config: per-engine env overrides (RADV driver tuning for Polaris hangs)

New server.engine_env_overrides (engine name → {VAR: val}), merged into
the engine's process env at spawn. Lets low-level driver knobs be set
without hardcoding — set radeon → RADV_DEBUG=syncshaders to serialize
RADV shader dispatch and test whether the Polaris compute-ring async
race (ring comp_x.y.z timeout) is what wedges the RX 580 under sustained
Vulkan compute.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014S8VtAvG499SsCbeESRK7V

codai/__init__.py

@@ -16,7 +16,7 @@
 
 # Canonical product version for CoderAI — single source of truth. Both the API
 # metadata and the admin web UI read from here.
-__version__ = "0.1.55"
+__version__ = "0.1.56"
 
 # Configure the CUDA caching allocator BEFORE torch is imported anywhere.
 # expandable_segments lets the allocator return freed pages to the driver even

codai/config.py

@@ -47,6 +47,11 @@ class ServerConfig:
     # Levels: auto | low | high | manual | profile_standard | profile_peak.
     # Best-effort — an unprivileged container logs the manual host command.
     dpm_force_performance_level_overrides: dict = field(default_factory=dict)
+    # Extra environment variables injected per engine (engine name → {VAR: val}),
+    # merged into that engine's process env at spawn. For low-level driver tuning
+    # without hardcoding — e.g. {"radeon": {"RADV_DEBUG": "syncshaders"}} to
+    # serialize RADV shader dispatch and avoid Polaris async compute-ring hangs.
+    engine_env_overrides: dict = field(default_factory=dict)
     # ─── Frontend/engine split ───────────────────────────────────────────────
     # coderai boots a thin, always-responsive *front* reverse proxy on the public
     # host/port and supervises one or more *engine* subprocesses (which do all

codai/frontproxy/engine_supervisor.py

@@ -347,6 +347,19 @@ class EngineSupervisor:
         # values are honoured (e.g. CUDA_VISIBLE_DEVICES="" hides all CUDA cards).
         for k, v in (engine.env or {}).items():
             env[str(k)] = str(v)
+        # Config-driven per-engine env overrides (low-level driver tuning, e.g.
+        # RADV_DEBUG=syncshaders on the radeon). Applied AFTER the auto-detected
+        # block so the user's value wins.
+        try:
+            _eov = (getattr(self.config.server, "engine_env_overrides", None)
+                    or {}).get(engine.name) or {}
+            for k, v in _eov.items():
+                env[str(k)] = str(v)
+            if _eov:
+                print(f"[front] {engine.name}: env overrides "
+                      f"{ {k: v for k, v in _eov.items()} }", flush=True)
+        except Exception:
+            pass
         # The global host-RAM cap (offload.max_ram_gb) is SHARED across all engines,
         # not split: tell each engine the front's PID so it measures the whole
         # fleet's RAM (front + every engine + workers) against the one cap.