Thanks, I tried it, and after modifying the Connection configuration value, the
timeout happens sooner.
I've just found a partial solution: I'm working with PHP in fastcgi mode (php-fpm), and after adding
a header('Connection: close'); to the end of the script, the timeouts are reduced drastically. It's
interesting that in the http response headers coming from httpd the Connection header remains
Keep-alive.
It's also interesting that after reviewing httpd's source code, error 408 is bound to tls somehow.
Though it happens independently from using tls.
Gábor Hidvégi
Hiltjo Posthuma wrote:
On Mon, Jan 15, 2018 at 12:15:48PM +0100, Hidvégi Gábor wrote:
Hello there,
if you look at the access log, you can notice that the shadow request is
exactly one minute after a real query, like:
OpenBSD teszt 192.168.255.1 - - [12/Jan/2018:16:16:42 +0100] "GET / HTTP/1.1"
200 5
OpenBSD teszt 192.168.255.1 - - [12/Jan/2018:16:17:42 +0100] "<UNKNOWN> " 408 0
or
OpenBSD teszt 192.168.255.1 - - [12/Jan/2018:20:22:46 +0100] "GET / HTTP/1.1"
200 5
OpenBSD teszt 192.168.255.1 - - [12/Jan/2018:20:23:46 +0100] "<UNKNOWN> " 408 0
It might be a timeout or a second request started in the background.
Gábor Hidvégi
I think this is because the default timeout in httpd is 1 minute:
"
request timeout seconds
Specify the inactivity timeout for HTTP operations
between client and server, for example the maximum time
to wait for a request from the client. The default
timeout is 60 seconds (1 minute). The maximum is
2147483647 seconds (68 years).
"
Can you try to change this as described below?
Hiltjo Posthuma wrote:
On Fri, Jan 12, 2018 at 05:15:24PM +0100, Hidvégi Gábor wrote:
Synopsis: httpd randomly gives error 408
Category: httpd
Environment:
System : OpenBSD 6.2
Details : OpenBSD 6.2 (GENERIC) #132: Tue Oct 3 21:18:21 MDT 2017
[email protected]:/usr/src/sys/arch/amd64/compile/GENERIC
Architecture: OpenBSD.amd64
Machine : amd64
Description:
httpd sometimes responds errordocument 408 on a random request
How-To-Repeat:
This bug is common, not amd64 only, can be reproduced on armv7 as well. All
you need to do is to open index.html in the browser (tried with Internet
Explorer, Firefox and Chrome) and keep pressing F5 (refresh) many times.
When you look at the access.log, you randomly see lines like this one:
OpenBSD teszt 192.168.0.19 - - [12/Jan/2018:16:59:16 +0100] "<UNKNOWN> " 408 0
httpd.conf:
default type text/html
prefork 2
server "OpenBSD teszt" {
listen on egress port 80
directory no auto index
}
types {
text/html html
}
Fix:
check httpd source
Gábor Hidvégi
Hey,
I can also reproduce it on -current and firefox-esr. It is related to the
timeout handling, but I don't know a fix yet.
A faster way to reproduce it is to add to httpd.conf:
server "*" {
connection {
request timeout 5
}
...
}
Then do a single request (no need to spam it) and wait 5 seconds.
Then (when run with httpd -d -vv) a line appears:
server *, client 1 (1 active), 127.0.0.1:43314 -> 127.0.0.1:6970,
timeout (408 Request Timeout)
