feat(evpp): implement LB_IpHash load balance strategy

[ithewei]

What

Implements the LB_IpHash load balance strategy, which was already declared in load_balance_e (event/hloop.h) but left as // Not Implemented in EventLoopThreadPool::nextLoop().

With LB_IpHash, all connections from the same client IP are consistently dispatched to the same worker loop.

Changes

• base/hsocket.{h,c}: add sockaddr_ip_hash() (C layer, so C users can also use it). FNV-1a over the IP bytes only — port excluded — so the same client IP maps to a stable value. IPv4 hashes 4 bytes, IPv6 hashes 16 bytes.
• evpp/EventLoopThreadPool.h: nextLoop() gains an optional uint32_t hash = 0 arg (backward compatible). LB_IpHash selects idx = hash % numLoops; the previously empty else now falls back to RoundRobin instead of always returning loop 0.
• evpp/TcpServer.h: onAccept() computes the peer IP hash via hio_peeraddr() when the strategy is LB_IpHash.
• docs/cn/TcpServer.md: document the available load balance options.

Drive-by bug fixes in base/hsocket.c

Two pre-existing bugs spotted while working in this file:

• sockaddr_compare: the IPv6 branch compared only 4 bytes (sizeof(struct in_addr)) instead of the full 16-byte in6_addr, so IPv6 addresses differing beyond the first 4 bytes were incorrectly treated as equal.
• ResolveAddr: the IPv6-literal branch was missing a return 0;, causing a redundant (potentially blocking) getaddrinfo() call that then overwrote the already-parsed literal result.

Notes

• LB_UrlHash is still unimplemented: at accept time there is no application-layer data, so the URL is unavailable. It belongs to the HTTP layer, not TcpServer. The else fallback keeps behavior safe (RoundRobin) if it is ever selected.

Testing

• make libhv ✅
• make evpp ✅ (all C++ test binaries link)
• IpHash dispatch behavior (multiple workers + multiple client IPs mapping to a stable loop) is intended for manual verification.

[Copilot]

🔵 Needs a closer look

An automated behavioral test for LB_IpHash is still needed.

Pull request overview

Implements client-IP-based load balancing for TcpServer, with socket hashing, IPv6 fixes, and documentation updates.

Changes:

• Adds FNV-1a hashing for IPv4 and IPv6 addresses.
• Implements LB_IpHash routing with RoundRobin fallback.
• Fixes IPv6 comparison and literal address resolution.
• Documents load-balance strategies.

File summaries

File	Summary
evpp/TcpServer.h	Computes peer-IP hashes during accept.
evpp/EventLoopThreadPool.h	Adds hash-based dispatch and fallback behavior.
docs/cn/TcpServer.md	Documents load-balance strategies.
base/hsocket.h	Declares the IP hash helper.
base/hsocket.c	Implements hashing and socket-address fixes.

Review details

Suppressed comments (2)

evpp/EventLoopThreadPool.h:32

• The public API reference in docs/cn/EventLoop.md:92 still declares nextLoop(load_balance_e lb = LB_RoundRobin) and omits both the new hash parameter and LB_IpHash. A caller following that documentation cannot provide the key and will silently use the default hash 0, routing every LB_IpHash call to worker 0; update that reference and its strategy list with this API change.

    EventLoopPtr nextLoop(load_balance_e lb = LB_RoundRobin, uint32_t hash = 0) {

evpp/EventLoopThreadPool.h:48

• The new LB_IpHash branch has no automated behavioral test: EventLoopThreadPool_test.cpp only prints selected loop IDs, and CI does not execute that binary. Add a deterministic test covering equal IP hashes (including different ports) and hashes that map to different workers before relying on manual verification.

            idx = hash % numLoops;

• Files reviewed: 5/5 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.