On Fri, Nov 6, 2015 at 11:59 AM, Cindy Cicalese cindom@gmail.com wrote:

We are noticing a performance issue that we have traced back to our squid configuration, but we are unsure how to fix it. The symptom is that every time we save a wiki page, it takes about 7 seconds. Otherwise traversing the wiki and rendering pages is not affected. We are running the following software versions:

MediaWiki 1.25.2 PHP 5.5.30 MySQL 5.5.46 Squid 3.1.23

We followed the instructions at https://www.mediawiki.org/wiki/Manual:Squid_caching#Configuring_MediaWiki. In particular, we are using the following in LocalSettings.php:

$wgUseSquid = true; $wgSquidServers = array('<your IPv4 address>'); $wgSquidServersNoPurge = array('127.0.0.1');

with our IP address inserted. When we comment out the $wgSquidServers line, saving reverts to desired speedy performance, so we have concluded that the problem involves purging the modified page. Supporting this conclusion, we see the following in our debug log:

[squid] SquidPurgeClientPool: timeout (5s)

The wiki in question is only accessible by https. We do not see the same problem on wikis that respond to http. Is there a configuration setting that we need to accommodate this situation, or is that unrelated?

Thanks for any help in sorting this out.

By default, SquidPurgeClient connects to the Squid server on port 80. It sounds like SquidPurgeClient::getSocket() is hanging for you possibly because of an intervening firewall that is dropping port 80 traffic.If you have shell access on the MW servers that are timing out sending these purge requests, you can attempt to verify using telnet or similar socket connections.

You can use a ":<port number>" suffix on each element of $wgSquidServers to specify an alternate port (eg '10.11.12.13:443').

Bryan