VPS crashes with the dreaded oom message....

[Floren]

Nobody asked the most important question... How is your MPM setup? Are you using mod_php?
All those fancy graphics are not helping, edit the sar config to log everything every minute and do a ps dump of TOP10 processes every second if you really want to know what broke your server when it happens.

I would solve the issue easy by installing Nginx, is available in Debian. It is way more flexible, with a very small memory footprint.

 

[p4guru]

Indeed even setting up Nginx as a reverse proxy to Apache for static files would help lower total memory usage too

 

[TPerry]

Or install OpenLiteSpeed. I've got it running on several of my VPS's now and it seems to be very comparable from my brief reviews of nginx.

 

[craigiri]

OK, I may have caught one hint. Look at what happened at 9am this morning.


There is no reason I can think of for the server load to peak like that.
So, my questions.....

1. Since "system" is the bad guy here, does that relate in any way which would help me locate it? In other words, do cron jobs use CPU or system, etc.? Can I narrow it.

2. The XF board is very small - less than 500 members and only a few posts a day at most. Would I be able to eliminate these XF cron jobs as the cause?

Any suggestions on what to look at would be appreciated. RAM and Network use also shot up at the same time, although only by a factor of perhaps 4X as opposed to this increase of maybe 10-20X (usual system is 4% or so).

Thanks!

Update - no system messages and no increase in web pages loaded.
Update 2: Network increased, but only about 4X.....see enclosed.

 

[craigiri]

Indeed even setting up Nginx as a reverse proxy to Apache for static files would help lower total memory usage too

As per above, there is no problem 99.9% of the time. It appears to be just a single blip which occurs at almost random times every week or two. This blip did not shut the processes down.....but some of them probably do.

Based on that, changing the server type and tuning may not do the job....rather finding the source of that 10-20X momentary increase in "system" which occurs randomly might?

 

[Floren]

The graphics don't help at all.
Post a ps -def output, you will see exactly what generates the load, or at least a top -M.