| Summary: | KeepAliveTimeout is ignored in all but the first VirtualHost | ||
|---|---|---|---|
| Product: | Apache httpd-2 | Reporter: | J <jaeschke> |
| Component: | Documentation | Assignee: | HTTP Server Documentation List <docs> |
| Status: | RESOLVED FIXED | ||
| Severity: | minor | CC: | jeremyk, takashi.asfbugzilla |
| Priority: | P4 | Keywords: | PatchAvailable |
| Version: | 2.5-HEAD | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Attachments: | add a note | ||
|
Description
J
2005-09-22 13:52:48 UTC
I'm currently experiencing this bug as well. I do have a little more information about it, however. It appears that the limit is not per server overall, it's per IP address. Say we have something like this: <VirtualHost 127.0.1.2:80> ServerName virt1.example.com KeepAliveTimeout 20 </VirtualHost> <VirtualHost 127.1.2.3:80> ServerName virt2.example.com KeepAliveTimeout 14 </VirtualHost> <VirtualHost 127.0.1.2:80> ServerName virt3.example.com KeepAliveTimeout 3 </VirtualHost> <VirtualHost 127.1.2.3:80> ServerName virt4.example.com KeepAliveTimeout 12 </VirtualHost> Then the keepalivetimeouts per domain will be the following: virt1.example.com: 20 virt2.example.com: 14 virt3.example.com: 20 virt4.example.com: 14 Additionally, if we remove the KeepAliveTimeout setting from the virt1.example.com virtualdomains, then both virt1.example.com and virt3.example.com will revert to the default of 15 (or the global setting specified outside of a virtualhost) Coincidentally, you can still toggle keepalives on and off per domain, regardless of load order. If we change our config to look something like this: <VirtualHost 127.0.1.2:80> ServerName virt1.example.com KeepAliveTimeout 20 </VirtualHost> <VirtualHost 127.1.2.3:80> ServerName virt2.example.com KeepAliveTimeout 14 KeepAlive Off </VirtualHost> <VirtualHost 127.0.1.2:80> ServerName virt3.example.com KeepAliveTimeout 3 </VirtualHost> <VirtualHost 127.1.2.3:80> ServerName virt4.example.com KeepAliveTimeout 12 </VirtualHost> then virt2.example.com will not allow keepalives, and virt4.example.com will have a 14 second keepalive timeout. I can provide a full config file if need be. I'm using apache 2.0.54 with the following modules loaded (not in any particular order): mod_autoindex mod_auth mod_auth_digest mod_auth_mysql libphp4 mod_access mod_actions mod_alias mod_cgi mod_deflate mod_dir mod_env mod_expires mod_headers mod_include mod_log_config mod_logio mod_mime mod_negotiation mod_rewrite mod_ssl mod_suexec mod_fastcgi mod_security mod_setenvif mod_encoding mod_dav mod_dav_fs mod_cband mod_dav_svn mod_authz_svn mod_limitipconn Some of those modules might be in-house, although I don't think so. If one isn't recognized, let me know and I will find out where it came from. If you need any additional information, I can provide anything required, just say the words :) This works as designed. KeepaliveTimeout is a connection property (you can sent requests to different name based virtual hosts over the same TCP/IP connection) thus is cannot be set differently in name based virtual hosts. As you have seen it is perfectly possible to adjust it individually for IP/Port based virtual hosts. In the case of name based virtual hosts it is the usual behaviour to use the setting of the first name based virtual host (the default host) for such properties. I admit that this is not clearly expressed in the documentation. So I leave this report open as a documentation bug. it's strange though, that you can *disable* keepalives on a per virtualhost basis, but not change the timeout? (In reply to comment #3) > it's strange though, that you can *disable* keepalives on a per virtualhost > basis, but not change the timeout? This is a partial valid point, but If a request goes to a name based host that has set keepalive off the connection can be shutdown by the server and this seems to be the right thing to do in this situation. On the other hand if a request goes to a name based virtual host and the next request on this connection goes to a different name based virtual host. What is the correct keepalive timeout to use? You could find arguments for both options: 1. It should be the one of the host that just processed the request. 2. It should be the one of the host that processes the next request. (In reply to comment #4) > This is a partial valid point, but > > If a request goes to a name based host that has set keepalive off the connection > can be shutdown by the server and this seems to be the right thing to do in this > situation. Agreed. > On the other hand if a request goes to a name based virtual host and the next > request on this connection goes to a different name based virtual host. What is > the correct keepalive timeout to use? > You could find arguments for both options: > 1. It should be the one of the host that just processed the request. > 2. It should be the one of the host that processes the next request. we can't possibly know what host is going to process the next request though, since we haven't even gotten the request. I would say it makes the most sense to use the keepalivetimeout setting for the virtualhost that processed the most recent request. Not to mention that such a scenario is probably pretty highly unlikely. Created attachment 21432 [details]
add a note
|