NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #129 by repository stars
Latest stable
v1.79.0
Last release today
22 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Some releases are documented
notes for 16 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
11 years old
2052 releases · first in 2015
Drag across the ruler to compare, or tap a release to open it.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.79.0 adds a WebSocket server and client package and direct access to HTTP trailer values. It also fixes cross-stream data leaks in compression, str
v1.79.0 adds a WebSocket server and client package and direct access to HTTP trailer values. It also fixes cross-stream data leaks in compression, strengthens request forwarding and HTTP framing, and improves shutdown, multipart streaming, and static-file cache isolation.
github.com/valyala/fasthttp/websocket provides Accept, Dial, context-based reads and writes, streaming message readers and writers, ping handling, and close handshakes. Optional permessage-deflate supports both compression modes, with and without context takeover; compression is disabled by default. The default message limit is 4 MiB. Read returns connection-owned data valid until the next Read or Reader call. Follow-up fixes correct URL credential decoding and connection closing, and reduce buffer sizes for large messages. #2493, #2545RequestHeader and ResponseHeader now expose PeekTrailer and PeekAllTrailer. These search only the trailer section, so an upfront header with the same name is not mistaken for a trailer. For streamed bodies, received trailers are available after the body reaches EOF. #2534Close and Reset, including when the worker queue is full. Stackless zstd encoding finishes its writes before output buffers are released, and pooled zstd decoding cannot return another caller's output after malformed input. #2555, #2554, #2564net/http requests use the same resource interpretation as fasthttp. Incoming request authorities containing userinfo are rejected before server callbacks run; client-constructed URLs still support credentials. #2572, #2575, #2580, #2579=, preventing a nameless cookie from becoming a named one. #2570, #2574, #2571, #2586Content-Type values are rejected before reading the body; identical repeats remain accepted. Chunk parsing bounds optional padding on individual size lines and excess framing overhead across chunks, applying the same limits to buffered bodies, streams, and unread-body drains. #2576, #2573CompressRoot are separated by source root, preventing one FS handler from serving another root's cached content. Existing flat cache entries are regenerated on demand. #2581, #2584101 Switching Protocols instead of reusing them for another HTTP request. #2562, #2568, #2569, #2566MaxRequestBodySize remain on the body stream so the handler can enforce its own limit before form buffering or temporary-file creation. Timed PipelineClient requests preserve pre-parsed uploads when copying work for asynchronous execution. #2583, #2563Other fixes report io.ErrShortWrite from affected writers and compression helpers, separate IPv4-only and dual-stack DNS cache entries, support HTTP proxy URLs with a trailing slash, and correct generated directory links, modification-time handling, and Windows CONNECT redirects. The Server.MaxRequestBodySize documentation now explains that buffered requests can reserve memory from the declared body size before payload data arrives.
Buffered chunk sizes are parsed directly instead of using one ReadByte call per digit. In the PR's Apple M5 benchmarks, two-chunk responses took about 10% less time, two-chunk requests 9% less time, and bodies with 200 one-byte chunks 17% less time. Results varied by platform and code layout; readers delivering one byte per call were 2–3% slower on the tested AMD EPYC system. These are focused PR measurements, not a release-wide comparison. #2539
The new WebSocket implementation also reduces excess buffer capacity for large messages. In the follow-up PR's examples, reading a compressed 4 MiB message used a 4 MiB buffer instead of 16 MiB, and compressing 1 MiB of repetitive text used a 64 KiB output buffer instead of 2 MiB. #2545
Full Changelog: v1.78.0...v1.79.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.78.0 fixes lost responses during graceful shutdown and makes streamed bodies report truncation reliably. It also improves redirect handling, preser
v1.78.0 fixes lost responses during graceful shutdown and makes streamed bodies report truncation reliably. It also improves redirect handling, preserves multipart uploads when copying requests, and speeds up path normalization, query parsing, header operations, and concurrent client requests.
io.ErrUnexpectedEOF if the peer closes at a chunk boundary before the terminating chunk arrives. If the trailer section is incomplete, the stream returns ErrBrokenChunk wrapping io.ErrUnexpectedEOF. #2515, #2538net/http handlers preserve the final response after a 1xx status. A handler that writes 103 Early Hints followed by 200 OK now sends the 200 and its body. The adaptor does not forward the interim status; 101 Switching Protocols remains final. #2516Request.KeepBodyBuffer and Response.KeepBodyBuffer settings before resetting for the next keep-alive request, so oversized buffers can be released according to the configured pool limit. #2508mailto: and native-app callbacks. RequestCtx.Redirect keeps targets with a scheme but no // intact. When the client receives a non-HTTP target of that form, it returns the 3xx response with its Location and no error. For references matching the base scheme, URI.Update resolves the remainder as relative: http:next against http://host/a/b resolves to http://host/a/next. #2503, #2506, #2507Location cannot be parsed, the client returns the redirect response and the parse error. The new URI.UpdateBytesErr exposes that error to callers; URI.Update and URI.UpdateBytes retain their existing behavior. #2523Authorization header takes precedence over URL credentials. When the header is empty, Basic authentication uses the percent-decoded username and password from the URI: p%40ss becomes p@ss, and a literal + stays +. URI.Parse also validates IPv6 literals with zone identifiers and rejects an empty zone. #2526, #2529, #2535boundary parameter name are matched case-insensitively. Multipart parsing accepts optional whitespace around ; and preserves semicolons inside quoted parameter values. Request.CopyTo also preserves uploads parsed directly from the incoming body, so forwarding the copy no longer sends an empty body. #2511, #2512, #2517, #2521Expires again accepts RFC 1123 dates with zone names such as UTC. An unusable Domain or Path is ignored without discarding the cookie or skipping later attributes such as HttpOnly. #2519, #2524FS can again serve compressed content when Root names a file, keeping compressed copies in its parent directory. Body decoders match Content-Encoding case-insensitively, and zstd level 0 selects the default level instead of panicking. #2505, #2509, #2510Path normalization skips unnecessary dot-segment checks for filenames such as /app.min.js. Query parsing locates escapes while splitting arguments, header operations avoid redundant scans, and Client.Do reduces contention between concurrent callers. #2513
In the PR benchmarks, parsing paths with file extensions took 40% less time, query parsing took 29% less time, and concurrent Client.Do dispatch with a no-op transport took 19% less time at GOMAXPROCS=4. End-to-end improvements were smaller: 4–6% less time for cached small files and 6–8% less time for in-memory client requests. Allocation counts were unchanged, and the pipelined server benchmarks showed no significant change. #2513
Full Changelog: v1.77.0...v1.78.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.77.0 strengthens static-file confinement, fixes hijacked connection handling, and improves HTTP and cookie compatibility. It also adds an opt-in bo
v1.77.0 strengthens static-file confinement, fixes hijacked connection handling, and improves HTTP and cookie compatibility. It also adds an opt-in bounded stream reader, configurable body-buffer retention, and deadline support for adapted net/http handlers.
Static-file symlink behavior changes by default. The default OS-backed FS now rejects symlinks that leave Root or CompressRoot, as well as absolute symlinks even when their targets are inside the root. Relative symlinks that stay inside the root continue to work. Set FS.AllowSymlinksOutsideRoot to restore the previous behavior when external symlinks are intentional. Custom FS.FS implementations are unaffected. #2490
os.Root to prevent symlink traversal outside Root and CompressRoot. Each root is confined independently; empty roots with AllowEmptyRoot and whole-volume roots retain their existing behavior. #2490FS.FS previously caused an os.Remove call using its filesystem-relative name, potentially deleting a matching file in the process working directory. Custom filesystems now regenerate stale compressed content in memory without removing a disk file. #2489KeepHijackedConns and ReduceMemoryUsage enabled, the hijacked reader is detached from the context before the hijack handler runs. This prevents later reads from panicking or accessing another client's connection after context reuse. #2495NewBoundedStreamReader works with SetBodyStream and applies backpressure when consumers fall behind. Active copied payload is limited to 160 KiB per bridge, or 320 KiB across the two bridges used by its bounded compression path. These limits exclude caller buffers, codecs, sockets, and other overhead; wrapping the reader loses the combined compression bound. Existing NewStreamReader and SetBodyStreamWriter behavior is unchanged. #2488Request.KeepBodyBuffer and Response.KeepBodyBuffer fields preserve body buffers across resets for reuse. Releasing a request or response to the pool clears the retention setting. #2451RequestCtx.SetReadDeadline and SetWriteDeadline are also exposed through fasthttpadaptor, enabling http.ResponseController deadline calls. The adaptor still buffers the request body before invoking the handler, so a read deadline set inside an adapted handler cannot govern the preceding upload. #2452Accept-Encoding entries such as gzip;q=0.8 now match, while q=0 makes a coding unacceptable. Matching remains case-insensitive and requires a complete coding token; quality-based preference ordering and wildcard matching are not added. #2491Expect: 100-continue is ignored for HTTP/1.0 requests, and 103 Early Hints is not sent to HTTP/1.0 clients. #2492, #2496DisableKeepalive and reaching MaxRequestsPerConn no longer suppress a registered hijack handler or add an unwanted Connection: close to an upgrade response. Explicit connection-close requests and other existing close conditions still apply. #2494fasthttputil.PipeConns now affects blocked reads and writes, and write-deadline updates no longer race with writes. #2471Max-Age attributes are ignored, Secure, HttpOnly, and Partitioned are recognized even with a value, and surrounding tabs are trimmed. Set-Cookie values without = are stored under the same empty key returned by Cookie.ParseBytes, keeping lookup and deletion consistent. #2497, #2498, #2500, #2501Proxy authentication now uses decoded credentials in Proxy-Authorization, and static-file 416 Range Not Satisfiable responses include Content-Range. #2499, #2502
Full Changelog: v1.76.0...v1.77.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.76.0 strengthens HTTP message framing, static-file isolation, and server connection handling. It also improves HTTP compatibility, avoids quadratic
v1.76.0 strengthens HTTP message framing, static-file isolation, and server connection handling. It also improves HTTP compatibility, avoids quadratic work when parsing large trailer sets, and adds configurable TCP keep-alives to TCPDialer.
Content-Length headers are synchronized with the outgoing body, duplicate or conflicting framing headers are rejected, and fixed-length request and response streams cannot write beyond their declared size. FS and SendFile also stop at the advertised Content-Length when a file grows during serving. #2470, #2462Content-Length and Transfer-Encoding close the connection. The connection is no longer reused after a request containing both framing headers. #2476NewVHostPathRewriter avoids decoding the host and path twice and handles invalid hosts safely, preventing crafted requests from escaping the virtual-host directory and accessing files directly under FS.Root. This affects FS handlers using NewVHostPathRewriter. #2487MaxConnsPerIP or Concurrency are closed without attempting an HTTP error response, which could otherwise trigger a TLS handshake and stall while waiting for a ClientHello. IPv6 clients are now also counted against MaxConnsPerIP. #2486, #2468Trailer headers and chunked trailer sections while preserving trailer order. #2465416; invalid single byte ranges still return 416. Range is also ignored when If-Range does not match, and both Range and If-Modified-Since are ignored for methods other than GET and HEAD. #2467, #2473, #2474, #2475PipelineClient skips the response body for HEAD requests. #2469ParseHTTPDate accepts obsolete HTTP-date formats, cookies with an invalid Expires attribute remain parseable, Expect: 100-continue is matched case-insensitively, and Accept-Encoding matches complete tokens. Asterisk-form request targets are rejected for methods other than OPTIONS. #2466, #2481, #2482, #2479, #2463TCPDialer.KeepAlive field sets the keep-alive period for connections created by the dialer; a negative value disables keep-alives. #2484Full Changelog: v1.75.0...v1.76.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.75.0 improves HTTP parsing and serialization, URI processing, cached file serving, and client connection handling. It also fixes cookie injection,
v1.75.0 improves HTTP parsing and serialization, URI processing, cached file serving, and client connection handling. It also fixes cookie injection, request/response framing problems, and excessive CPU or memory consumption from certain inputs.
Go 1.26 or newer is now required. #2395
Server and Date lines, faster HTTP date formatting, and fewer flushes around chunked trailers. Large buffered responses can use writev on TCP and Unix connections; TLS and streamed bodies retain the buffered path. #2425, #2415, #2434, #2426, #2433/, faster quoting and validation, and linear-time path normalization. #2411, #2423, #2453FS; the existing cache metadata is separate. Changes to these files become visible after cache expiration (CacheDuration, 10 seconds by default). #2446Server.LazyRequestTime avoids reading the clock until RequestCtx.Time() is called; with this option, the returned time is the time of the first call. The benchmark comparison below uses the default setting. #2439, #2404StreamRequestBody enabled, the server drains unread body data before reusing the connection, with a 256 KiB drain limit, or closes the connection when safe draining is not possible. Responses are flushed before draining so early replies reach clients still uploading, and response streams can consume the request body before its remaining size is checked. #2368, #2461100-continue requests close the connection. A denied request could previously leave its body unread and have those bytes interpreted as the next request, desynchronizing a reused connection. #2450RequestHeader.SetCookie and its byte variants now sanitize semicolons in names and values, so one logical cookie cannot become multiple cookies on the wire. #2393BodyUnzstdWithLimit limits now select a memory-limited decoder, with caps rounded up in 8 MiB buckets; exceeding the decoder limit returns ErrBodyTooLarge. #2407Content-Length and Transfer-Encoding on 1xx/204 responses. Explicit content-type metadata and permitted 304 metadata remain supported. #2441, #2447, #2457PipelineClient.MaxResponseBodySize to a positive value to reject oversized responses, including with Response.StreamBody; the default remains unlimited. Idle workers now retire after errors, abandoned requests release their streams, and failed connection establishment is throttled. #2427, #2458Argument sorting now honors any negative comparator result in Args.Sort and Args.SortKeys, and static-file serving rejects zero-length suffix byte ranges (bytes=-0) with 416 Range Not Satisfiable. #2459, #2460
Other significant changes include opt-in ETags for FS, a net/http request conversion function, improved adaptor handling of flushes, aborts and trailers, and fixes to shutdown and per-IP connection lifetimes. #2399, #2104, #2429, #2445, #2352, #2402
Compared v1.74.0 with master 6141d3a on Ubuntu 26.04, an AMD EPYC 9454, and Go 1.27.1. Each benchmark has ten samples per revision, run alternately with the same inputs and CPU affinity. Negative percentages mean less time per operation. All rows below have p < 0.001.
| Benchmark | v1.74.0 | Release candidate | Time/op change |
|---|---|---|---|
| Minimal GET, pipeline depth 64 | 484.30 ns | 278.05 ns | -42.6% |
| Browser-shaped GET, pipeline depth 16 | 1.080 µs | 722.55 ns | -33.1% |
| Request-header parsing | 561.80 ns | 348.15 ns | -38.0% |
| Response-header parsing | 466.95 ns | 277.55 ns | -40.6% |
| Request-header writing | 49.84 ns | 37.45 ns | -24.8% |
| Response-header writing | 68.64 ns | 40.99 ns | -40.3% |
| Cached 1 KiB file, directory | 2.324 µs | 611.85 ns | -73.7% |
Cached 1 KiB file, os.DirFS |
3.805 µs | 612.30 ns | -83.9% |
| URI path with query string | 91.39 ns | 46.06 ns | -49.6% |
| HTTP date formatting | 108.25 ns | 28.59 ns | -73.6% |
| Trickled ~4 KiB headers, one byte per read | 1385.784 µs | 191.257 µs | -86.2% |
8 KiB path of repeated / |
352.529 µs | 54.330 µs | -84.6% |
~8 KiB path of repeated /a/.. |
96.733 µs | 22.253 µs | -77.0% |
| TCP client/server round trips, small response | 5.193 µs | 5.030 µs | -3.1% |
| TCP client/server round trips, 64 KiB response | 15.302 µs | 13.478 µs | -11.9% |
The pipelined serving and file benchmarks use an in-memory connection with writes discarded. The two TCP rows include real loopback networking. These results describe the selected workloads; TCP gains are smaller than the gains in parsing and serialization. Server.LazyRequestTime remains disabled in this comparison.
Full Changelog: v1.74.0...v1.75.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.