Union Station is back online after facing scaling issues on Day 1

blog.phusion.nl

19 pointsvijaydev5 comments

This reads like a Ted Dziuba performance art project.

I mean designing your setup as a ~15 server cluster on one box? on a raid-1 of sata drives!?! then xfs on raid-0!!!

This url is getting saved now as my canonical 'overengineering leads to faceplants' example.

I'm not sure what you're getting at. We were getting so much traffic that this setup is necessary, and now that we've scaled I'm very, very glad that we went with this "overengineered" setup since day 1. We also never explicitly chose RAID-1; it was just the default setup that our hosting provider gave us.

But yeah if you're running a blog, don't do this.

RabbitMQ if deployed in the wrong manner, or stuck in the wrong place for the wrong reasons will bite you. I see that it looks like a transaction manager because Mongo can't handle tx very well? I beleive MySQL does, and so and Cassandra. Seems more a fit for that type of job.

We don't use RabbitMQ as a transaction manager. We use it as a worker queue.

RabbitMQ was not the bottleneck. In fact RabbitMQ doesn't perform much I/O at all.

I do see that now thx.

How many rails backends are you running?

Reply on news.ycombinator.com