Re: Socket error (EAGAIN) at high load

--===============6979547216118279976==
Content-Type: multipart/alternative; boundary=e89a8ff1c52a3781dd04c585dfe0

--e89a8ff1c52a3781dd04c585dfe0
Content-Type: text/plain; charset=ISO-8859-1

The comment I found talks about Linux 2.0  ( "This works around a problem
on Linux 2.0.x" http://www.fastcgi.com/devkit/libfcgi/os_unix.c  )
Have you asked about it on the linux dev mailing lists?
Since we're at 3.5 now you might think it had been looked at by now.


On Mon, Jul 23, 2012 at 3:55 PM, Dean Banks <[email protected]> wrote:

> Hi all,
>
> I'm running an FCGI 2.4.1-based  C application over a Unix domain socket
> behind nginx on Linux 2.6.32.  At periods of high load, I get a lot of
> EAGAIN errors from nginx, implying that my app is not draining the receive
> buffer fast enough.
>
> I have a high number of FDs configured and socket buffers are
>
> SO_RCVBUF = 256960
> SO_SNDBUF = 256960
>
> It appears that reducing READABLE_UNIX_FD_DROP_DEAD_**TIMEVAL from the
> default of 2.0 seconds may be a reasonable next step to reduce the volume
> of EAGAIN errors under load, but other than a "conservative" value of 2.0
> and a "doesn't work" value of 0.0, I'm having a hard time picking a
> starting point to iterate around.
>
> As an aside, the source mentions that the timeout is required to work
> around a kernel bug,  though without additional information it's not clear
> if the bug has been addressed.  Is the timeout still required?
> ---

*"Why do you have that banana in your ear?"
"To keep away the alligators."
"But there are no alligators here."
"See! It's working!"*

--e89a8ff1c52a3781dd04c585dfe0
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

The comment I found talks about Linux 2.0 =A0( &quot;<span style=3D"white-s=
pace:pre-wrap">This works around a problem on Linux 2.0.x&quot;   </span>
<a href=3D"http://www.fastcgi.com/devkit/libfcgi/os_unix.c">http://www.fast=
cgi.com/devkit/libfcgi/os_unix.c</a>=A0=A0<span style=3D"white-space:pre-wr=
ap">)=A0</span><div><span style=3D"white-space:pre-wrap">Have you asked abo=
ut it on the linux dev mailing lists?</span></div>
<div><span style=3D"white-space:pre-wrap">Since we&#39;re at 3.5 now you mi=
ght think it had been looked at by now.</span></div><div><span style=3D"whi=
te-space:pre-wrap"><br></span><br><div class=3D"gmail_quote">On Mon, Jul 23=
, 2012 at 3:55 PM, Dean Banks <span dir=3D"ltr">&lt;<a href=3D"mailto:conne=
[email protected]" target=3D"_blank">[email protected]</a>&gt;</span> wrot=
e:<br>
<blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1p=
x #ccc solid;padding-left:1ex">Hi all,<br>
<br>
I&#39;m running an FCGI 2.4.1-based =A0C application over a Unix domain soc=
ket behind nginx on Linux 2.6.32. =A0At periods of high load, I get a lot o=
f EAGAIN errors from nginx, implying that my app is not draining the receiv=
e buffer fast enough.<br>

<br>
I have a high number of FDs configured and socket buffers are<br>
<br>
SO_RCVBUF =3D 256960<br>
SO_SNDBUF =3D 256960<br>
<br>
It appears that reducing READABLE_UNIX_FD_DROP_DEAD_<u></u>TIMEVAL from the=
 default of 2.0 seconds may be a reasonable next step to reduce the volume =
of EAGAIN errors under load, but other than a &quot;conservative&quot; valu=
e of 2.0 and a &quot;doesn&#39;t work&quot; value of 0.0, I&#39;m having a =
hard time picking a starting point to iterate around.<br>

<br>
As an aside, the source mentions that the timeout is required to work aroun=
d a kernel bug, =A0though without additional information it&#39;s not clear=
 if the bug has been addressed. =A0Is the timeout still required?<br>---</b=
lockquote>
</div><em style=3D"font-family:Verdana,Helvetica,sans-serif;font-size:12px;=
background-color:rgb(255,255,255)">&quot;Why do you have that banana in you=
r ear?&quot;=A0<br>&quot;To keep away the alligators.&quot;<br>&quot;But th=
ere are no alligators here.&quot;<br>
&quot;See! It&#39;s working!&quot;</em>
<div><br></div>
</div>

--e89a8ff1c52a3781dd04c585dfe0--

--===============6979547216118279976==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

_______________________________________________
FastCGI-developers mailing list
FastCGI-developers-xGejAJT2w6xVgU18Zptdi0EOCMrvLtNR@public.gmane.org
http://mailman.fastcgi.com/mailman/listinfo/fastcgi-developers

--===============6979547216118279976==--