From 944907bfeeb45bb650907f0d7277f54024c69a86 Mon Sep 17 00:00:00 2001
From: Nick Sweeting <git@sweeting.me>
Date: Wed, 2 Sep 2026 23:56:46 -0700
Subject: [PATCH] Keep OpenCode proxy traffic out of recovery

---
 archivebox/opencode/views.py            | 44 ++++++++++++++++---------
 archivebox/tests/test_opencode_agent.py | 25 ++++++++++++++
 2 files changed, 53 insertions(+), 16 deletions(-)

diff --git a/archivebox/opencode/views.py b/archivebox/opencode/views.py
index 286a54e9..1f8770c2 100644
--- a/archivebox/opencode/views.py
+++ b/archivebox/opencode/views.py
@@ -105,6 +105,7 @@ def _stop_owned_process(process: subprocess.Popen | None = None) -> None:
     if owned_process is None:
         return
     if owned_process.poll() is None:
+        _signal_owned_process(owned_process, signal.SIGCONT)
         _signal_owned_process(owned_process, signal.SIGTERM)
         try:
             owned_process.wait(timeout=5)
@@ -369,11 +370,15 @@ def _ensure_default_session(settings: dict) -> str:
     return session_id
 
 
-def _health(settings: dict) -> bool:
+def _owned_process_running() -> bool:
+    return _PROCESS is not None and _PROCESS.poll() is None
+
+
+def _health(settings: dict, timeout: float = 2) -> bool:
     try:
         response = requests.get(
             f"{settings['origin']}/global/health",
-            timeout=2,
+            timeout=timeout,
         )
         return response.status_code == 200
     except requests.RequestException:
@@ -399,15 +404,20 @@ def _ensure_opencode(settings: dict) -> tuple[bool, str]:
     workdir = settings["workdir"].resolve()
 
     with _PROCESS_LOCK:
-        if _health(settings):
-            return True, ""
-        if _PROCESS is not None and _PROCESS.poll() is None:
+        if _owned_process_running():
             deadline = time.monotonic() + min(_PROCESS_HEALTH_GRACE, settings["timeout"])
-            while time.monotonic() < deadline and _PROCESS.poll() is None:
-                if _health(settings):
+            while _owned_process_running():
+                remaining = deadline - time.monotonic()
+                if remaining <= 0:
+                    break
+                if _health(settings, timeout=min(2, remaining)):
                     return True, ""
-                time.sleep(0.25)
+                remaining = deadline - time.monotonic()
+                if remaining > 0:
+                    time.sleep(min(0.25, remaining))
             _stop_owned_process(_PROCESS)
+        elif _health(settings):
+            return True, ""
 
         try:
             binary, git_binary, binary_env = _resolve_binary(
@@ -609,11 +619,12 @@ async def _event_chunks(
     params: tuple[tuple[str, str], ...],
     headers: dict[str, str],
 ):
-    ok, error = await sync_to_async(_ensure_opencode, thread_sensitive=False)(settings)
-    if not ok:
-        _LOGGER.warning("OpenCode event stream unavailable: %s", error)
-        yield b'event: error\ndata: {"error":"OpenCode upstream unavailable"}\n\n'
-        return
+    if not _owned_process_running():
+        ok, error = await sync_to_async(_ensure_opencode, thread_sensitive=False)(settings)
+        if not ok:
+            _LOGGER.warning("OpenCode event stream unavailable: %s", error)
+            yield b'event: error\ndata: {"error":"OpenCode upstream unavailable"}\n\n'
+            return
 
     timeout = httpx.Timeout(settings["timeout"], read=None)
     url = _proxy_url(settings, path)
@@ -735,9 +746,10 @@ def opencode_proxy_view(request: HttpRequest, path: str | None = None):
         response.headers["X-Accel-Buffering"] = "no"
         return response
 
-    ok, error = _ensure_opencode(settings)
-    if not ok:
-        return _proxy_error_response(error)
+    if path == "global/health" or not _owned_process_running():
+        ok, error = _ensure_opencode(settings)
+        if not ok:
+            return _proxy_error_response(error)
 
     try:
         method = request.method or "GET"
diff --git a/archivebox/tests/test_opencode_agent.py b/archivebox/tests/test_opencode_agent.py
index c369f100..2bb6b34c 100644
--- a/archivebox/tests/test_opencode_agent.py
+++ b/archivebox/tests/test_opencode_agent.py
@@ -317,7 +317,9 @@ def test_opencode_preserves_a_transiently_unhealthy_owned_process(live_opencode)
     try:
         with ThreadPoolExecutor(max_workers=1) as executor:
             resumed = executor.submit(resume_process)
+            started_at = time.monotonic()
             ok, error = views._ensure_opencode(live_opencode.settings)
+            elapsed = time.monotonic() - started_at
             resumed.result()
     finally:
         views._signal_owned_process(process, signal.SIGCONT)
@@ -325,6 +327,29 @@ def test_opencode_preserves_a_transiently_unhealthy_owned_process(live_opencode)
     assert ok, error
     assert views._PROCESS is process
     assert process.poll() is None
+    assert elapsed < views._PROCESS_HEALTH_GRACE
+
+
+def test_opencode_proxy_does_not_wait_for_recovery_lock(admin_client, live_opencode):
+    from archivebox.opencode import views
+
+    workdir = quote(str(live_opencode.config.data_dir.resolve()))
+    executor = ThreadPoolExecutor(max_workers=1)
+    views._PROCESS_LOCK.acquire()
+    try:
+        request = executor.submit(
+            admin_client.get,
+            f"/admin/agent/opencode/path?directory={workdir}",
+            HTTP_HOST=ADMIN_TEST_HOST,
+            HTTP_SEC_FETCH_SITE="same-origin",
+        )
+        response = request.result(timeout=5)
+    finally:
+        views._PROCESS_LOCK.release()
+        executor.shutdown(wait=True)
+
+    assert response.status_code == 200
+    assert str(live_opencode.config.data_dir.resolve()).encode() in response.content
 
 
 def test_opencode_proxy_sse_response_is_unbuffered(admin_client, live_opencode):