From 6cf5c3e5a9bf9225aa0254ecaa5a0552efd845a0 Mon Sep 17 00:00:00 2001
From: Story Crater Bot <19826264+Riotpiaole@users.noreply.github.com>
Date: Fri, 21 Aug 2026 20:06:10 -0700
Subject: [PATCH] fix(server): stop WriteTimeout from killing in-progress LLM
 streams

http.Server.WriteTimeout is an absolute deadline over the whole
response, not an inactivity timeout -- 15s was cutting off SSE streams
from the reasoning model mid-generation, surfacing to clients as a
"terminated" error well before the model finished. Switch to
ReadHeaderTimeout (protects against slow headers without capping body
duration) and raise WriteTimeout to match the edge nginx Ingress's
proxy-read/send-timeout of 3600s.
---
 internal/server/server.go | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/internal/server/server.go b/internal/server/server.go
index 86f59e0..b616f7e 100644
--- a/internal/server/server.go
+++ b/internal/server/server.go
@@ -21,11 +21,17 @@ type Server struct {
 func New(listenAddr string, shutdownTimeout time.Duration, handler http.Handler) *Server {
 	return &Server{
 		httpServer: &http.Server{
-			Addr:         listenAddr,
-			Handler:      handler,
-			ReadTimeout:  15 * time.Second,
-			WriteTimeout: 15 * time.Second,
-			IdleTimeout:  60 * time.Second,
+			Addr:    listenAddr,
+			Handler: handler,
+			// ReadHeaderTimeout (not ReadTimeout) and a long WriteTimeout: both
+			// ReadTimeout and WriteTimeout are absolute deadlines covering the
+			// whole request/response body, not inactivity timeouts -- a 15s
+			// WriteTimeout here was killing in-progress LLM SSE streams (proxy.go's
+			// outbound transport deliberately avoids this same mistake). Mirrors
+			// the edge nginx Ingress's proxy-read/send-timeout of 3600s.
+			ReadHeaderTimeout: 15 * time.Second,
+			WriteTimeout:      1 * time.Hour,
+			IdleTimeout:       60 * time.Second,
 		},
 		shutdownTimeout: shutdownTimeout,
 		healthChecker:   NewHealthChecker(false, false),