experimaestro 2.0.0a8__py3-none-any.whl → 2.0.0b4__py3-none-any.whl

experimaestro/notifications.py

@@ -115,14 +115,27 @@ class Reporter(threading.Thread):
 
     def check_urls(self):
         """Check whether we have new schedulers to notify"""
-        mtime = os.path.getmtime(self.path)
+        # Check if path exists (it might have been deleted during cleanup)
+        if not self.path.exists():
+            return
+
+        try:
+            mtime = os.path.getmtime(self.path)
+        except (OSError, FileNotFoundError):
+            # Path was deleted while we were checking
+            return
+
         if mtime > self.lastcheck:
             for f in self.path.iterdir():
                 self.urls[f.name] = ListenerInformation(f.read_text().strip())
                 logger.info("Added new notification URL: %s", self.urls[f.name].url)
                 f.unlink()
 
-            self.lastcheck = os.path.getmtime(self.path)
+            try:
+                self.lastcheck = os.path.getmtime(self.path)
+            except (OSError, FileNotFoundError):
+                # Path was deleted during iteration
+                return
 
     def run(self):
         logger.info("Running notification thread")
@@ -236,12 +249,21 @@ class Reporter(threading.Thread):
 
 
 def progress(value: float, level=0, desc: Optional[str] = None, console=False):
-    """When called from a running task, report the progress
+    """Report task progress to the experimaestro server.
+
+    Call this function from within a running task to report progress.
+    Progress is displayed in the web UI and TUI monitors.
+
+    Example::
 
-    Args:
-        level: The level (starting from 0)
-        value: The current value
-        desc: An optional description of the current task
+        for i, batch in enumerate(dataloader):
+            train(batch)
+            progress(i / len(dataloader), desc="Training")
+
+    :param value: Progress value between 0.0 and 1.0
+    :param level: Nesting level for nested progress bars (default: 0)
+    :param desc: Optional description of the current operation
+    :param console: If True, also print to console when no server is available
     """
     if TaskEnv.instance().slave:
         # Skip if in a slave process
@@ -255,8 +277,21 @@ def report_eoj():
 
 
 class xpm_tqdm(std_tqdm):
-    """XPM wrapper for experimaestro that automatically reports progress to the
-    server"""
+    """Experimaestro-aware tqdm progress bar.
+
+    A drop-in replacement for ``tqdm`` that automatically reports progress
+    to the experimaestro server. Use this instead of the standard ``tqdm``
+    in your task's ``execute()`` method.
+
+    Example::
+
+        from experimaestro import tqdm
+
+        class MyTask(Task):
+            def execute(self):
+                for batch in tqdm(dataloader, desc="Training"):
+                    train(batch)
+    """
 
     def __init__(self, iterable=None, file=None, *args, **kwargs):
         # Report progress bar
@@ -279,14 +314,29 @@ class xpm_tqdm(std_tqdm):
 
 
 @overload
-def tqdm(**kwargs) -> xpm_tqdm:
-    ...
+def tqdm(**kwargs) -> xpm_tqdm: ...
 
 
 @overload
-def tqdm(iterable: Optional[Iterator[T]] = None, **kwargs) -> Iterator[T]:
-    ...
+def tqdm(iterable: Optional[Iterator[T]] = None, **kwargs) -> Iterator[T]: ...
 
 
 def tqdm(*args, **kwargs):
+    """Create an experimaestro-aware progress bar.
+
+    A drop-in replacement for ``tqdm.tqdm`` that automatically reports progress
+    to the experimaestro server. Use this in task ``execute()`` methods.
+
+    Example::
+
+        from experimaestro import tqdm
+
+        for epoch in tqdm(range(100), desc="Epochs"):
+            for batch in tqdm(dataloader, desc="Batches"):
+                train(batch)
+
+    :param iterable: Iterable to wrap (optional)
+    :param kwargs: Additional arguments passed to tqdm
+    :return: A progress bar iterator
+    """
     return xpm_tqdm(*args, **kwargs)  # type: ignore

experimaestro/progress.py

@@ -106,8 +106,6 @@ class StateFile:
 
 
 class ProgressFileWriter:
-    # TODO: Implement buffering and flushing
-
     def __init__(
         self, task_path: Path, max_entries_per_file: int = DEFAULT_MAX_ENTRIES_PER_FILE
     ):

experimaestro/rpyc.py

@@ -1,5 +1,4 @@
 import atexit
-import shutil
 import tempfile
 from pathlib import Path
 from subprocess import Popen, PIPE, run

experimaestro/run.py

@@ -9,6 +9,7 @@ from typing import List
 import fasteners
 from experimaestro.notifications import progress, report_eoj
 from experimaestro.utils.multiprocessing import delayed_shutdown
+from experimaestro.exceptions import GracefulTimeout
 from .core.types import ObjectType
 from experimaestro.utils import logger
 from experimaestro.core.objects import ConfigInformation
@@ -41,9 +42,6 @@ def run(parameters: Path):
         task = ConfigInformation.fromParameters(params["objects"])
         task.__taskdir__ = Path.cwd()
 
-        # Set the tags
-        task.__tags__ = params["tags"]
-
         # Notify that the task has started
         progress(0)
 
@@ -92,9 +90,20 @@ class TaskRunner:
                 report_eoj()
             logger.info("Finished cleanup")
 
-    def handle_error(self, code, frame_type):
-        logger.info("Error handler: finished with code %d", code)
-        self.failedpath.write_text(str(code))
+    def handle_error(self, code, frame_type, reason: str = "failed", message: str = ""):
+        """Handle task error and write failure information.
+
+        Args:
+            code: Exit code
+            frame_type: Signal frame type (unused)
+            reason: Failure reason (e.g., "failed", "timeout")
+            message: Optional message with details
+        """
+        logger.info("Error handler: finished with code %d, reason=%s", code, reason)
+        failure_info = {"code": code, "reason": reason}
+        if message:
+            failure_info["message"] = message
+        self.failedpath.write_text(json.dumps(failure_info))
         self.cleanup()
         logger.info("Exiting")
         delayed_shutdown(60, exit_code=code)
@@ -147,6 +156,10 @@ class TaskRunner:
                 # Everything went OK
                 logger.info("Task ended successfully")
                 sys.exit(0)
+        except GracefulTimeout as e:
+            logger.info("Task requested graceful timeout: %s", e.message)
+            self.handle_error(1, None, reason="timeout", message=e.message)
+
         except Exception:
             logger.exception("Got exception while running")
             self.handle_error(1, None)