computefence 0.1.2__tar.gz → 0.2.0__tar.gz

{computefence-0.1.2 → computefence-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: computefence
-Version: 0.1.2
+Version: 0.2.0
 Summary: Pre-flight validation for GPU training runs on rented infrastructure
 License-Expression: MIT
 Requires-Python: >=3.8

computefence-0.2.0/computefence/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.2.0"

computefence-0.2.0/computefence/checks/storage.py

@@ -0,0 +1,136 @@
+import os
+import shutil
+from pathlib import Path
+
+# Persistent volume mount points used by the common GPU rental providers.
+MOUNT_CANDIDATES = ["/workspace", "/runpod-volume", "/vast"]
+
+GB = 1024 ** 3
+BLOCKER_BELOW_GB = 5
+WARNING_BELOW_GB = 20
+DISK_FIX = "Free up disk space or attach a larger volume before launching"
+
+
+def check_storage():
+    results = []
+
+    # Read HuggingFace cache environment variables
+    hf_home = os.environ.get("HF_HOME")
+    hf_hub_cache = os.environ.get("HF_HUB_CACHE")
+    xdg_cache = os.environ.get("XDG_CACHE_HOME")
+
+    # Detect mounted persistent volumes
+    mounted_volumes = []
+    for path in MOUNT_CANDIDATES:
+        if Path(path).exists():
+            mounted_volumes.append(path)
+
+    # Check HF_HOME
+    if hf_home is None:
+        if mounted_volumes:
+            results.append({
+                "status": "warn",
+                "message": f"HF_HOME is not set. Mounted volume detected at {mounted_volumes[0]} but HuggingFace will cache to default local path.",
+                "fix": f"export HF_HOME={mounted_volumes[0]}/.cache/huggingface"
+            })
+        else:
+            results.append({
+                "status": "warn",
+                "message": "HF_HOME is not set. HuggingFace will use default local cache.",
+                "fix": "Set HF_HOME to a persistent volume path before training"
+            })
+    elif hf_home.startswith("/root") or hf_home.startswith("/tmp"):
+        results.append({
+            "status": "warn",
+            "message": f"HF_HOME appears to point to ephemeral storage: {hf_home}",
+            "fix": f"export HF_HOME=/workspace/.cache/huggingface"
+        })
+    elif mounted_volumes and not any(hf_home.startswith(v) for v in mounted_volumes):
+        results.append({
+            "status": "warn",
+            "message": f"HF_HOME ({hf_home}) does not appear to be on a mounted volume.",
+            "fix": f"export HF_HOME={mounted_volumes[0]}/.cache/huggingface"
+        })
+    else:
+        results.append({
+            "status": "pass",
+            "message": f"HF_HOME is set to {hf_home}"
+        })
+
+    # Report mounted volumes found
+    if mounted_volumes:
+        results.append({
+            "status": "pass",
+            "message": f"Persistent volume(s) detected: {', '.join(mounted_volumes)}"
+        })
+    else:
+        results.append({
+            "status": "warn",
+            "message": "No mounted persistent volumes detected at /workspace, /runpod-volume, or /vast",
+            "fix": "Attach a network volume before training to persist checkpoints and cache"
+        })
+
+    return results
+
+def check_disk_headroom():
+    """Report free space on the workspace volume(s) and the root disk.
+
+    Uses shutil.disk_usage rather than shelling out to df: it returns the same
+    numbers as exact bytes, with no output parsing and no subprocess.
+    """
+    results = []
+
+    # Build the target list, skipping any path that shares a device with one
+    # already queued so a single filesystem is not reported twice.
+    targets = []
+    seen_devices = set()
+    for path in MOUNT_CANDIDATES + ["/"]:
+        candidate = Path(path)
+        if not candidate.exists():
+            continue
+        try:
+            device = candidate.stat().st_dev
+        except OSError:
+            device = None
+        if device is not None:
+            if device in seen_devices:
+                continue
+            seen_devices.add(device)
+        targets.append(candidate)
+
+    for target in targets:
+        label = "Root disk (/)" if str(target) == "/" else f"Volume {target}"
+
+        try:
+            usage = shutil.disk_usage(target)
+        except OSError as e:
+            results.append({
+                "status": "warn",
+                "message": f"{label} — could not read disk usage: {e}",
+                "fix": DISK_FIX
+            })
+            continue
+
+        free_gb = usage.free / GB
+        total_gb = usage.total / GB
+        summary = f"{label} — {free_gb:.1f} GB free of {total_gb:.1f} GB"
+
+        if free_gb < BLOCKER_BELOW_GB:
+            results.append({
+                "status": "fail",
+                "message": f"{summary} (below {BLOCKER_BELOW_GB} GB)",
+                "fix": DISK_FIX
+            })
+        elif free_gb < WARNING_BELOW_GB:
+            results.append({
+                "status": "warn",
+                "message": f"{summary} (below {WARNING_BELOW_GB} GB)",
+                "fix": DISK_FIX
+            })
+        else:
+            results.append({
+                "status": "pass",
+                "message": summary
+            })
+
+    return results

{computefence-0.1.2 → computefence-0.2.0}/computefence/cli.py

@@ -3,7 +3,7 @@ from rich.console import Console
 
 from computefence import __version__
 from computefence.checks.environment import check_environment
-from computefence.checks.storage import check_storage
+from computefence.checks.storage import check_disk_headroom, check_storage
 from computefence.checks.dataset import check_dataset
 
 console = Console()
@@ -53,7 +53,7 @@ def doctor():
     # Run every check up front so the verdict line can be counted and printed
     # above the individual results.
     env_results = check_environment()
-    storage_results = check_storage()
+    storage_results = check_storage() + check_disk_headroom()
     dataset_results = check_dataset(
         dataset_path=dataset,
         input_column=input_column,

{computefence-0.1.2 → computefence-0.2.0}/computefence.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: computefence
-Version: 0.1.2
+Version: 0.2.0
 Summary: Pre-flight validation for GPU training runs on rented infrastructure
 License-Expression: MIT
 Requires-Python: >=3.8

computefence-0.1.2/computefence/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.1.2"

computefence-0.1.2/computefence/checks/storage.py

@@ -1,64 +0,0 @@
-import os
-from pathlib import Path
-
-
-def check_storage():
-    results = []
-
-    # Read HuggingFace cache environment variables
-    hf_home = os.environ.get("HF_HOME")
-    hf_hub_cache = os.environ.get("HF_HUB_CACHE")
-    xdg_cache = os.environ.get("XDG_CACHE_HOME")
-
-    # Detect mounted persistent volumes
-    mounted_volumes = []
-    for path in ["/workspace", "/runpod-volume", "/vast"]:
-        if Path(path).exists():
-            mounted_volumes.append(path)
-
-    # Check HF_HOME
-    if hf_home is None:
-        if mounted_volumes:
-            results.append({
-                "status": "warn",
-                "message": f"HF_HOME is not set. Mounted volume detected at {mounted_volumes[0]} but HuggingFace will cache to default local path.",
-                "fix": f"export HF_HOME={mounted_volumes[0]}/.cache/huggingface"
-            })
-        else:
-            results.append({
-                "status": "warn",
-                "message": "HF_HOME is not set. HuggingFace will use default local cache.",
-                "fix": "Set HF_HOME to a persistent volume path before training"
-            })
-    elif hf_home.startswith("/root") or hf_home.startswith("/tmp"):
-        results.append({
-            "status": "warn",
-            "message": f"HF_HOME appears to point to ephemeral storage: {hf_home}",
-            "fix": f"export HF_HOME=/workspace/.cache/huggingface"
-        })
-    elif mounted_volumes and not any(hf_home.startswith(v) for v in mounted_volumes):
-        results.append({
-            "status": "warn",
-            "message": f"HF_HOME ({hf_home}) does not appear to be on a mounted volume.",
-            "fix": f"export HF_HOME={mounted_volumes[0]}/.cache/huggingface"
-        })
-    else:
-        results.append({
-            "status": "pass",
-            "message": f"HF_HOME is set to {hf_home}"
-        })
-
-    # Report mounted volumes found
-    if mounted_volumes:
-        results.append({
-            "status": "pass",
-            "message": f"Persistent volume(s) detected: {', '.join(mounted_volumes)}"
-        })
-    else:
-        results.append({
-            "status": "warn",
-            "message": "No mounted persistent volumes detected at /workspace, /runpod-volume, or /vast",
-            "fix": "Attach a network volume before training to persist checkpoints and cache"
-        })
-
-    return results