From c59f028150bcc76fd9147b9302fd5ad8a7fb3092 Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Sun, 15 Mar 2026 06:54:40 +0000
Subject: [PATCH] studio: kill orphaned llama-server processes on startup

When the studio process is killed (SIGTERM/SIGKILL), atexit handlers
may not run in the subprocess orchestrator, leaving llama-server
processes orphaned and holding GPU memory. This caused OOM errors when
trying to load a new model after a studio restart.

On init, LlamaCppBackend now runs pgrep to find and SIGKILL any stale
llama-server processes before starting fresh.
---
 studio/backend/core/inference/llama_cpp.py | 28 ++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index b853bcc5a6..0dde135fd4 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -49,6 +49,7 @@ class LlamaCppBackend:
         self._stdout_lines: list[str] = []
         self._stdout_thread: Optional[threading.Thread] = None
 
+        self._kill_orphaned_servers()
         atexit.register(self._cleanup)
 
     # ── Properties ────────────────────────────────────────────────
@@ -706,6 +707,33 @@ class LlamaCppBackend:
                 self._stdout_thread.join(timeout = 2)
                 self._stdout_thread = None
 
+    @staticmethod
+    def _kill_orphaned_servers():
+        """Kill any orphaned llama-server processes from previous studio runs."""
+        import os
+        import signal
+
+        try:
+            result = subprocess.run(
+                ["pgrep", "-f", "llama-server"],
+                capture_output = True, text = True, timeout = 5,
+            )
+            if result.returncode != 0:
+                return
+            for line in result.stdout.strip().splitlines():
+                pid = int(line.strip())
+                if pid == os.getpid():
+                    continue
+                try:
+                    os.kill(pid, signal.SIGKILL)
+                    logger.info(f"Killed orphaned llama-server process (pid={pid})")
+                except ProcessLookupError:
+                    pass
+                except PermissionError:
+                    pass
+        except Exception:
+            pass
+
     def _cleanup(self):
         """atexit handler to ensure llama-server is terminated."""
         self._kill_process()