Hidden Productivity Costs of Parallel Run

2021-09-27

All CPUs are Fully Used

PHPStan parallel implementation uses your CPU cores to process multiple files at once. When I run it on my PC, it looks like this:

That's great! 16 threads, all fully used. It means the static analysis process of any project ends up in 15 seconds. That's fast, and then I'll be able to continue.

The problem is, for these 15 seconds, the laptop does not have any spare computation power. It only focuses on tasks given by PHPStan, but everything else is slowed down.

Instant feedback is essential to stay in the flow and focused on the project. Any lag can destroy state of beautiful deep work.

Short but Complete Lag?

Imagine you're writing a post or a code, and then you have to pause for 15 seconds... or even 10 or 5 seconds. Your brain has to stop thinking, start caching the thoughts and code you want to write.

Would you use PHPStorm
if it has 5 seconds lag everytime you create a new class?

What if you want to run PHPStan and one more tool, e.g., PHPUnit or ECS? The latter will suffer from low computation power and take much longer than parallel of PHPStan.

Keep 1 CPU Always Free

There is a workaround for this. We can tell PHPStan to use fewer threads than is our maximum. That way, there will always be one free:

# phpstan.neon
parameters:
    parallel:
        # for Tomas: 16 threads - 1
        maximumNumberOfProcesses: 15

Max Full !== Min Free

Well, there will be one always free on my pc. The problem is, all the developers share the main phpstan.neon config. But what is the chance the developers use computers with the same amount of CPU threads?

It would be much better to have the option to see the number of accessible CPUs. Something like:

# phpstan.neon
parameters:
    parallel:
        minimumNumberOfFreeThreads: 1

P.S.: If you know of such a feature, let me know.

All at Once of One by One?

Imagine you have to compute 5 algebraic formulas. You can choose one from 2 ways to receive them:

• A) you'll get one every 10 minutes
• B) you'll get all formulas at once

Which will cost you more time and which more energy?

Parallel Drains your Battery Much Faster

Before parallel run was added to PHPStan, my laptop worked for only 4-5 hours of coding. I could go for coffee in the morning, see a client, do little mentoring, and it was still running in the afternoon.

As only 1 core was working, there was lot of energy left.

That changed since parallel. When you develop PHPStan rules or work on various projects, PHPStan often needs to analyze the whole project from scratch. That's fast in time but costly in energy.

Suddenly, my laptop battery duration dropped from 5 hours to 2,5 hours. It turned off, charger at home and I was free to meditate.

The battery duration estimates were jumping like crazy from 6 hours to 30 minutes and back. First, I assumed it's some bug in my Ubuntu. Then I found out it depended if PHPStan was running or not.

Charger is Your Friend

This is not a life-ending situation, but it's something you should be aware of. If you're a cable type and stay in one place always plugged in, you don't mind.

But if you love to travel, work, or write posts on the train like me, keep it in mind.

Happy coding!

---

Do you learn from my contents or use open-souce packages like Rector every day?
Consider supporting it on GitHub Sponsors. I'd really appreciate it!