From 74df7d2494c83e15af623661ffc1253bbe3cc6c7 Mon Sep 17 00:00:00 2001
From: autcir <autcir@gmail.com>
Date: Sat, 10 Oct 2026 13:47:22 +0200
Subject: [PATCH] fix(proxy): HTTP/2 verso l'upstream disattivabile con
 OPENVIDIA_HTTP2=0

Con HTTP/2 un solo GOAWAY di NVIDIA (ConnectionTerminated) chiude insieme
tutti gli stream in corso: su soliso, con 5 sessioni opencode in parallelo,
ogni chiusura faceva fallire tutte le sessioni con
"Invalid openvidia/openai-compatible-chat stream event". Con
OPENVIDIA_HTTP2=0 ogni richiesta ha la sua connessione HTTP/1.1 e una
chiusura costa uno stream solo. Il default resta HTTP/2.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 openvidia/config.py          | 11 +++++++++++
 openvidia/proxy_app.py       |  2 +-
 tests/test_upstream_http2.py | 17 +++++++++++++++++
 3 files changed, 29 insertions(+), 1 deletion(-)
 create mode 100644 tests/test_upstream_http2.py

diff --git a/openvidia/config.py b/openvidia/config.py
index f9b1346..6b4953d 100644
--- a/openvidia/config.py
+++ b/openvidia/config.py
@@ -77,6 +77,17 @@ def httpx_timeout_kwargs() -> dict[str, float]:
 UPSTREAM_CHAT = UPSTREAM_BASE + "chat/completions"
 
 
+def upstream_http2() -> bool:
+    """Whether upstream requests share one multiplexed HTTP/2 connection.
+
+    With HTTP/2 a single GOAWAY from the upstream (``ConnectionTerminated``)
+    kills every in-flight stream at once, and a client mid-stream sees a
+    broken event. ``OPENVIDIA_HTTP2=0`` gives each request its own HTTP/1.1
+    connection, so a closed connection costs one stream, not all of them.
+    """
+    return os.environ.get("OPENVIDIA_HTTP2", "1").strip().lower() not in ("0", "false", "no", "off")
+
+
 def outbound_proxy() -> str | None:
     """Return outbound HTTP/SOCKS5 proxy URL for upstream requests.
 
diff --git a/openvidia/proxy_app.py b/openvidia/proxy_app.py
index 0e7e51a..c8a76f3 100644
--- a/openvidia/proxy_app.py
+++ b/openvidia/proxy_app.py
@@ -313,7 +313,7 @@ def create_app(state: ProxyState, web_dir: Path | None = None) -> FastAPI:
     limits = httpx.Limits(max_keepalive_connections=100, max_connections=200, keepalive_expiry=30.0)
     proxy_url = config.outbound_proxy()
     client = httpx.AsyncClient(
-        http2=True,
+        http2=config.upstream_http2(),
         proxy=proxy_url if proxy_url else None,
         limits=limits,
         timeout=httpx.Timeout(**config.httpx_timeout_kwargs()),
diff --git a/tests/test_upstream_http2.py b/tests/test_upstream_http2.py
new file mode 100644
index 0000000..68c7f28
--- /dev/null
+++ b/tests/test_upstream_http2.py
@@ -0,0 +1,17 @@
+from openvidia import config
+
+
+def test_http2_is_the_default(monkeypatch):
+    monkeypatch.delenv("OPENVIDIA_HTTP2", raising=False)
+    assert config.upstream_http2() is True
+
+
+def test_http2_can_be_switched_off(monkeypatch):
+    for value in ("0", "false", "no", "off", " OFF "):
+        monkeypatch.setenv("OPENVIDIA_HTTP2", value)
+        assert config.upstream_http2() is False
+
+
+def test_any_other_value_keeps_http2(monkeypatch):
+    monkeypatch.setenv("OPENVIDIA_HTTP2", "1")
+    assert config.upstream_http2() is True