I recently migrated a small web app to OLS. There are a couple of html files and two primary Perl/FastCGI scripts as external apps using UDS, directly called by the command parameter. Everything works, is operating in FastCGI mode, and is generally lightning fast. However, I have a situation I haven't been able to figure out. A common action in the app fires off four async xhr requests to one external app (this is the one with the problem), and one xhr to the other external app. One can do this many times, over and over, and it all works as expected. I've put logging into the external apps so I can see the time actual request processing starts and finishes. All four requests arrive at the same time, the processing start times match up with the times of request arrival in the access log, and the responses occurs milliseconds later. But every now and then, the fourth request doesn't start processing until ~7 seconds after the access log time-- I am assuming it wasn't 'assigned' until then. Processing takes the normal amount of time after that. It's always the fourth request of the group, and is usually the same 'call', but not always.
I've fiddled with the OLS server and vhost (template) parameters to no avail. The virtual server it runs on is 4cpu/8G running Ubuntu 24 wth Mariadb as a backend. Both OLS admin and top show the server very lightly loaded: 0.1 or less usually with no observable spikes. I did load up the server to 9+ once and it still worked. Though I've allowed many more, the pool for the 4-hit app is rarely over 11-13, and one rarely sees more than 1 or 2 in use in the dashboard. Response data is not enormous.
It's been a while since I've run a web server, and OLS is new to me, so I expect I've messed something up, but I cannot figure out what. Anyone had a similar experience, or have any ideas?
I've fiddled with the OLS server and vhost (template) parameters to no avail. The virtual server it runs on is 4cpu/8G running Ubuntu 24 wth Mariadb as a backend. Both OLS admin and top show the server very lightly loaded: 0.1 or less usually with no observable spikes. I did load up the server to 9+ once and it still worked. Though I've allowed many more, the pool for the 4-hit app is rarely over 11-13, and one rarely sees more than 1 or 2 in use in the dashboard. Response data is not enormous.
It's been a while since I've run a web server, and OLS is new to me, so I expect I've messed something up, but I cannot figure out what. Anyone had a similar experience, or have any ideas?
