[Bug 192889] accept4 socket hangs in CLOSED (memcached)

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=192889

--- Comment #5 from Arne Claus <arne.claus at trivago.com> ---
We did another test with FreeBSD9, memcached 1.4.20 and libevent 2.x.
This time the server did not get stuck.

We do see different behavior in both servers though.
On FreeBSD 9 we see the CLOSED and CLOSE_WAIT values staying at one number (0
or an initial, low value) for a long time. On FreeBSD 10 they are constantly
increasing and decreasing.
We poll these values in intervals of 1 second so it looks like FreeBSD 10 holds
the connections in these states a lot longer?

What we also see on both systems (in much higher quantities on FreeBSD 10) is
that connections get stuck in CLOSE_WAIT and move to CLOSED after quite a long
time (about two hours).

Can someone explain this behavior or describe what's the differences between
the two systems in that area?

It might help solving the bug or find a faster reproduction rule.
I assume that there is a race condition - involving the libevent code of
memcached - which is just happening more often on FreeBSD 10 due to the
seemingly slower handling of closed connections.

-- 
You are receiving this mail because:
You are the assignee for the bug.