fix(disk-check): remove redundant /tmp check in containers

- Simplified disk check to only verify root (/) filesystem
- In containers, /tmp is just a subdirectory of / (same disk)
- Eliminates duplicate log messages with identical results
- Updated tests to verify only root filesystem is checked
- Updated documentation to reflect container behavior
- Reduces check overhead with single disk_usage() call

Author: deanq

docs/serverless/worker_fitness_checks.md

@@ -255,16 +255,15 @@ Memory check passed: 12.00GB available (of 16.00GB total)
 
 ### Disk Space
 
-Verifies adequate disk space on root filesystem and /tmp (common for model downloads).
+Verifies adequate disk space on root filesystem.
 
-Requires free space to be at least a percentage of total disk size, which automatically scales to different machine sizes.
+In containers, the root (/) filesystem is typically the only mount point. The check requires free space to be at least a percentage of total disk size, which automatically scales to different machine sizes.
 
 - **Default**: 10% of total disk must be free
 - **Configure**: `RUNPOD_MIN_DISK_PERCENT=15` (or any percentage 0-100)
 
 What it checks:
 - Root filesystem (/) free space percentage
-- Temporary directory (/tmp) free space percentage
 - Automatic scaling based on total disk size
 
 Scaling examples with 10% default:
@@ -274,7 +273,7 @@ Scaling examples with 10% default:
 
 Example log output:
 ```
-Disk space check passed on /: 50.00GB free (50.0% available)
+Disk space check passed: 50.00GB free (50.0% available)
 ```
 
 ### Network Connectivity

runpod/serverless/modules/rp_system_fitness.py

@@ -121,41 +121,33 @@ def _check_memory_availability() -> None:
 
 def _check_disk_space() -> None:
     """
-    Check disk space availability on root and /tmp.
+    Check disk space availability on root filesystem.
 
+    In containers, root (/) is typically the only filesystem.
     Requires free space to be at least MIN_DISK_PERCENT% of total disk size.
 
     Raises:
         RuntimeError: If insufficient disk space
     """
-    paths_to_check = ["/", "/tmp"]
+    try:
+        usage = shutil.disk_usage("/")
+        total_gb = usage.total / (1024**3)
+        free_gb = usage.free / (1024**3)
+        free_percent = 100 * (free_gb / total_gb)
 
-    for path in paths_to_check:
-        try:
-            usage = shutil.disk_usage(path)
-            total_gb = usage.total / (1024**3)
-            free_gb = usage.free / (1024**3)
-            used_percent = 100 * (1 - free_gb / total_gb)
-            free_percent = 100 * (free_gb / total_gb)
-
-            # Check if free space is below the required percentage
-            if free_percent < MIN_DISK_PERCENT:
-                raise RuntimeError(
-                    f"Insufficient disk space on {path}: {free_gb:.2f}GB free "
-                    f"({free_percent:.1f}%), {MIN_DISK_PERCENT}% required"
-                )
-
-            log.info(
-                f"Disk space check passed on {path}: {free_gb:.2f}GB free "
-                f"({free_percent:.1f}% available)"
+        # Check if free space is below the required percentage
+        if free_percent < MIN_DISK_PERCENT:
+            raise RuntimeError(
+                f"Insufficient disk space: {free_gb:.2f}GB free "
+                f"({free_percent:.1f}%), {MIN_DISK_PERCENT}% required"
             )
-        except FileNotFoundError:
-            # /tmp may not exist on some systems
-            if path == "/":
-                # Root always exists
-                raise
-            # Skip /tmp if it doesn't exist
-            log.debug(f"Path {path} not found, skipping disk check")
+
+        log.info(
+            f"Disk space check passed: {free_gb:.2f}GB free "
+            f"({free_percent:.1f}% available)"
+        )
+    except FileNotFoundError:
+        raise RuntimeError("Could not check disk space: / filesystem not found")
 
 
 async def _check_network_connectivity() -> None:

tests/test_serverless/test_modules/test_system_fitness.py

@@ -125,8 +125,8 @@ def test_insufficient_disk_fails(self, mock_disk_usage):
 
     @patch("runpod.serverless.modules.rp_system_fitness.MIN_DISK_PERCENT", 10.0)
     @patch("shutil.disk_usage")
-    def test_checks_both_root_and_tmp(self, mock_disk_usage):
-        """Test that both root and /tmp are checked."""
+    def test_checks_root_filesystem(self, mock_disk_usage):
+        """Test that root filesystem is checked."""
         mock_usage = MagicMock()
         # 100GB total, 50GB free (50% free) - should pass
         mock_usage.total = 100 * 1024**3
@@ -135,10 +135,8 @@ def test_checks_both_root_and_tmp(self, mock_disk_usage):
 
         _check_disk_space()
 
-        # Verify both paths were checked
-        assert mock_disk_usage.call_count >= 2
-        paths_checked = [call[0][0] for call in mock_disk_usage.call_args_list]
-        assert "/" in paths_checked
+        # Verify root filesystem was checked
+        mock_disk_usage.assert_called_once_with("/")
 
 
 # ============================================================================