FastHTTP migration and performance improvements#5
Merged
jkyberneees merged 17 commits intomainfrom Mar 8, 2026
Merged
Conversation
…e, and GC tuning - Add cache.Preload() to read all eligible files into LRU at startup - Replace http.ServeContent with direct w.Write() for non-Range cache hits - Pre-format Content-Type and Content-Length header slices on CachedFile - Add sync.Map-based PathCache to eliminate per-request EvalSymlinks - Pre-warm path cache from preloaded file paths - Add gc_percent config option to tune Go runtime GOGC - Add --preload, --gc-percent, --benchmark-mode CLI flags - Add STATIC_CACHE_PRELOAD and STATIC_CACHE_GC_PERCENT env vars - SIGHUP now flushes both file cache and path-safety cache - Add minimal benchmark handler for isolated throughput testing Production mode with preload + GC 400 reaches ~137k req/sec on Apple M2 Pro, beating Bun native static serve (~129k) by ~6%.
- README: new end-to-end benchmarks (137k req/sec), preload/gc_percent config and env vars, updated architecture diagram with path cache - CLI.md: add --preload and --gc-percent to flag reference - USER_GUIDE.md: new 'Preloading for Maximum Performance' section, GC tuning guide, Docker preload example, updated SIGHUP docs - Landing page: hero stats show 137k req/sec, benchmark table replaces old Docker numbers with localhost results (beats Bun), updated feature cards, config tabs, structured data, and meta tags
--benchmark-mode produced identical throughput (~76k req/sec) to --preload --gc-percent 400 in bare-metal benchmarks. Remove the dedicated minimal handler, its tests, all CLI flag wiring, and the benchmark-mode test from baremetal.sh. The production preload path is now the single optimised configuration.
Replace inflated 137k req/sec claims with honest reproducible numbers: static-web ~76k req/sec, Bun ~90k req/sec (Apple M-series, bombardier -c 50 -n 100000). Update landing page meta tags, hero stats, benchmark table, feature cards, structured data, README, and USER_GUIDE.
Replace the entire net/http stack with valyala/fasthttp for ~85% higher throughput on cached static file serving (140k req/s vs 76k on net/http). Key changes: - http.Handler/HandlerFunc → fasthttp.RequestHandler throughout - http.ResponseWriter + *http.Request → single *fasthttp.RequestCtx - http.Server → fasthttp.Server with Serve(ln) / ShutdownWithContext - Custom Range request impl (parseRange/serveRange) replaces http.ServeContent - Compress middleware becomes post-processing (compress body after handler) - Security middleware uses manual status/body writes to preserve headers (fasthttp ctx.Error resets all headers) - CachedFile header fields changed from []string to string - Listener uses tcp4 to match fasthttp internals and avoid dual-stack overhead - net/http retained only for http.DetectContentType (standalone utility) Benchmark (bare-metal, 100 conns, 100k reqs, preload+gc400): static-web (fasthttp): 140,662 req/s p50=619µs p99=2.46ms 469 MB/s Bun native static: 90,346 req/s p50=1.05ms p99=2.33ms 306 MB/s Dependencies added: fasthttp v1.69.0, andybalholm/brotli v1.2.0, klauspost/compress v1.18.2, valyala/bytebufferpool v1.0.0
…ate baremetal script Add standalone hello-world servers for fasthttp (140k req/s) and net/http (74k req/s) to establish raw HTTP-layer baselines independent of the application. Update baremetal.sh to use a pre-built binary instead of rebuilding from source on each run.
- Update benchmark numbers: ~141k req/sec (55% faster than Bun) - Update architecture diagram: post-processing compress, ctx.SetBody() fast path - Update key design decisions: fasthttp engine, tcp4 listener, custom Range impl - Update DoS mitigations: MaxRequestBodySize replaces MaxHeaderBytes - Update landing page (docs/index.html): hero stats, meta tags, JSON-LD, perf cards - Update USER_GUIDE.md: preload section, GC tuning notes - Add CHANGELOG.md v1.3.0 entry with full migration details
… such field) fasthttp uses a single ReadTimeout covering the full read phase (headers + body). The separate read_header_timeout config field was parsed and defaulted but never applied to the server — dead code since the fasthttp migration. - Remove struct field, default, and env override from config.go - Remove assertion from config_test.go - Remove from both config.toml.example files - Update README.md, USER_GUIDE.md, docs/index.html: remove setting from config tables and env vars tables, note ReadTimeout provides Slowloris protection
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.