v0.2.3

Patch changes

• f524020 Prebuilt deb and rpm packages for amd64 and arm64 on Cloudsmith: nginx-module-mzstd for nginx.org's stable and mainline nginx, on every distribution release nginx.org packages for, and libnginx-mod-http-mzstd and nginx-mod-mzstd for the nginx of Debian, Ubuntu, RHEL-compatible distributions and Fedora, from 1.25.4 on. See "Installation from packages" in the README. — Thanks @Mnwa!

v0.2.2

Patch changes

• Bump zstd with perf optimizations

v0.2.1

Patch changes

• 03f5719 Update mzstd to 0.2.5. Levels -5 to 2 compress 5-10% faster, and levels 13-15 up to 21% faster. A response compressed without a dictionary allocates about 8 KB less for its encoder. Output is unchanged. — Thanks @Mnwa!

v0.2.0

Minor changes

• 4834385 zstd_dict_file now compresses responses with a dictionary; before, it was ignored. It accepts Zstandard-format dictionaries (such as zstd --train writes) and raw-content ones, as tokers/zstd-nginx-module does, and also works in server and location blocks, where zstd_dict_file off turns it off. The file is read and checked when the configuration is loaded, so a missing or malformed dictionary now fails the configuration instead of logging a warning. Each file is prepared once for every level and window that locations use it with. Responses with a dictionary are compressed in the event loop, also when zstd_thread_pool is set.

  Update mzstd to 0.2.4. Responses compressed on a thread pool change: each segment now loads the input before it as history instead of compressing it again, up to 27% faster at levels 1-5 and 18-22, with frames up to 0.3% larger. Serial output is unchanged. — Thanks @Mnwa!

v0.1.2

Patch changes

• 146f881 Update mzstd to 0.2.3. Compression is faster at every level: the fast and double-fast match finders (levels up to 4), the lazy finders and the optimal parser (levels 16–22) do less work per byte, and sequence coding executes fewer instructions. Output is unchanged and still byte-identical to zstd 1.5.7. — Thanks @Mnwa!

v0.1.1

Patch changes

• 935b72a Update mzstd to 0.2.2. The encoder now writes straight into uninitialized output buffers, so each new zstd_buffers buffer is no longer zeroed first. A response whose body does not match its Content-Length is still aborted, now by the encoder itself. Output is unchanged. — Thanks @Mnwa!

v0.1.0

Minor changes

• b4da15a Initial release: ngx_http_zstd_filter_module and ngx_http_zstd_static_module, a drop-in replacement for tokers/zstd-nginx-module built on ngx-rust and the pure-Rust mzstd encoder. Supports its directives (zstd, zstd_types, zstd_buffers, zstd_comp_level including negative levels, zstd_min_length, zstd_static; zstd_dict_file is accepted and ignored) and the $zstd_ratio variable, and adds those of ngx_mbrotli: zstd_window, zstd_reuse_memory, zstd_retention_policy, zstd_idle_encoders, zstd_thread_pool, zstd_parallel_segment_size, zstd_parallel_tasks and zstd_parallel_compressors. The window is limited to 8 MB, as RFC 9659 requires and browsers enforce, so levels 20–22 no longer produce responses Chrome rejects. With ngx_mbrotli loaded too, zstd is preferred when a client accepts both encodings. — Thanks @Mnwa for your first contribution 🎉!