Living 20 minutes from downtown Philadelphia, I’m a big fan of our hockey mascot, Gritty. Recently I’ve been playing around with a website called character.ai, and one of the neat things that site lets you do is create bots based on characters, real or imaginary. For example, there is one character based on Albert Einstein, and another is based on Darth Vader. So I decided I would create a character based on Gritty.
I immediately regretted that.
The AI powering that site is… frightfully good, to say the least. After seeding the character with just a handful of tweets from Gritty’s Twitter feed, the bot quickly took on a life of its own and said things that I would absolutely expect the real Gritty to say.
For example, let’s start with the no-fly list:
Next I asked Gritty about his diet, and the answers the bot gave were concerning, to say the least:
I’m a big fan of Amazon S3 for storage, and I like it so much that I use Odrive to sync folders from my hard drive into S3 use S3 to store copies of all of my files from Dropbox as a form of backup. I only have about 20 GB of data that I truly care about, so that should be less than a dollar per month for hosting, right? Well…
Close to 250 GB billed for last month. How did that happen?
I’m a big fan of the Discord Musicbot, and run it on some Discord servers that I admin. Wanting to run it on a server, I first created an Ansible playbook and launched a server on Digital Ocean. But after a few months, I noticed that the server was sitting over 90% idle. Surely there had to be a better way.
So I next tried Docker, and created a Dockerized version of the Musicbot. I was quite happy with how much easier it was to spin up the bot, but still didn’t want to run it on a dedicated server on Digital Ocean. Aside from having unused capacity, if that machine were to go down, I’d have to do something manually.
I thought about running it in some sort of hosted Docker cluster, and came across Amazon’s container service. So this post is about creating your own cluster in ECS and hosting a Docker container in it. I found the process slightly confusing when I did it the first time, and wanted to share my experience here.
While S3 is a great storage platform, what happens if you accidentally delete some important files? Well, S3 has a mechanism to recover deleted files, and I’d like to go into that in this post.
First, make sure you have versioning enabled on your bucket. This can be done via the API, or via the UI in the “properties” tab for your bucket. Versioning saves every change to a file (including deletions) as a separate version of said object, with the most recent version taking precedence. In fact, a deletion is also a version! It is a zero-byte version which has a “DELETE” flag set. And the essence of recovering undeleted files simply involves removing the latest version with the “DELETE” flag.
This is what that would look like in the UI:
To undelete these files, we’ll use a script I created called s3-undelete.sh, which can be found over on GitHub:
At my day job, I get to write a bit of code. I’m fortunate that my employer is pretty cool about letting us open source what we write, so I’m happy to announce that two of my projects have been open sourced!
The first project is an app which I wrote in PHP, it can be used to compare an arbitrary number of .ini files on a logical basis. What this means is that if you have ini files with similar contents, but the stanzas and key/value pairs are all mixed up, this utility will read in all of the .ini files that you specify, put the stanzas and their keys and values into well defined data structures, perform comparisons, and let you know what the differences are. (if any) In production, we used this to compare configuration files for Splunk from several different installations that we wanted to consolidate. Given that we had dozens of files, some having hundreds of lines, this utility saved us hours of effort and eliminated the possibility of human error. It can be found at:
As a service to the Philly tech community (and because folks asked), I took notes at tonight’s presentation, called “Security Practices for DevOps Teams”. It was presented by Chris Merrick, VP of Engineering at RJMetrics.
Security is a “cursed role”
…in the sense that if you’re doing a really good job as a security engineer, no one knows you exist.
It isn’t sexy
It’s hard to quantify
It’s never done
As DevOps engineers, we are all de facto security engineers
Some tips to avoid ending up like this [Picture of a dismembered C3PO]
Obscurity is not Security
“A secret endpoint on your website is not security”
“Don’t rely on randomness to secure things”
Do not give more privileges than are needed
If you talk to an insecure system, you’re at risk
Breaking into datacenters
Any vector that comes through an application you developed
Applications you didn’t write
Phishing, social engineering
Authentication is who you are
Authorization is what you can access
Don’t access production directory
Good news: this is our job anyways
Don’t spread private keys around
Don’t put in your Dropbox
Don’t let it leave the machine you generated it on
Use SSH agent forwarding
ssh -A you@remote
Don’t use shared accounts
Be able to revoke access quickly
Time yourself. Go.
We use Amazon OpsWorks to help us achieve these goals
Chef+AWS, with some neat tricks: simple autoscaling, application deployment, and SSH user management
“Logs are your lifeline”
When you get into a high pressure security investigation, you start with your logs
Capture all authentication events, privilege, escalations, and state changes.
From your Os and all running applications
Make sure you can trust your logs
Remember – they’re your lifeline
Have a retention policy
We keep 30 days “hot”, 90 days “cold”
Logging – ELK
We use ELK for hot log searching
Kibana creates logs and lets you monitor your application in real time
Keep unencrypted secrets out of code
Otherwise, a MongoLab exploit becomes your exploit
Don’t keep old code around
Make deployment and rollback easy
More good news: this is our job anyways
When dealing with a security issue, the last thing we need a “hard last step” in order to get the fix out
Don’t use your root account, ever.
Set a long password and lock it away
Set a strong password policy and require MFA
Don’t create API keys where API access isn’t needed
Same goes for a console password
Use Managed Policies
To make management easier
Use Roles to gran taccess to other systems
No need to deploy keys, auto-rotates
IAM Policy Pro Tips
Don’t use explicit DENY policies
Keep in mind that everything is denied by default
Don’t assume your custom policy is correct just because it saves – the interface only confirms the JSON is valid
Use the policy simulator
Know Thy Enemy
People are out there scanning for AWS keys – treat your private key like a private SSH key
In this post, I’m going to discuss how to create a GitHub repo and upload (or “push”) it to GitHub, a popular service for hosting Git repositories.
What is revision control and why do I need it?
The concept of revision control is a system which tracks changes to files. In programming, that is usually program code, but documents and text files can also be tracked. Using revision control will give the following benefits:
You will know what was changed, when it was changed, and who changed it
Multiple people can collaborate on a project without fear of overwriting each others’ changes.
Protection against accidentally deleting a critical file. (revision history is usually read-only)
In GitHub, we store revisions in “repositories” or “repos” for short. As of this writing, the #1 service for storing Git repositories is GitHub. They offer free hosting for Git repositories.
Earlier tonight, I had the pleasure of attending a presentation from Chris Munns of Amazon at the offices of First Round Capital about scaling your software on AWS past the first 10 million users. I already had some experience with AWS, but I learned quite a few new things about how to leverage AWS, so I decided to write up my notes in a blog post for future reference, and as a service to other members of the Philadelphia tech community.
Without further preamble, here are my notes from the presentation: