Craft in K8S

Enterprise adoption

Gartner Research 2022

Projects for Organisations will increasingly be deployed in-house K8s clusters.

Access to K8s is getting easier

🤔

Best Practices

K8s (tries to) force the use of best practices in application design

From 12 Factor Apps:

• Environment stored config
• External backing services
• Build, release, run
• Stateless processes
• Port binding
• Disposability
• Log event streams

Back in 2019, Docker started to become widely adopted for packaging and running applications.

Everything started getting containerised.

Enterprises were no exception.

Large companies had thousands of apps packaged as containers.

K8s acts as an abstraction over server resources.

Scheduling - Networking - Storage - Resources
Scaling (servers and containers) - Deployments - Server Upgrades (kinda)

K8s organises containers.

Our first step is to make sure we use them.

Then export them as 'images'.

Containers have their own filesystems.

Containers restart often.

Each time, their filesystem resets to match the state.

Anything we write to the filesystem at run-time will be lost, unless we prepare for it.

Outside of Assets, Craft is kind to us.

It has been built to work well in this type of environment.

Just set `allowUpdates => false`.

Plugins... not so much.

The one assumption Craft/Yii does make...

That both PHP and nginx are using the same filesystem and can share cpresources files.

?

- admin-cp.js

?

- admin-cp.js

Remember ephemeral filesystems? Things on the filesystem don't last long!

Logs should go to stdout or stderr.

K8s will collect and organise them.

The k8s ops team might provide some way for you to export or access them.

CRAFT_STREAM_LOG=true

Craft's default cache and session handlers use the filesystem.

Uh oh, ephemeral filesystems + load balancing...

Super easy to deploy to k8s.

Super easy to configure in Craft.

But watch out for a yii2-redis bug that we ran into:

https://servd.host/blog/craft-yii-and-redis-session-absentee

When deploying to k8s we need to define our expected CPU and memory usages, so k8s can schedule our containers.

Processing images using ImageMagick or GD requires a lot of memory!

Save yourself a headache. Use an external service: Imgix, ServerlessSharp

We have two options:

1. Set resources to match what we expect PHP to use.
   K8s will kill the container when it processes an image and uses much more than expected.
2. Set resources to match the max required to resize an image.
   K8s will reserve way more resources than the application is likely to require. Especially when scaling horizontally. Ops team unhappy :(

Running Craft on k8s takes a bit of planning.

But it's do-able.

Now you're ready to deploy, scale, and play nicely with all the benefits k8s offers.