Back in May, I wrote a blog post about how I had moved a number of Dancer2 applications to a new server and had, in the process, created a standardised procedure for deploying Dancer2 apps. It’s been about six weeks since I did that and I thought it would be useful to give a little update on how it all went and talk about a few little changes I’ve made.

I mentioned that I was moving the apps to a new server. What I didn’t say was that I was convinced my old server was overpowered (and overpriced!) for what I needed, so the new server has less RAM and, I think, a slower CPU than the old one. And that turned out to be a bit of a problem. It turned out there was a time early each morning when there were too many requests coming into the server and it ran out of memory. I was waking up most days to a dead server. My previous work meant that fixing it wasn’t hard, but it really wasn’t something that I wanted to do most mornings.

So I wanted to look into reducing the amount of memory used by the apps. And that turned out to be a two-stage approach.

You might recall that the apps were all controlled using a standardised driver program called “app_service”. It looked like this:

We’re deferring most of the clever stuff to Daemon::Control. But we’re building the parameters to pass to the constructor. And two of the parameters (“program” and “program_args”) control how the service is run. You’ll see I’m using Starman. The first fix was obvious when you look at my code. Starman is a pre-forking server and we always start with 10 copies of the app. Now, I’m very proud of some of my apps, but I think it’s optimistic to think my Klortho server will ever need to respond to 10 simultaneous requests. Honestly, I’m pleasantly surprised if it gets 10 requests in a month. So the first change was to make it easy to change the number of workers.

In the previous article, I talked about using ENV::Util to load environment variables from a “.env” file. And we can continue to use that approach here. I rewrote the “program_args” code to be this:

Here we look for an environment variable (defined in “.env”) and use either that value or a default of 10.

I made similar changes to all the “app_service” files, added appropriate environment variables to all the “.env” files and restarted all the apps. Immediately, I could see an improvement as I was now running maybe a third of the app processes on the server. But I felt I could do betters. So I had a close look at the Starman documentation to see if there was anything else I could tweak. That’s when I found the “–preload-app” command-line option.

Starman works by loading a main driver process which then fires up as many worker processes as you ask it for. Without the “–preload-app” option, each of those processes loads a copy of the application. But with this option, each worker process reads the main driver’s copy of the application and only loads its own copy when it wants to write something. This can be a big memory saving – although it’s important to note that the documentation warns:

Enabling this option can cause bad things happen when resources like sockets or database connections are opened at load time by the master process and shared by multiple children.

I’m pretty sure that most of my apps are not in any danger here, but I’m keeping a close eye on the situation and if I see any problems, it’s easy enough to turn preloading off again.

When adding the preloading option to “app_service”, I realised I should probably completely rewrite the code that builds the program arguments. It now looks like this:

The observant among you will notice that I’ve subtly changed the behaviour of the worker count environment variable. Previously, a missing variable would use a default value of 10. Now, it just omits the argument which uses Starman’s default value of 5.

I’ve made similar changes in all my “app_service” programs and set environment variables to turn preloading on. And now my apps use substantially less memory. The server hasn’t died since I implemented this stuff at the start of this week. So that makes me very happy.

But programming is the pursuit of minimisation. I’ve already seen two places where I can make these programs smaller and simpler.

1. That last code snippet looks too repetitive. It should be a loop iterating over a hash. The keys are the names of the environment variables and the values are references to arrays containing the values that are added to the program arguments if that environment variable is set.
2. I now have five or six “app_service” programs that look very similar. I must be able to turn them into one standard program. Do those environment variables really need to include the application name?

The Klortho service driver program is on GitHub. Can you suggest any more improvements?