RadixTree.searchLongest previously allocated &MatchResult{} on the
heap every time it encountered a handler-bearing node during tree
traversal — potentially N allocations per lookup with only 1 surviving.
Changes:
- searchLongest now tracks best *RadixNode (stack pointer) instead of
allocating MatchResult at every handler node
- FindLongestPrefix allocates a single pooled MatchResult via sync.Pool
only when returning a match
- Add ReleaseMatchResult() for callers to return MatchResult to pool
- LocationEngine.Match releases pooled results after use
Result: 0 B/op, 0 allocs/op on tree traversal benchmarks (sequential
and parallel).
Remove sync.Mutex from Target's failure tracking hot path:
- failCount int64 → atomic.Int64
- failedUntil int64 → atomic.Int64
- IsAvailable(): Load() instead of Lock/Unlock, eliminating mutex
from every per-target check in every LB Select() call
- RecordFailure(): Add(1) + Store() instead of Lock/Unlock
- RecordSuccess(): Store(0) instead of Lock/Unlock
This removes the only mutex acquisition in the load balancing
selection path. Every Select() call iterates targets and calls
IsAvailable() on each — previously acquiring failMu per target
when MaxFails > 0. Now fully lock-free.
- Add comments for lua/api_log.go HTTP status codes and log levels
- Add comments for variable/builtin.go and ssl.go constants
- Add comments for utils/httperror.go error variables
- Add comments for matcher/matcher.go location types
- Add comments for compression/compression.go algorithms
- Include author attribution (xfy)
- proxyDebugLog: move Enabled() guard to call sites to avoid allocations
- proxyDebugLog: add default case for unsupported types
- static routes: remove unintended regex support to match original behavior
- Fix FD leak in DupListener: close *os.File after net.FileListener
- Add cleanup of partially-duped listeners on DupListener failure
- Make reload timeout configurable via shutdown.reload_timeout
- Handle filepath.Abs errors in processIncludes instead of ignoring
- Use net.ParseIP in isAnyAddr for robust IPv6 support
Replace depth-only detection with path-based visited set tracking.
Detects cycles immediately on first revisit instead of after 10 depth
iterations. Supports diamond patterns (A->B->shared, A->C->shared)
via backtracking. Add self-include and diamond tests. Document that
only servers/stream/variables are merged in defaults.go.
Make AddNamed return *ConflictError for consistency with other Add*
methods so handleRegistrationError treats named location conflicts as
warnings instead of fatal errors. Add tests for handleRegistrationError
covering both conflict and fatal error paths.
Remove VHost fallback during graceful upgrade. Serialize listener
creation before parallel router/middleware setup to prevent concurrent
inherited listener consumption. Fix tcpAddrMatch to match when either
side is any-addr (0.0.0.0/::).
createListener now checks pre-set s.listeners (Path 2) for hot reload,
not just upgradeManager.IsChild() (Path 1). Add DupListener to dup FDs
so old/new servers own independent listeners. Reload rebuilds HTTP/2
and HTTP/3. Add matchInheritedListener with TCP any-addr matching.
Add requiresFullRestart with VHost server count detection.
