VPS servers can easily run out of memory during traffic spikes.
For example, most people don’t change Apache’s default setting which allows 150 clients to connect simultaneously. This is way too large a number for a typical VPS server. Let’s do the math. Apache’s processes are typically ~25MB each. If our website gets a temporary traffic spike and 150 processes launch, we’ll need 3750MB of memory on our server. If we don’t have this much (and we don’t!), then the OS will grind to a halt as it swaps memory to disk to make room for new processes, but then immediately swaps the stuff on disk back into memory.
No useful work gets done once swapping happens. The server can be stuck in this state for hours, even after the traffic rush has subsided. During this time, very few web requests will get serviced.
It’s very important to configure your applications so memory swapping does not occur. If you use Apache, you should set MaxClients to something more reasonable like 20 or 30. There are many other optimizations to make, too. Linode has a Library article with optimizations for Apache, MySQL, and PHP.
Reboot server on out of memory condition
Still, in cases where something goes awry, it is good to automatically reboot your server when it runs out of memory. This will cause a minute or two of downtime, but it’s better than languishing in the swapping state for potentially hours or days.
You can leverage a couple kernel settings and Lassie to make this happen on Linode.
Adding the following two lines to your /etc/sysctl.conf will cause it to reboot after running out of memory:
The vm.panic_on_oom=1 line enables panic on OOM; the kernel.panic=10 line tells the kernel to reboot ten seconds after panicking.
Read more about rebooting when out of memory on Linode’s wiki.