A bunch of changes...

[python_utils.git] / lockfile.py

diff --git a/lockfile.py b/lockfile.py
index b6a832ee0e3a9c5eadb27cc3b1955538fb49c345..ca190df09f9245bca7153945f0fee61859daa1d7 100644 (file)
--- a/lockfile.py
+++ b/lockfile.py
@@ -9,9 +9,20 @@ import signal
 import sys
 from typing import Optional
 
+import config
 import decorator_utils
 
 
+cfg = config.add_commandline_args(
+    f'Lockfile ({__file__})',
+    'Args related to lockfiles')
+cfg.add_argument(
+    '--lockfile_held_duration_warning_threshold_sec',
+    type=float,
+    default=10.0,
+    metavar='SECONDS',
+    help='If a lock is held for longer than this threshold we log a warning'
+)
 logger = logging.getLogger(__name__)
 
 
@@ -75,7 +86,7 @@ class LockFile(object):
             return True
         except OSError:
             pass
-        logger.debug(f'Failed; I could not acquire {self.lockfile}.')
+        logger.warning(f'Could not acquire {self.lockfile}.')
         return False
 
     def acquire_with_retries(
@@ -108,12 +119,19 @@ class LockFile(object):
 
     def __enter__(self):
         if self.acquire_with_retries():
+            self.locktime = datetime.datetime.now().timestamp()
             return self
         msg = f"Couldn't acquire {self.lockfile}; giving up."
         logger.warning(msg)
         raise LockFileException(msg)
 
     def __exit__(self, type, value, traceback):
+        if self.locktime:
+            ts = datetime.datetime.now().timestamp()
+            duration = ts - self.locktime
+            if duration >= config.config['lockfile_held_duration_warning_threshold_sec']:
+                str_duration = datetime_utils.describe_duration_briefly(duration)
+                logger.warning(f'Held {self.lockfile} for {str_duration}')
         self.release()
 
     def __del__(self):
@@ -151,15 +169,16 @@ class LockFile(object):
                     try:
                         os.kill(contents.pid, 0)
                     except OSError:
-                        logger.debug('The pid seems stale; killing the lock.')
+                        logger.warning(f'Lockfile {self.lockfile}\'s pid ({contents.pid}) is stale; ' +
+                                       'force acquiring')
                         self.release()
 
                     # Has the lock expiration expired?
                     if contents.expiration_timestamp is not None:
                         now = datetime.datetime.now().timestamp()
                         if now > contents.expiration_datetime:
-                            logger.debug('The expiration time has passed; ' +
-                                         'killing the lock')
+                            logger.warning(f'Lockfile {self.lockfile} expiration time has passed; ' +
+                                           'force acquiring')
                             self.release()
         except Exception:
             pass