Fix race condition in logging (#150)

* Fix race condition in logging

* bump version v25.37.0 -> v26.8.0

* Bump black version

* New tag

* bump version v26.8.0 -> v26.8.1

Author: adammcdonagh

.pre-commit-config.yaml

@@ -32,7 +32,7 @@ repos:
     hooks:
       - id: prettier
   - repo: https://github.com/psf/black
-    rev: 25.1.0
+    rev: 26.1.0
     hooks:
       - id: black
         args:

CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+# v25.8.1
+
+- Handle race condition when creating directories in logging module, triggered byu many threads trying to create the same directory at the same time.
+- Enforce log output bufering to flush after newline characters.
+
 # v25.37.0
 
 - Add additional logging to `sftp` & `local` protocol.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "opentaskpy"
-version = "v25.37.0"
+version = "v26.8.1"
 authors = [{ name = "Adam McDonagh", email = "adam@elitemonkey.net" }]
 license-files = [ "LICENSE" ]
 
@@ -42,7 +42,7 @@ requires-python = ">=3.11"
 dev = [
     "types-requests >=2.28",
     "types-paramiko >=3.0",
-    "black == 25.1.0",
+    "black == 26.1.0",
     "isort",
     "pytest",
     "bumpver",
@@ -71,7 +71,7 @@ otf-batch-validator = "opentaskpy.cli.batch_validator:main"
 profile = 'black'
 
 [tool.bumpver]
-current_version = "v25.37.0"
+current_version = "v26.8.1"
 version_pattern = "vYY.WW.PATCH[-TAG]"
 commit_message = "bump version {old_version} -> {new_version}"
 commit = true

src/opentaskpy/cli/task_run.py

@@ -18,8 +18,7 @@ def main() -> None:
     """Parse args and call TaskRun class."""
     parser = argparse.ArgumentParser(
         formatter_class=argparse.RawTextHelpFormatter,
-        epilog=dedent(
-            """\
+        epilog=dedent("""\
                                 Environment Variables:
 
                                 There are several environment variables that can be used to impact the behaviour:
@@ -44,8 +43,7 @@ def main() -> None:
 
                                 e.g. OTF_OVERRIDE_TRANSFER_DESTINATION_0_PROTOCOL_CREDENTIALS_USERNAME
 
-                                """
-        ),
+                                """),
     )
 
     parser.add_argument(

src/opentaskpy/otflogging.py

@@ -4,6 +4,7 @@
 import logging
 import os
 import re
+import threading
 from datetime import datetime
 
 OTF_LOG_FORMAT = (
@@ -393,7 +394,11 @@ def close_log_file(logger__: logging.Logger, result: bool = False) -> None:
         # Now everything is closed, we can rename the log file
         # If result is True, then rename the file and remove _running from the name
         if new_log_filename:
-            os.rename(log_file_name, new_log_filename)
+            # Use lock to prevent race condition when multiple threads rename same file
+            with _file_operations_lock:
+                # Check file still exists before renaming (another thread may have renamed it)
+                if os.path.exists(log_file_name):
+                    os.rename(log_file_name, new_log_filename)
 
             # Change the basename of the handler to match the new filename, in case it
             # wants to log anything else
@@ -443,6 +448,9 @@ def redact(log_message: str) -> str:
 
 logger = init_logging(__name__)
 
+# Thread lock for directory creation and file operations
+_file_operations_lock = threading.Lock()
+
 
 # mypy: ignore-errors
 class TaskFileHandler(logging.FileHandler):
@@ -461,3 +469,22 @@ def __init__(self, filename, mode="a", encoding=None, delay=True):
         """
         _mkdir(os.path.dirname(filename))
         logging.FileHandler.__init__(self, filename, mode, encoding, delay)
+
+    def _open(self):
+        """Open the log file, ensuring the directory exists.
+
+        Overrides the FileHandler._open to ensure directory creation is thread-safe
+        even when delay=True is used. Also uses line buffering for real-time log updates,
+        which is important when using network filesystems like EFS.
+        """
+        # Ensure directory exists with thread safety before opening file
+        with _file_operations_lock:
+            _mkdir(os.path.dirname(self.baseFilename))
+        # Open with line buffering (buffering=1) for real-time log visibility on network filesystems
+        return open(
+            self.baseFilename,
+            self.mode,
+            buffering=1,
+            encoding=self.encoding,
+            errors=self.errors,
+        )