It's so difficult to scale for this type of load...

What is this graph what does it measure?

Is it routine? If you know it's coming at the same time you can scale 30 minutes before.

Isn’t this what serverless is for?

look like a random spike... is it some kind of shop or social media app?

is it unforeseen legitimate traffic? Perhaps you don't even need to scale for it -- just filter it out? And if it's legit and foreseen - scale for it in advance.

The load is expected. This specifically is is operations on our Redis database that we use for caching. Our platform is for conferences, so the majority of the time we have no users, until suddently we have thousands of users simultaneously online for it.

We can sort of scale for it in advance, unless too much is going on at the same time (we're two developers managing the environment + managing other developers + prorgamming).

Spiky load is always fun.

Application wise we use Kubernetes and tend to run with a fairly large overhead, so auto scaling can catch up.

The E in EC2 stands for Elastic; EC2 instances accumulate CPU credit which can then be used to handle spikes.

@devs this is a fun project to plan and build for.

If you have enough data you could actually train an AI to look for early indicators that would determine these spikes and scale out to accommodate.

With a combination of data sources like the traffic you're showing along with app logs, database content based on venue dates it's possible to accomplish this.

This is no different than how Netflix handles their own scalability problems.

@sariel we usually go from 0 users to >5000 users within 2-3 minutes.

@devs then there's plenty of time to scale up. 😂

@jonas-w my dicki

@karma your dicki in the morning

Hard wood ™️

Gamestop?