-
Notifications
You must be signed in to change notification settings - Fork 872
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
CPU pinned / spinning on epoll_wait #53
Comments
Haven't seen this. What version of libevent? |
As an update, I believe we've run into the same behavior. Note that this elevated CPU usage does not correspond with an increase in throughput. Mean reserves per second is 150, though it does spike to 1000/s for very short periods of time (a few minutes). Here's what out CPU usage looked like before and after this problem cropped up. We're running a 32-bit EC2 instance (a small):
CPU Info uname -a: |
We don't use libevent at all any more, so it's possible the bug is Can anyone reproduce this with recent code from git (since April 19)? |
Keith, would you consider trunk be "production quality" at this point? We still have the IO pinned instance in prod, which I'd love to fix up, but given that I've had a very had time trying to synthetically reproduce this behavior.. don't want to break other stuff while trying building from master. ;-) |
Understood. Yes, I'd consider it production quality, but if you're |
Ok, thanks, we'll give it a shot and let you know what we find out. |
Hi Keith, compiled and deployed the tip of beanstalkd as of June 10, 2011 and I'm seeing the same behavior as described in the bug. It's not completely pegging the cpu, but we're getting the following from an strace. Didn't see the same behavior on 1.4.6. We have a large number of clients doing a 'reserve-with-timeout'. Would that cause timestamp calls in a tight for-loop? epoll_wait(6, {}, 500, 10) = 0 |
This output looks normal. The process wakes up at least once Beanstalkd used to compute the exact amount of time to wait, |
See https://groups.google.com/d/topic/beanstalk-talk/MZNodUgzGHM for Edit: fixed link. |
Just to keep it in the github issue tracker here is my current situation with Here is the loop eating up to 100% CPU (1 core) CALL kevent(0x6,0,0,0x7f7ffffc5f80,0x1,0x829e80) |
This was separately reported as #96, and it's fixed now. |
Haven't been able to identify the exact culprit, but out of the blue one of our beanstalk processes is pinning the CPU (system). Doing an strace shows:
Running 1.4.6 -- anyone else experienced this?
The text was updated successfully, but these errors were encountered: