Hacker News Reader

Comment by jimmaswell | Hacker News Reader

From my experience I would never recommend giving up control of your servers to some third party. Weeks wasted waiting for useless support teams to get back to you on something you could have fixed in 10 minutes if you had root. Opaque configuration issues you can't debug without said useless support team. Needing permission and approval for every little thing on prod. If I was ever at a high level in a company I'd never go farther than some AWS load balancing or whatever on cloud instances we still have root on.

View on HN

Yes, the absolute opaqueness of so-called serverless is a huge hit to productivity.

Numerous times there's something weird going on and you're stuck trying to guess and retry based on largely useless logs until it somehow works better but you never really know what the root cause truly was.

Meanwhile on my own server I'll ssh in and have complete visibility, I can trace and dump network traffic, bpftrace userspace and kernel code, attach debuggers, there's no limit to visibility.

Yes lambda/serverless saves you a day or three in initial setup but you'll pay that time back with 10-100x interest as soon as you need to debug anything.

> If I was ever at a high level in a company I'd never go farther than some AWS load balancing or whatever on cloud instances we still have root on.

Your competitors would salivate at this statement, fyi. Speed is a competitive advantage. AWS is not "let's rent a big ball of EC2 servers and call it a day", and anyone who treats it like that is going to get eaten alive. If you have not looked at -- for example -- Dynamo, you should. If you have not looked at SQS, you should. The ability to have predictable, scalable services for your engineers to use and deploy against is like dumping kerosene onto a fire, it unlocks abilities and velocity that more traditional software dev shops just can't compete against.

Our customer (300k+ employees) switched to AWS a couple of years ago and I simply hate it so much. The “predictable, scalable” service we’re using, MSK, is a pain to develop against. Log files are all over the place, thanks to microservices and because no one, myself included, has a clue on how to make them manageable again. AWS’s graphical user interface is a constantly changing mess. I hate clicking my way through a GUI just so I can download individual log files manually.

I wonder how you folks manage to work with AWS and not hate it.

This is just a bad setup.

Ship logs others places, clicking in the GUI outside of testing things out is a mistake, use Terraform or CDK or Cloudformation.

I've managed very large fleets of servers and its dead simple if you have a good setup.

Reply on news.ycombinator.com