Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I agree with most of these points. They are a good starting point. I'd probably steer clear of kubernetes until you actually really really need it. ECS is good enough for most things, and a fucktonne more simple to look after.

but, where I hard diverge is the lack of metrics. Everything should be generating metrics. Logs are great, but they are crap for giving you near realtime trends.

Everytime your container gets a connection: increment a counter. Every message processed, counter. every message failed, counter. Every KB of data sent out, counter. Every service call, counter by service.

metric all the things, in a sensible, mostly automated way, with a decent schema.

Then you can combine all those metrics into a dashboard that shows your system performing against business goals.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: