AutoSpotting
A simple and easy to use tool designed to significantly lower your Amazon AWS costs by automating the use of spot instances.
Why?
We believe that AWS EC2 is often pricier than it should be, and that the pricing models that can significantly reduce costs are hard to be reliably used by humans and are better handled by automation.
There are many way to automate the use of spot instances, some offered out of the box by AWS and some from third parties, each with its own characteristics and drawbacks.
Unlike all those options, we developed a novel, simple but effective way to make it much easier to convert existing infrastructure within minutes, with minimal configuration changes, negligible additional infrastructure and runtime costs, safely and securely and without any vendor lock-in.
Note: it's not necessarily the most cost-efficient for a given AutoScaling group, although it does perform pretty well in practice. The main focus is the ease of adoption on large infrastructure such as environments with hundreds of AutoScaling groups or even hundreds of AWS accounts where it can be rolled out and enabled in almost no-time compared to any other tool out there. Once you tested it and you are confident with it, it can even be enabled against all the groups from an AWS account without touching their configuration.
It also tries to be as cheap as possible to run, with negligible runtime costs and being open source the software is free of charge if you build it from source. In addition we offer inexpensive enterprise-grade support plans that should barely be noticeable at scale, but just enough to support further development, unlike similar commercial tools that claim a very significant chunk of your savings.
This approach allows a large number of companies and individuals to significantly reduce their infrastructure costs or get more bang for the same buck. They can now easily get access to cheap compute capacity so they can spend their scarce resources developing innovative products not paying for overpriced compute capacity.
How does it work?
Once installed and enabled by tagging existing on-demand AutoScaling groups, AutoSpotting gradually replaces their on-demand instances with spot instances that are usually much cheaper, at least as large and identically configured to the group's members, without changing the group configuration in any way. For your peace of mind, you can also keep running a configurable number of on-demand instances given as percentage or absolute number.
This can be seen in action below, you can click to expand the animation:
It implements some complex logic aware of spot and on demand prices, including for different spot products and configurable discounts for reserved instances or large volume customers. It also considers the specs of all instance types and automatically places bids to instance types and prices chosen based on flexible configuration set globally or overridden at the group level using additional tags, but these overrides are often not needed.
A single installation can handle all enabled groups in parallel across all available AWS regions, but can be restricted to fewer regions if desired.
Your groups will then monitor and use these spot instances just like they would do with your on-demand instances. They will automatically join your load balancer and start receiving traffic once passing the health checks, and the traffic would automatically be drained on termination.
What savings can I expect?
The savings it generates are often in the 60-80% range, but sometimes even up to 90%, like you can see in the graph below.
What's under the hood?
The entire logic described above is implemented in a Lambda function deployed using CloudFormation or Terraform stacks that can be installed and configured in just a few minutes.
The stack assigns the function the minimal set of IAM permissions required for it to work and requires no admin-like cross-account permissions. The entire code base can be audited to see how these permissions are being used and even locked down further if your audit discovers any issues. This is not a SaaS, there's no component that calls home and reveals any details about your infrastructure.
The Lambda function is written in the Go programming language and the code is compiled as a static binary compressed and uploaded to S3. For evaluation or debugging purposes, the same binary can run out of the box locally on Linux machines or as a Docker container on Windows or macOS. Some people even run these containers on their existing Kubernetes clusters assuming the other resources provided by the stack are implemented in another way on Kubernetes.
The stack also consists of a Cron-like CloudWatch event, that runs the Lambda function periodically to take action against the enabled groups. Between runs your group is entirely managed by AutoScaling (including any scaling policies you may have) and load balancer health checks, that can trigger instance launches or replacements using the original on-demand launch configuration. These instances will be replaced later by better priced spot instances when they are available on the spot market.
Read here for more information and implementation details.
FAQs
Frequently asked questions about the project are answered in the FAQ, please read this first before asking for support.
If you have additional questions not covered there, they can be easily added to the source of the FAQ by editing in the browser and creating a pull request, and we'll answer them while reviewing the pull request.
Getting Started
Just like in the above animation, it's as easy as launching a CloudFormation (or
Terraform)
stack and setting the (configurable) spot-enabled tag on the AutoScaling
groups where you want it enabled to true.
All the required infrastructure and configuration will be created automatically, so you can get started as fast as possible.
For more detailed information you can read this document
Note
- the binaries launched by this stack are distributed under a proprietary license, and are free to use for evaluation, up to $1000 monthly savings. Once you reach this limit you'll need to either switch to the inexpensive supported binaries (designed to cost a small fraction of around 1% of your total savings for supporting further development), or you can build your own binaries based on the open source code and run it for free.
Support
Community support is available on the gitter chat room, where the main authors and other users are likely to help you solve issues.
Note: This is offered on a best effort basis and under certain conditions, such as using the latest version of the evaluation binaries.
If you need more comprehensive support you will need to purchase a support plan.
Contributing
Unlike multiple commercial products in this space that cost a lot of money and attempt to lock you in, this project is fully open source and developed in the open by a vibrant community.
We urge you to support us on Github Sponsors if this software helps you save any significant amount of money, this will greatly help further development.
Financial sponsorship is voluntary, it's also fine if you just try it out and give feedback, report issues, improve the documentation, write some code or assign a developer to work on it, or even just spread the word among your peers who might be interested in it. Any sort of support would be greatly appreciated and would make a huge difference to the project.
Note: Non-trivial code should be submitted according to the contribution guidelines.
Proprietary binaries
The source code is and will always be open source, so you can build and run it yourself, see how it works and even enhance it if you want.
But if you want to conveniently get started or update within minutes without setting up and maintaining a build environment or any additional infrastructure, we have pre-built evaluation binaries that will save you significant amounts of time and effort.
These can be used for evaluation purposes as long as the generated monthly savings are less than $1000. Once you reach this level you will need to either purchase an inexpensive stable build that doesn't have this limitation, and also comes with a support plan, or you can build AutoSpotting from source code.
The support license costs vary by group, region and AWS account coverage and can also be paid through Patreon.
Individuals and companies supporting the development of the open source code get free of charge support and stable build access for a year since their latest contribution to the project.
Note:
- even though these evaluation builds are usually stable enough, they may not have been thoroughly tested yet and come with best effort community support.
- the docker images available on DockerHub are also distributed under the same binary license and the costs are the same.
Stable builds
Carefully tested builds suitable for Enterprise use will be communicated to Patreon backers as soon as they join.
They come with support from the author, who will do his best to help you successfully run AutoSpotting on your environment so you can get the most out of it. The feature requests and issues will also be prioritized based on the Patreon tier.
Please get in touch on gitter if you have any questions about these stable builds.
Compiling and Installing
It is recommended to use the evaluation or stable binaries, which are easy to install, support further development of the software and allow you to get support.
But if you have some special needs that require some customizations or you don't want to rely on the author's infrastructure or contribute anything for longer term use of the software, you can always build and run your customized binaries that you maintain on your own, just keep in mind that those won't be supported in any way.
More details are available here
Users
Autospotting is already used by hundreds of individuals and organizations around the world, and we estimate to already save them in the millions of dollars monthly. Some of them we know of are mentioned in the list of notable users.
The following deserve a special mention for contributing significantly to the development effort (listed in alphabetical order):
License
This software is distributed under the terms of the OSL-3.0 license.
The official binaries are licensed under this proprietary license.
