Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
From:   Omar Kilani <omar.kilani@gmail.com>
Date:   Mon, 15 Apr 2019 11:02:42 -0700
Message-ID: <CA+8F9hicnF=kvjXPZFQy=Pa2HJUS3JS+G9VswFHNQQynPMHGVQ@mail.gmail.com>
Subject: Strange issues with epoll since 5.0
To:     linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi there,

I=E2=80=99m still trying to piece together a reproducible test that trigger=
s
this, but I wanted to post in case someone goes =E2=80=9Chmmm... change X
might have done this=E2=80=9D.

Basically, something=E2=80=99s broken (or at least, has changed enough to
cause problems in user space) in epoll since 5.0. It=E2=80=99s still broken=
 in
5.1-rc5.

It doesn=E2=80=99t happen 100% of the time. It=E2=80=99s sort of hard to pi=
n down but
I=E2=80=99ve observed the following:

* nginx not accepting connections under load
* A java app which uses netty / NIO having strange writability
semantics on channels, which confuses netty / java enough to not
properly flush written data on the socket.

I went and tested these Linux kernels:

4.20.17
4.19.32
4.14.111

And the issue(s) do not show up there.

I=E2=80=99m still actively chasing this up, and will report back =E2=80=94 =
I haven=E2=80=99t
touched kernel code in 15 years so I=E2=80=99m a little rusty. :)

Regards,
Omar