fix: run hyperhealth migrations once, and never stamp on failure

agents/hyperhealth/Dockerfile

@@ -46,5 +46,8 @@ EXPOSE 8090
 HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=5 \
     CMD curl -f http://localhost:8090/health || exit 1
 
-# Run migrations on container start: upgrade head, fallback to stamp if needed
-CMD ["sh", "-c", "alembic upgrade head || alembic stamp 001 && exec uvicorn main:app --host 0.0.0.0 --port 8090 --workers 2"]
+# Migrate once, here, before uvicorn forks its workers. Never `|| alembic stamp`:
+# 001 is head, so stamping on failure marks an unmigrated DB as fully migrated and
+# every later `upgrade head` becomes a no-op. On failure we exit; compose has
+# restart: unless-stopped and waits for postgres to be healthy, so it retries.
+CMD ["sh", "-c", "alembic upgrade head && exec uvicorn main:app --host 0.0.0.0 --port 8090 --workers 2"]

agents/hyperhealth/main.py

@@ -7,7 +7,6 @@
 import os
 import sys
 import uuid
-import subprocess
 from contextlib import asynccontextmanager
 from datetime import datetime, timedelta
 from pathlib import Path
@@ -105,23 +104,11 @@ def _get_session_factory():
 async def lifespan(app: FastAPI):
     log.info("hyperhealth.startup", environment=ENVIRONMENT)
     engine = _get_engine()
-
-    # Run Alembic migrations on startup
-    try:
-        result = subprocess.run(
-            ["alembic", "upgrade", "head"],
-            cwd=_HERE,
-            capture_output=True,
-            text=True,
-            timeout=30
-        )
-        if result.returncode == 0:
-            log.info("hyperhealth.migrations_completed")
-        else:
-            log.warning("hyperhealth.migrations_warning", stderr=result.stderr)
-    except Exception as e:
-        log.warning("hyperhealth.migrations_error", error=str(e))
-
+
+    # Migrations run once in the container CMD, before uvicorn forks its
+    # workers. lifespan executes per worker, so running alembic here meant
+    # concurrent `upgrade head` calls racing each other on every boot.
+
     log.info("hyperhealth.db_ready")
     await _ping_healer()
     yield