nateberkopec

nateberkopec, 26 days ago to random

Almost everyone uses an exception tracker (e.g. Sentry) to track errors in production. These become bug tickets, sometimes even automatically. Bugs get fixed.

Do you have something like this for performance? If you don't, are you wondering why your engineers never fix perf?

tenderlove, 7 months ago to random

First time driving on the left! 😵😵😵

nateberkopec, 8 months ago to random

Most of the time, you don't want "performance" testing. These setups are brittle and bad at predicting production usage.

Far better to feature flag, roll out to small groups (1%, then 5%, etc) and see what happens.

nateberkopec, 9 months ago to random

Aaron Patterson and John Hawthorn have created a new profiler for Ruby:

1. It can profile multiple threads
2. It has "markers" for when events happen, independent of the callstack/timeline

https://github.com/jhawthorn/vernier

nateberkopec, 9 months ago to random

Redis is not faster than your SQL database, it's just different.

It's usually under less load. It's key-value, not relational, so looking up simple things can feel faster (but would sometimes also be fast if you just used a hash index on your SQL database).

It's not magical performance dust, just different.

nateberkopec, 9 months ago to random

@tenderlove “sampling profiling is vibes based profiling” legendary cool stuff livestream quote

nateberkopec, 9 months ago to random

I find that Rails apps that use serializers have more problems than apps that use templates (jbuilder, or just no JSON, HTML templates only).

Serializers cause a little too much abstraction between dev and data, leading to careless mistakes about data access and N+1s galore

nateberkopec, 9 months ago to random

The first step to a scalable web service is automatic autoscaling.

Without autoscaling, you're just waiting for a flash sale, social media post, or other "white swan" event to take down your site at just the wrong moment.

nateberkopec, 11 months ago to random

I would be skeptical about advice on how to run a company from someone that caused a third of their company to voluntarily quit en-masse.

nateberkopec, 11 months ago to random

100% not interested in a new Zuck social network, thanks.

nateberkopec, 1 year ago to random

Everyone should be using Spot instances for all web and background job workloads.

Your requests should be timing out after 60 seconds tops, and your background jobs should be resumable (or at least idempotent). Spot is not a problem. Take the 70% savings off your AWS bill.

nateberkopec, 1 year ago to random

ActiveRecord::QueryCache is no longer of unbounded size!

https://github.com/rails/rails/pull/48110

I predict a lot of "my memory usage went way down!" posts after the next Rails release...

nateberkopec, 1 year ago to random

What is the logical equivalent of !(predicate? && predicate2?) without parentheses?

A: https://simple.wikipedia.org/wiki/De_Morgan's_laws?wprov=sfti1

#railsconf