Cache resiliency

msal_extensions/cache_lock.py

@@ -2,9 +2,16 @@
 import os
 import sys
 import errno
+import time
+import logging
+
+logger = logging.getLogger(__name__)
+
 import portalocker
 from distutils.version import LooseVersion
 
+current_time = getattr(time, "monotonic", time.time)
+
 
 class CrossPlatLock(object):
     """Offers a mechanism for waiting until another process is finished interacting with a shared
@@ -25,7 +32,29 @@ def __init__(self, lockfile_path):
             flags=portalocker.LOCK_EX | portalocker.LOCK_NB,
             **open_kwargs)
 
+    def try_to_create_lock_file(self):
+        timeout = 5
+        check_interval = 0.25
+        timeout_end = current_time() + timeout
+        while timeout_end > current_time():
+            try:
+                with open(self._lockpath, 'x'):
+                    return True
+            except OSError as err:
+                if err.errno == errno.EEXIST:
+                    logger.warning("Lock file exists, trying again after some time")
+                    time.sleep(check_interval)
+                else:
+                    raise
+            except ValueError:
+                logger.warning("Python 2 does not support atomic creation of file")
+                return False
+
+        return False
+
     def __enter__(self):
+        if not self.try_to_create_lock_file():
+            logger.warning("Failed to create lock file")
         file_handle = self._lock.__enter__()
         file_handle.write('{} {}'.format(os.getpid(), sys.argv[0]).encode('utf-8'))
         return file_handle

tests/test_cache_lock_file_perf.py

@@ -67,7 +67,7 @@ def test_lock_for_high_workload(temp_location):
 
 
 def test_lock_for_timeout(temp_location):
-    num_of_processes = 10
+    num_of_processes = 30
     sleep_interval = 1
     _run_multiple_processes(num_of_processes, temp_location, sleep_interval)
     count = _validate_result_in_cache(temp_location)