, c, bodySize) c.WriteStream(reader) if err != nil { c.Errorf("%s %s %s %s %s Failed to copy response body: %v", c.ClientIP(), c.Request.Method, u, c.UserAgent(), c.Request.Proto, err) diff --git a/proxy/handler.go b/proxy/handler.go index b15f1b5..ac89662 100644 --- a/proxy/handler.go +++ b/proxy/handler.go @@ -9,6 +9,32 @@ import ( "github.com/infinite-iroha/touka" ) +// buildHandlerPath 使用 strings.Builder 来高效地构建最终的 URL. +// 这避免了使用标准字符串拼接时发生的多次内存分配. +func buildHandlerPath(path, matcher string) string { + var sb strings.Builder + sb.Grow(len(path) + 50) + + if matcher == "blob" && strings.HasPrefix(path, "github.com") { + sb.WriteString("https://raw.githubusercontent.com") + if len(path) > 10 { // len("github.com") + pathSegment := path[10:] // skip "github.com" + if i := strings.Index(pathSegment, "/blob/"); i != -1 { + sb.WriteString(pathSegment[:i]) + sb.WriteString("/") + sb.WriteString(pathSegment[i+len("/blob/"):]) + } else { + sb.WriteString(pathSegment) + } + } + } else { + sb.WriteString("https://") + sb.WriteString(path) + } + + return sb.String() +} + var re = regexp.MustCompile(`^(http:|https:)?/?/?(.*)`) // 匹配http://或https://开头的路径 func NoRouteHandler(cfg *config.Config) touka.HandlerFunc { @@ -32,21 +58,16 @@ func NoRouteHandler(cfg *config.Config) touka.HandlerFunc { } // 制作url - rawPath = "https://" + matches[2] - - var ( - user string - repo string - matcher string - ) - + path := matches[2] var matcherErr *GHProxyErrors - user, repo, matcher, matcherErr = Matcher(rawPath, cfg) + user, repo, matcher, matcherErr := Matcher("https://"+path, cfg) if matcherErr != nil { ErrorPage(c, matcherErr) return } + rawPath = buildHandlerPath(path, matcher) + shoudBreak = listCheck(cfg, c, user, repo, rawPath) if shoudBreak { return @@ -57,11 +78,7 @@ func NoRouteHandler(cfg *config.Config) touka.HandlerFunc { return } - // 处理blob/raw路径 if matcher == "blob" { - rawPath = rawPath[18:] - rawPath = "https://raw.githubusercontent.com" + rawPath - rawPath = strings.Replace(rawPath, "/blob/", "/", 1) matcher = "raw" } diff --git a/proxy/nest.go b/proxy/nest.go index 675e4a3..921af3a 100644 --- a/proxy/nest.go +++ b/proxy/nest.go @@ -2,14 +2,78 @@ package proxy import ( "bufio" + "bytes" "fmt" "ghproxy/config" "io" "strings" + "sync" "github.com/infinite-iroha/touka" ) +var ( + prefixGithub = []byte("https://github.com") + prefixRawUser = []byte("https://raw.githubusercontent.com") + prefixRaw = []byte("https://raw.github.com") + prefixGistUser = []byte("https://gist.githubusercontent.com") + prefixGist = []byte("https://gist.github.com") + prefixAPI = []byte("https://api.github.com") + prefixHTTP = []byte("http://") + prefixHTTPS = []byte("https://") +) + +func EditorMatcherBytes(rawPath []byte, cfg *config.Config) (bool, error) { + if bytes.HasPrefix(rawPath, prefixGithub) { + return true, nil + } + if bytes.HasPrefix(rawPath, prefixRawUser) { + return true, nil + } + if bytes.HasPrefix(rawPath, prefixRaw) { + return true, nil + } + if bytes.HasPrefix(rawPath, prefixGistUser) { + return true, nil + } + if bytes.HasPrefix(rawPath, prefixGist) { + return true, nil + } + if cfg.Shell.RewriteAPI { + if bytes.HasPrefix(rawPath, prefixAPI) { + return true, nil + } + } + return false, nil +} + +func modifyURLBytes(url []byte, host []byte, cfg *config.Config) []byte { + matched, err := EditorMatcherBytes(url, cfg) + if err != nil || !matched { + return url + } + + var u []byte + if bytes.HasPrefix(url, prefixHTTPS) { + u = url[len(prefixHTTPS):] + } else if bytes.HasPrefix(url, prefixHTTP) { + u = url[len(prefixHTTP):] + } else { + u = url + } + + newLen := len(prefixHTTPS) + len(host) + 1 + len(u) + newURL := make([]byte, newLen) + + written := 0 + written += copy(newURL[written:], prefixHTTPS) + written += copy(newURL[written:], host) + written += copy(newURL[written:], []byte("/")) + copy(newURL[written:], u) + + return newURL +} + func EditorMatcher(rawPath string, cfg *config.Config) (bool, error) { // 匹配 "https://github.com"开头的链接 if strings.HasPrefix(rawPath, "https://github.com") { @@ -64,87 +128,126 @@ func modifyURL(url string, host string, cfg *config.Config) string { return url } -// processLinks 处理链接，返回包含处理后数据的 io.Reader -func processLinks(input io.ReadCloser, host string, cfg *config.Config, c *touka.Context) (readerOut io.Reader, written int64, err error) { - pipeReader, pipeWriter := io.Pipe() // 创建 io.Pipe +var bufferPool = sync.Pool{ + New: func() interface{} { + return new(bytes.Buffer) + }, +} + +// processLinksStreamingInternal is a link processing function that reads the input line by line. +// It is memory-safe for large files but less performant due to numerous small allocations. +func processLinksStreamingInternal(input io.ReadCloser, host string, cfg *config.Config, c *touka.Context) (readerOut io.Reader, written int64, err error) { + pipeReader, pipeWriter := io.Pipe() readerOut = pipeReader - go func() { // 在 Goroutine 中执行写入操作 + go func() { defer func() { - if pipeWriter != nil { // 确保 pipeWriter 关闭，即使发生错误 - if err != nil { - if closeErr := pipeWriter.CloseWithError(err); closeErr != nil { // 如果有错误，传递错误给 reader - c.Errorf("pipeWriter close with error failed: %v, original error: %v", closeErr, err) - } - } else { - if closeErr := pipeWriter.Close(); closeErr != nil { // 没有错误，正常关闭 - c.Errorf("pipeWriter close failed: %v", closeErr) - if err == nil { // 如果之前没有错误，记录关闭错误 - err = closeErr - } - } - } + if err != nil { + pipeWriter.CloseWithError(err) + } else { + pipeWriter.Close() } }() + defer input.Close() - defer func() { - if err := input.Close(); err != nil { - c.Errorf("input close failed: %v", err) - } + bufReader := bufio.NewReader(input) + bufWriter := bufio.NewWriterSize(pipeWriter, 4096) + defer bufWriter.Flush() - }() - - var bufReader *bufio.Reader - - bufReader = bufio.NewReader(input) - - var bufWriter *bufio.Writer - - bufWriter = bufio.NewWriterSize(pipeWriter, 4096) // 使用 pipeWriter - - //确保writer关闭 - defer func() { - if flushErr := bufWriter.Flush(); flushErr != nil { - c.Errorf("writer flush failed %v", flushErr) - // 如果已经存在错误，则保留。否则，记录此错误。 - if err == nil { - err = flushErr - } - } - }() - - // 使用正则表达式匹配 http 和 https 链接 for { line, readErr := bufReader.ReadString('\n') - if readErr != nil { - if readErr == io.EOF { - break // 文件结束 - } - err = fmt.Errorf("读取行错误: %v", readErr) // 传递错误 - return // Goroutine 中使用 return 返回错误 + if readErr != nil && readErr != io.EOF { + err = fmt.Errorf("read error: %w", readErr) + return } - // 替换所有匹配的 URL modifiedLine := urlPattern.ReplaceAllStringFunc(line, func(originalURL string) string { - return modifyURL(originalURL, host, cfg) // 假设 modifyURL 函数已定义 + return modifyURL(originalURL, host, cfg) }) - n, writeErr := bufWriter.WriteString(modifiedLine) - written += int64(n) // 更新写入的字节数 - if writeErr != nil { - err = fmt.Errorf("写入文件错误: %v", writeErr) // 传递错误 - return // Goroutine 中使用 return 返回错误 + var n int + n, err = bufWriter.WriteString(modifiedLine) + written += int64(n) + if err != nil { + err = fmt.Errorf("write error: %w", err) + return } - } - // 在返回之前，再刷新一次 (虽然 defer 中已经有 flush，但这里再加一次确保及时刷新) - if flushErr := bufWriter.Flush(); flushErr != nil { - if err == nil { // 避免覆盖之前的错误 - err = flushErr + if readErr == io.EOF { + break } - return // Goroutine 中使用 return 返回错误 } }() - return readerOut, written, nil // 返回 reader 和 written，error 由 Goroutine 通过 pipeWriter.CloseWithError 传递 + return readerOut, written, nil +} + +// processLinks acts as a dispatcher, choosing the best processing strategy based on file size. +// It uses a memory-safe streaming approach for large or unknown-size files, +// and a high-performance buffered approach for smaller files. +func processLinks(input io.ReadCloser, host string, cfg *config.Config, c *touka.Context, bodySize int) (readerOut io.Reader, written int64, err error) { + const sizeThreshold = 256 * 1024 // 256KB + + // Use streaming for large or unknown size files to prevent OOM + if bodySize == -1 || bodySize > sizeThreshold { + c.Debugf("Using streaming processor for large/unknown size file (%d bytes)", bodySize) + return processLinksStreamingInternal(input, host, cfg, c) + } else { + c.Debugf("Using buffered processor for small file (%d bytes)", bodySize) + return processLinksBufferedInternal(input, host, cfg, c) + } +} + +// processLinksBufferedInternal a link processing function that reads the entire content into a buffer. +// It is optimized for performance on smaller files but carries an OOM risk for large files. +func processLinksBufferedInternal(input io.ReadCloser, host string, cfg *config.Config, c *touka.Context) (readerOut io.Reader, written int64, err error) { + pipeReader, pipeWriter := io.Pipe() + readerOut = pipeReader + hostBytes := []byte(host) + + go func() { + // 在 goroutine 退出时, 根据 err 是否为 nil, 带错误或正常关闭 pipeWriter + defer func() { + if closeErr := input.Close(); closeErr != nil { + c.Errorf("input close failed: %v", closeErr) + } + }() + defer func() { + if err != nil { + if closeErr := pipeWriter.CloseWithError(err); closeErr != nil { + c.Errorf("pipeWriter close with error failed: %v", closeErr) + } + } else { + if closeErr := pipeWriter.Close(); closeErr != nil { + c.Errorf("pipeWriter close failed: %v", closeErr) + } + } + }() + + buf := bufferPool.Get().(*bytes.Buffer) + buf.Reset() + defer bufferPool.Put(buf) + + // 将全部输入读入复用的缓冲区 + if _, err = buf.ReadFrom(input); err != nil { + err = fmt.Errorf("reading input failed: %w", err) + return + } + + // 使用 ReplaceAllFunc 和字节版本辅助函数, 实现准零分配 + modifiedBytes := urlPattern.ReplaceAllFunc(buf.Bytes(), func(originalURL []byte) []byte { + return modifyURLBytes(originalURL, hostBytes, cfg) + }) + + // 将处理后的字节写回管道 + var n int + n, err = pipeWriter.Write(modifiedBytes) + if err != nil { + err = fmt.Errorf("writing to pipe failed: %w", err) + return + } + written = int64(n) + }() + + return readerOut, written, nil } diff --git a/proxy/nest_bench_test.go b/proxy/nest_bench_test.go new file mode 100644 index 0000000..d254290 --- /dev/null +++ b/proxy/nest_bench_test.go @@ -0,0 +1,65 @@ +package proxy + +import ( + "ghproxy/config" + "io" + "strings" + "testing" +) + +const benchmarkInput = ` +Some text here. +Link to be replaced: http://github.com/user/repo +Another link: https://google.com +And one more: http://example.com/some/path +This should not be replaced: notalink +End of text. +` + +func BenchmarkProcessLinksStreaming(b *testing.B) { + cfg := &config.Config{} + host := "my-proxy.com" + + b.ReportAllocs() + b.ResetTimer() + + for i := 0; i < b.N; i++ { + b.StopTimer() + input := io.NopCloser(strings.NewReader(benchmarkInput)) + b.StartTimer() + + reader, _, err := processLinksStreamingInternal(input, host, cfg, nil) + if err != nil { + b.Fatalf("processLinksStreamingInternal failed: %v", err) + } + + _, err = io.ReadAll(reader) + if err != nil { + b.Fatalf("Failed to read from processed reader: %v", err) + } + } +} + +func BenchmarkProcessLinksBuffered(b *testing.B) { + cfg := &config.Config{} + host := "my-proxy.com" + + b.ReportAllocs() + b.ResetTimer() + + for i := 0; i < b.N; i++ { + b.StopTimer() + input := io.NopCloser(strings.NewReader(benchmarkInput)) + b.StartTimer() + + reader, _, err := processLinksBufferedInternal(input, host, cfg, nil) + if err != nil { + b.Fatalf("processLinksBufferedInternal failed: %v", err) + } + + _, err = io.ReadAll(reader) + if err != nil { + b.Fatalf("Failed to read from processed reader: %v", err) + } + } +} diff --git a/proxy/routing.go b/proxy/routing.go index 7a5748f..48a4c69 100644 --- a/proxy/routing.go +++ b/proxy/routing.go @@ -7,6 +7,33 @@ import ( "github.com/infinite-iroha/touka" ) +// buildRoutingPath 使用 strings.Builder 来高效地构建最终的 URL. +// 这避免了使用标准字符串拼接时发生的多次内存分配. +func buildRoutingPath(rawPath, matcher string) string { + var sb strings.Builder + // 预分配内存以提高性能 + // (This comment is in Chinese as requested by the user) + sb.Grow(len(rawPath) + 30) + sb.WriteString("https://") + + if matcher == "blob" { + sb.WriteString("raw.githubusercontent.com") + if len(rawPath) > 10 { // len("github.com") + pathSegment := rawPath[10:] + if i := strings.Index(pathSegment, "/blob/"); i != -1 { + sb.WriteString(pathSegment[:i]) + sb.WriteString("/") + sb.WriteString(pathSegment[i+len("/blob/"):]) + } else { + sb.WriteString(pathSegment) + } + } + } else { + sb.WriteString(rawPath) + } + return sb.String() +} + func RoutingHandler(cfg *config.Config) touka.HandlerFunc { return func(c *touka.Context) { @@ -44,17 +71,11 @@ func RoutingHandler(cfg *config.Config) touka.HandlerFunc { return } - // 处理blob/raw路径 + rawPath = buildRoutingPath(rawPath, matcher) if matcher == "blob" { - rawPath = rawPath[10:] - rawPath = "raw.githubusercontent.com" + rawPath - rawPath = strings.Replace(rawPath, "/blob/", "/", 1) matcher = "raw" } - // 为rawpath加入https:// 头 - rawPath = "https://" + rawPath - switch matcher { case "releases", "blob", "raw", "gist", "api": ChunkedProxyRequest(ctx, c, rawPath, cfg, matcher)