From 7204c0ab09d7f033f61757a0a1ae9c3162e88c59 Mon Sep 17 00:00:00 2001
From: Matthias Diener <mdiener@illinois.edu>
Date: Mon, 13 Dec 2021 14:35:03 -0600
Subject: [PATCH] Cache: reduce sleep time (#504)
* Cache: reduce sleep time
* Update cache.py
* print warnings fix
* restructure
* add comment about our choice for the timeout
---
pyopencl/cache.py | 22 ++++++++++++++++++----
1 file changed, 18 insertions(+), 4 deletions(-)
diff --git a/pyopencl/cache.py b/pyopencl/cache.py
index 6e01c90e..582a7908 100644
--- a/pyopencl/cache.py
+++ b/pyopencl/cache.py
@@ -88,18 +88,32 @@ class CacheLockManager(CleanupBase):
except OSError:
pass
+ # This value was chosen based on the py-filelock package:
+ # https://github.com/tox-dev/py-filelock/blob/a6c8fabc4192fa7a4ae19b1875ee842ec5eb4f61/src/filelock/_api.py#L113
+ # When running pyopencl in an application with multiple ranks
+ # that share a cache_dir, higher timeouts can lead to
+ # application stalls even with low numbers of ranks.
+ # cf. https://github.com/inducer/pyopencl/pull/504
+ wait_time_seconds = 0.05
+
+ # Warn every 10 seconds if not able to acquire lock
+ warn_attempts = int(10/wait_time_seconds)
+
+ # Exit after 60 seconds if not able to acquire lock
+ exit_attempts = int(60/wait_time_seconds)
+
from time import sleep
- sleep(1)
+ sleep(wait_time_seconds)
attempts += 1
- if attempts > 10:
+ if attempts % warn_attempts == 0:
from warnings import warn
warn("could not obtain cache lock--delete '%s' if necessary"
% self.lock_file)
- if attempts > 3 * 60:
- raise RuntimeError("waited more than three minutes "
+ if attempts > exit_attempts:
+ raise RuntimeError("waited more than one minute "
"on the lock file '%s'"
"--something is wrong" % self.lock_file)
--
GitLab