Last modified: 2014-09-26 07:12:48 UTC
Shahyar Ghobadpour in his local MediaWiki-vagrant, and me in the http://flow-tests.wmflabs.org instance are both experiencing http timeouts. The index.php content loads, the first one or two load.php succeeds, but then subsequent load.php requests for that page spin for a lonnnnnnnng time, and then regular index.php and api.php requests time out as well. We both pulled latest vagrant repository and ran `{labs-,}vagrant provision`. `sudo apachectl restart` resets the web server for one request, then it starts timing out again. I just did `sudo restart hhvm` `sudo /etc/init.d/redis-server restart` and then another `sudo apachectl stop` `sudo apachectl restart` and the flow-tests labs server has stayed up for several requests. FWIW flow-tests:/var/log/apache2/error.log has lines like [Mon Aug 11 19:09:57.267877 2014] [proxy_fcgi:error] [pid 25259] (22)Invalid argument: [client 127.0.0.1:57582] AH01075: Error dispatching request to : [Mon Aug 11 19:11:57.393869 2014] [proxy_fcgi:error] [pid 24268] [client 10.68.16.65:44516] AH01067: Failed to read FastCGI header, referer: http://flow-tests.wmflabs.org/wiki/Talk:Sandbox [Mon Aug 11 19:11:57.393931 2014] [proxy_fcgi:error] [pid 24268] (70014)End of file found: [client 10.68.16.65:44516] AH01075: Error dispatching request to :, referer: http://flow-tests.wmflabs.org/wiki/Talk:Sandbox these might be a false alarm from all the restarting I was doing.
I think I might be experiencing the same thing. It seems to be happening more often (after fewer requests) on articles imported from other wikis (possibly containing more elements than just text or a single image, such as templates).
This sounds a lot like bug 68574, which I thought was fixed in our latest hhvm builds. What version of hhvm do you have installed? The best way to tell at the moment is to do `dpkg -l|grep hhvm` so you can get the full build number. The latest package is 3.3-dev+20140728+wmf5
(In reply to Bryan Davis from comment #2) > What version of hhvm do you have installed? The best way to > tell at the moment is to do `dpkg -l|grep hhvm` so you can get the full > build number. This morning the flow-tests wmflabs server was at 3.1+20140723-1+wmf1 It has been more stable only running a few browser tests on it. > The latest package is 3.3-dev+20140728+wmf5 I `labs-vagrant enable <random feature>` and did a labs-vagrant provision; I had to follow the suggestion "You might want to run 'apt-get -f install' to correct these <errors>". Now flow-tests is at 3.3-dev+20140728+wmf5 I restarted hhvm and apache, we'll see :)
S: please re-open if this is still an issue.