gpu-dev 0.6.3__tar.gz → 0.6.5__tar.gz

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpu-dev
-Version: 0.6.3
+Version: 0.6.5
 Summary: CLI tool for PyTorch GPU developer server reservations
 Author: PyTorch Team
 Requires-Python: >=3.10

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/cli-tools/gpu-dev-cli/gpu_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpu-dev
-Version: 0.6.3
+Version: 0.6.5
 Summary: CLI tool for PyTorch GPU developer server reservations
 Author: PyTorch Team
 Requires-Python: >=3.10

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/cli-tools/gpu-dev-cli/gpu_dev_cli/disks.py

@@ -154,12 +154,8 @@ def list_disks(user_id: str, config: Config) -> List[Dict]:
     List all disks for a user.
     Returns list of disk info dicts with: name, size, last_used, created_at, snapshot_count, in_use, reservation_id
     """
-    ec2_client = get_ec2_client(config)
-    dynamodb = get_dynamodb_resource(config)
-
-    # Query DynamoDB disks table for this user's disks (with pagination)
-    disks_table_name = config.disks_table if hasattr(config, 'disks_table') else f"{config.queue_name.rsplit('-', 1)[0]}-disks"
-    disks_table = dynamodb.Table(disks_table_name)
+    dynamodb = config.dynamodb
+    disks_table = dynamodb.Table(config.disks_table)
 
     dynamodb_disks = []
     response = disks_table.query(
@@ -208,9 +204,6 @@ def list_disks(user_id: str, config: Config) -> List[Dict]:
         is_deleted = disk_item.get('is_deleted', False)
         delete_date = disk_item.get('delete_date')
 
-        # Check current in_use status (check dynamically from reservations table)
-        is_in_use, reservation_id = get_disk_in_use_status(disk_name, user_id, config)
-
         disks.append({
             'name': disk_name,
             'size_gb': size_gb,
@@ -219,13 +212,37 @@ def list_disks(user_id: str, config: Config) -> List[Dict]:
             'last_used': last_used,
             'snapshot_count': snapshot_count,
             'pending_snapshot_count': pending_snapshot_count,
-            'in_use': is_in_use,
+            'in_use': bool(disk_item.get('in_use', False)),
             'is_backing_up': is_backing_up,
-            'reservation_id': reservation_id,
+            'reservation_id': str(disk_item.get('attached_to_reservation', '')) or None,
             'is_deleted': is_deleted,
             'delete_date': delete_date,
         })
 
+    # Batch check: find all active reservations with disk_name set (single query)
+    try:
+        reservations_table = dynamodb.Table(config.reservations_table)
+        active_disks = {}
+        for status in ["active", "preparing", "queued", "pending"]:
+            resp = reservations_table.query(
+                IndexName="UserStatusIndex",
+                KeyConditionExpression="user_id = :uid AND #s = :status",
+                ExpressionAttributeNames={"#s": "status"},
+                ExpressionAttributeValues={":uid": user_id, ":status": status},
+                ProjectionExpression="reservation_id, disk_name",
+            )
+            for item in resp.get("Items", []):
+                dn = item.get("disk_name")
+                if dn:
+                    active_disks[dn] = str(item.get("reservation_id", ""))[:8]
+
+        for disk in disks:
+            if disk["name"] in active_disks:
+                disk["in_use"] = True
+                disk["reservation_id"] = active_disks[disk["name"]]
+    except Exception:
+        pass
+
     # Sort by last_used (most recent first)
     disks.sort(key=lambda d: d['last_used'] or datetime.min.replace(tzinfo=timezone.utc), reverse=True)
 

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gpu-dev"
-version = "0.6.3"
+version = "0.6.5"
 description = "CLI tool for PyTorch GPU developer server reservations"
 authors = [{name = "PyTorch Team"}]
 readme = "cli-tools/gpu-dev-cli/README.md"

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/sdk/python/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gpu-dev-sdk"
-version = "0.1.0"
+version = "0.6.5"
 description = "Python SDK for GPU development server reservations"
 readme = "README.md"
 requires-python = ">=3.10"

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/sdk/python/src/gpu_dev/__init__.py

@@ -57,4 +57,4 @@ __all__ = [
     "ExecResult",
 ]
 
-__version__ = "0.1.0"
+__version__ = "0.6.5"

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/sdk/python/src/gpu_dev/_backend/aws.py

@@ -166,7 +166,7 @@ class AwsBackend:
             "no_persistent_disk": params.get("no_persistent_disk", False),
             "github_user": params.get("github_user", ""),
             "preserve_entrypoint": params.get("preserve_entrypoint", False),
-            "version": "0.6.0",
+            "version": __import__("gpu_dev").__version__,
         }
         if params.get("disk_name"):
             message["disk_name"] = params["disk_name"]
@@ -189,13 +189,17 @@ class AwsBackend:
                 return self._item_to_info(item)
             return None
 
-        resp = self._reservations.query(
-            IndexName="UserIndex",
-            KeyConditionExpression="user_id = :uid",
-            FilterExpression="begins_with(reservation_id, :rid)",
-            ExpressionAttributeValues={":uid": user_id, ":rid": reservation_id},
-        )
+        query_kwargs = {
+            "IndexName": "UserIndex",
+            "KeyConditionExpression": "user_id = :uid",
+            "FilterExpression": "begins_with(reservation_id, :rid)",
+            "ExpressionAttributeValues": {":uid": user_id, ":rid": reservation_id},
+        }
+        resp = self._reservations.query(**query_kwargs)
         items = resp.get("Items", [])
+        while not items and "LastEvaluatedKey" in resp:
+            resp = self._reservations.query(**query_kwargs, ExclusiveStartKey=resp["LastEvaluatedKey"])
+            items = resp.get("Items", [])
         if len(items) == 1:
             return self._item_to_info(items[0])
         return None

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/sdk/python/src/gpu_dev/_sync/sandbox.py

@@ -136,11 +136,6 @@ class Sandbox:
         """Owner's user ID."""
         return self._info.user_id
 
-    @property
-    def node_ip(self) -> str | None:
-        """Node public IP."""
-        return self._info.node_ip
-
     @property
     def is_active(self) -> bool:
         """Whether the sandbox is running and ready for commands."""
@@ -322,6 +317,62 @@ class Sandbox:
         except Exception:
             return []
 
+    def timing(self) -> list[dict[str, str | float]]:
+        """Get a performance breakdown of reservation setup.
+
+        Computes time deltas between status transitions to show
+        where time was spent.
+
+        Returns:
+            List of ``{"step": "...", "duration": seconds, "timestamp": "..."}`` dicts.
+
+        Example::
+
+            for step in sandbox.timing():
+                print(f"{step['duration']:5.1f}s  {step['step']}")
+        """
+        from datetime import datetime
+
+        entries = self.logs()
+        if not entries:
+            return []
+
+        result = []
+        prev_ts = None
+        for entry in entries:
+            ts_str = entry.get("timestamp", "")
+            msg = entry.get("message", "")
+            try:
+                ts = datetime.fromisoformat(ts_str)
+            except (ValueError, TypeError):
+                continue
+            if prev_ts:
+                delta = (ts - prev_ts).total_seconds()
+                if delta >= 0.05:
+                    result.append({
+                        "step": msg,
+                        "duration": round(delta, 2),
+                        "timestamp": ts_str,
+                    })
+            else:
+                result.append({
+                    "step": msg,
+                    "duration": 0.0,
+                    "timestamp": ts_str,
+                })
+            prev_ts = ts
+
+        if result and len(entries) >= 2:
+            first_ts = datetime.fromisoformat(entries[0]["timestamp"])
+            last_ts = datetime.fromisoformat(entries[-1]["timestamp"])
+            result.append({
+                "step": "TOTAL",
+                "duration": round((last_ts - first_ts).total_seconds(), 2),
+                "timestamp": "",
+            })
+
+        return result
+
     def pod_logs(self, lines: int = 50) -> str:
         """Fetch container stdout from the running pod via SSH.
 

{gpu_dev-0.6.3 → gpu_dev-0.6.5}/terraform-gpu-devservers/lambda.tf

@@ -186,8 +186,8 @@ resource "aws_lambda_function" "reservation_processor" {
       HOSTED_ZONE_ID                     = local.effective_domain_name != "" ? local.hosted_zone_id : ""
       SSH_DOMAIN_MAPPINGS_TABLE          = local.effective_domain_name != "" ? aws_dynamodb_table.ssh_domain_mappings.name : ""
       SSL_CERTIFICATE_ARN                = local.effective_domain_name != "" ? aws_acm_certificate.wildcard[0].arn : ""
-      LAMBDA_VERSION                     = "0.6.0"
-      MIN_CLI_VERSION                    = "0.6.0"
+      LAMBDA_VERSION                     = "0.6.5"
+      MIN_CLI_VERSION                    = "0.6.5"
       # Comma-separated GPU types that require --spot flag, or "all" for every type.
       # Empty = no spot types (on-demand / reserved). Set per-workspace.
       ASG_NAME_PREFIX                    = "${var.prefix}-gpu-nodes"