[Bug 192889] New: accept4 socket hangs in CLOSED (memcached)
bugzilla-noreply at freebsd.org
bugzilla-noreply at freebsd.org
Thu Aug 21 14:40:08 UTC 2014
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=192889
Bug ID: 192889
Summary: accept4 socket hangs in CLOSED (memcached)
Product: Base System
Version: 10.0-STABLE
Hardware: amd64
OS: Any
Status: Needs Triage
Severity: Affects Some People
Priority: ---
Component: kern
Assignee: freebsd-bugs at FreeBSD.org
Reporter: arne.claus at trivago.com
We encountered that our memcached instances on FreeBSD 10 were rejecting
connections after being under heavy load. The machines showed several thousand
connections stuck in CLOSED (sometimes CLOSED_WAIT).
We could reproduce the issue several times using a load script.
There is a corresponding memcached bug with more information on the tests:
https://code.google.com/p/memcached/issues/detail?id=375
We assume that this issue is a FreeBSD 10 related because:
- The memcached source does not seem to do anything wrong (at first glance)
- The issue does not seem to happen when switching from accept4 to accept
(tests pending)
- According to our admins the connections were closed after enabling
net.inet.tcp.log_debug
- According to our admins this bug did not appear on FreeBSD 9 machines
We are still running tests to verify that this bug does only happen on FreeBSD
10 and is accept4 related. The current tests point in this direction.
accept4 uses the flag SOCK_NONBLOCK
Memcached version: 1.4.20
FreeBSD version: 10.0-STABLE #0 r269376
--
You are receiving this mail because:
You are the assignee for the bug.
More information about the freebsd-bugs
mailing list