CF Server just spikes, how do you see what it's running!

Report · May 22, 2009

we have 7 webservers in a cluster. The code is all the same via svn. Now we have a few websites, a few simple monitors that report the server load (these are all linux servers) and they all hover around .5 -> 2. Lunch time, early evening for a span of 30-90 minutes, one will start, go from a 1.5 to 13, then 24, then 45 killing the box. The loadbalancer sees it and takes it out of the loop so it won't continue to get new traffic but the load continues to rise as 'it' processes whatever it is running.

Is there a way to say in plain english, "Cold Fusion, what pages, queries, etc. are you running right now"? I tried using the CF Monitor, the alerts which send you a huge dump on info none of which really helps. I have thought about fusionreactor, which we looked at a long time ago, but had problems installing and didn't have time dedicated to really looking at it.

So is there an 'apachetop' type thing command line, or something to help debug. We get thousands of logs every 10 minutes so using a log parser like splunk seems almost impossible.

Thanks.

Report · May 22, 2009

I've had some success using mod_status (it allows the current

activitiy to be visible as site.com/server-status) and some simple

sort | uniq -c processing.

Mack

Report · May 22, 2009

Cool, now will that show what coldfusion is actually doing (page/query) or just that it's using the most resources? That I know, when the load hits, memmory goes out the window, deep into swap, and top, ps, etc. just show coldfusion taking up all the RAM. You can't actually tell what CF is actually doing though.

Will look into it.

Thanks.

Report · May 23, 2009

It shows the current activity of the apache server. To see what's

happening I would show only the .cfm pages to see if there's a

particular page that gets hit, then order by the number of

requests/IP, etc.

Mack

Report · May 29, 2009

You should look at FusionReactor again: http://www.fusion-reactor.com/. We run 10 instances across 2 servers and it's been invaluable in identifying bottleneck issues. I first tried it because we were having memory spike issues and just couldn't track it down. Once I installed it, I was able to identify the issue and even the specific piece of code the first week. Seems it's at least worth the effort in doing the free trial if you really want to track down your problem. Try it, you'll like it:) I personally like the Enterprise Dashboard and the current Running Requests and the ability to drill down into each requests details.

Oh, and it took me about 15 minutes to install it.