Post mortem of a failed HackerNews launch

Post mortem of a failed HackerNews launch

Faced the same thing with Solr once, I was running a Chef Server on an AWS Micro EC2 instance. Chef Server uses Solr for all the text searching. After a couple of days of Chef Clients making frequent connections to the Chef Server, the server crashed . Solr was becoming a memory hog.

The Micro EC2 instance is just a 613 MB memory machine and since the Chef Server is just a single instance which I use to provision my infrastructure which I have full control over, I didn’t drill down into what was the exact reason for the memory spike, instead cheated and upgraded to a bigger EC2 instance and things never failed from there. But I thought I should atleast search and find out some patterns with Solr and memory usage and I stumbled across this :

(@Cherian_Thomas : glad to see an MEC alumnus here! )

Micro instances frequently skip clock cycles and not in the least recommended.

The cpu clock cycles are shared in a micro - AWS only guarantees short bursts of CPU when available for t1.micro’s, but that’s okay for my servers running on micros anyways. I’m concerned about skipped clock cycles, though. Can you shed some light on that, if we’re not talking about the same thing ?

So if for some reason another set of preferred EC2 units need CPU, your micro instance will not get for that moment, which is basically skipping.

okay … i had the same idea , thanks anyways , Cherian