Tobias Schulz-Hess wrote:
Hi Rainer,
Rainer Jung schrieb:
Hi,
1) How many fds does the process have, so is the question "why can't
we use all those 4096 fds configured", or is it "Where do those 4096
fdsused by my process come from"?
The latter. We can actually see the 4096 fds are used (by port 8080 in
CLOSE_WAIT state...).
Well, we're pretty sure that the fds actually are the connections from
the HTTPConnector of Tomcat. The connector is set to use 200 connections
simultaneously. So the question is: Why aren't those connections closed?...
Are you using the tcnative APR connector?
2) CLOSE_WAIT means the remote side closed the connection and the
local side didn't yet close it. What's you remote side with respect to
TCP? Is it browsers, or a load balancer or stuff like that?
We have NGINX as a proxy in front of the tomcat (on another server). So
request from the Internet arrive at NGINX and are then forwarded to the
tomcat(s).
By now, we're pretty happy with NGINX, since it is really fast and has
low footprint, but could well be that it does not work well with tomcat.
We have the problems with our live servers, so the application, which
actually is initiating the connection is a browser.
3) Are you using keep alive (not implying that's the cause of your
problems, but keep alive makes the connection live cycle much more
complicated from the container point of view).
As far as I understood NGINX, we only use keep alive request for the
communication between client and NGINX. The communication between NGINX
and tomcat does not have settings for keep alive, so I assume: no.
This is the relevant part of the NGINX configuration:
location / {
proxy_pass http://verwandt_de;
proxy_redirect off;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For
$proxy_add_x_forwarded_for;
client_max_body_size 10m;
client_body_temp_path
/var/nginx/client_body_temp;
proxy_buffering off;
proxy_store off;
proxy_connect_timeout 30;
proxy_send_timeout 80;
proxy_read_timeout 80;
}
So any suggestions that I should move the topic forward to some NGINX
mailing list?
Not sure yet. It's interesting, that there is a 30 seconds timeout in
this config. Maybe you should investigate, what those 30 seconds.mean.
On the other hand, 30 seconds are not that rarely used as defaults ...
What about experimenting with maxKeepAliveRequests=1 in your http
connector (server.xml)?
Kind regards,
Tobias.
Regards,
Rainer
Tobias Schulz-Hess wrote:
Hi there,
we use the current Tomcat 6.0 on 2 machines. The hardware is brand
new and is really fast. We get lots of traffic which is usually
handled well by the tomcats and the load on those machines is between
1 and 6 (when we have lots of traffic).
The machines have debian 4.1/64 as OS.
However, sometimes (especially if we have lots of traffic) we get the
following exception:
INFO | jvm 1 | 2008/01/23 15:28:18 | java.net.SocketException:
Too many open files
INFO | jvm 1 | 2008/01/23 15:28:18 | at
java.net.PlainSocketImpl.socketAccept(Native Method)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
java.net.PlainSocketImpl.accept(PlainSocketImpl.java:384)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
java.net.ServerSocket.implAccept(ServerSocket.java:453)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
java.net.ServerSocket.accept(ServerSocket.java:421)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
org.apache.tomcat.util.net.DefaultServerSocketFactory.acceptSocket(DefaultServe
rSocketFactory.java:61)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
org.apache.tomcat.util.net.JIoEndpoint$Acceptor.run(JIoEndpoint.java:310)
INFO | jvm 1 | 2008/01/23 15:28:18 | at
java.lang.Thread.run(Thread.java:619)
I
We already have altered the ulimit from 1024 (default) to 4096 (and
therefore proofing: yes, I have used google and read almost
everything about that exception).
We also looked into the open files and all 95% of them are from or to
the Tomcat Port 8080. (The other 5% are open JARs, connections to
memcached and MySQL and SSL-Socket).
Most of the connections to port 8080 are in the CLOSE_WAIT state.
I have the strong feeling that something (tomcat, JVM, whatsoever)
relies that the JVM garbage collection will kill those open
connections. However, if we have heavy load, the garbage collection
is suspended and then the connections pile up. But this is just a guess.
How can this problem be solved?
Thank you and kind regards,
Tobias.
-----------------------------------------------------------
Tobias Schulz-Hess
---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
