Since 2019 the release notes, potential breaking changes and mitigation procedures have only been communicated via private channels to the Patreon subscribers when releasing a new stable version.
Sign up on Patreon in order to receive these release notifications, as well as to get access to these supported stable versions. In the future certain premium features will only be available in the stable builds.
Yet another really beefy entry, but most of this was also captured in much more detail in this blog post.
Many thanks to HERE Technologies for supporting a lot of the development effort, to the Patreon supporters and to everyone who contributed code to this release.
- The method by which AutoSpotting terminates existing Auto Scaling Group
instances has changed. By default, AutoSpotting now uses the EC2
TerminateInstanceInAutoScalingGroupAPI. This API call ensures that any Termination Lifecycle Hooks that might be configured on the Group are respected, which was not the case in previous AutoSpotting versions. Users who depend upon the legacy behavior, which was to detach the instance from the Auto Scaling Group and terminate it, can setinstance_termination_methodtodetachin their deployment configurations. - Better handling of out of capacity situations.
- Better handling of VPC, DefaultVPC and EC2 Classic security groups.
- Support running in opt-out mode.
- Tagging launched spot instances with
launched-by-autospotting=true. - Obey the scale-in protection and termination protection for on-demand instances that previously were being replaced.
- Terraform module in the Terraform Registry.
- Documentation for running as a Kubernetes cron job instead of Lambda.
- Instance type updates and regional expansions. Note: some recent instance types are still not supported yet due to missing upstream pricing information.
- AutoSpotting now launches spot instances using the
RunInstancesAPI call, which allows us to simplify the logic considerably and fix a number of bugs. - The spot bidding engine was heavily refactored, using less memory and being much more scalable on large installations.
- Fix compilation of macOS.
- Smaller binaries.
- Build AutoSpotting using Go 1.11
A lot of time passed since the previous Changelog update, so this is a really beefy entry, we should definitely do this more often.
There were a lot of contributions from many individuals, thank you all for helping improve this project!
Special thanks once again to @xlr-8 who contributed a lot throughout the previous year, and is now a very active co-maintainer of the project.
- Consider GPUs in the instance compatbility checks
- Instance type updates and regional expansions
- Aggressive bidding strategy based on the current spot price, by @kartik894
- Prevent termination of last instance, by @universam1
- Support volume discounts (think Reserved Instance), by @masneyb
- Instance type whitelisting, by @raravena80
- Instance type blacklisting, by @binarylogic
- Consider the EBS pricing surcharge, by @vecchp
- Deploy using Terraform, by @xlr-8
- Significant refactorings, especially done by @xlr-8
- We're now using the native Lambda Go runtime, again thanks to @xlr-8
- Scalability fixes: concurrency by @thebigjc, API pagination by @ahaverbuch, support handling more than 50 AutoScaling groups concurrently, by @chaner
- Much more unit test coverage, thanks to @xlr-8 and @artemnikitin
- Buildsystem changes, we now also have automated lint and vet checks
- Dependency vendoring, by @xlr-8
- Lots of bugfixes, special thanks to @xlr-8
- Countless documentation updates by too many individuals to mention here
- The CloudFormation stack needs to be updated after build 633, because of the change to the Go Lambda runtime
- Some of the AutoScaling group tags used to override the global configuration were renamed to be more consistent, please refer to the current state of the documentation.
~Cristian
I forgot to update this in a while, so this is a quite big changelog entry.
I got the first major code contributions by other developers, so from now on the changelog entries will be split by author. It may also have a header like this one in which it will contain a short summary or a message from the author.
Special thanks to Hugo Rosnet, who contributed a lot of code that implemented a number of major features, helped me with multiple code reviews and kept me motivated enough to constantly work on this project.
Also thanks to Jay Wineinger who contributed a non-trivial piece of code, and to the other folks who contributed documentation, raised or discussed various Github issues.
~Cristian
- Big code refactoring to make the code more maintainable and testable.
- Buildsystem improvements (and regressions, since fixed).
- Updated regional and instance type coverage, thanks to ec2instances.info.
- Support restricting the execution to a given set of regions.
- Expose all configuration options also as CloudFormation stack parameters.
- Documentation updates and improvements.
- Random small cleanups.
- Update Lambda function's IAM permissions.
- The algorithm now supports keeping on-demand instances in each AutoScaling group.
- The algorithm is now configurable using tags set on the group and based on flags when executing it locally as a CLI tool.
- Significant test coverage increases.
- Significant clean-up and refactoring of the core algorithm.
- Documentation improvements.
- Pagination fix, making it work for users having many ASGs.
- Documentation for Elastic Beanstalk.
Major, breaking compatibility, packaging update: now using eawsy/aws-lambda-go for packaging of the Lambda function
- Switch to the golang-native eawsy/aws-lambda-go for packaging of the Lambda function code.
- This is a breaking change, updating already running CloudFormation stacks will also need a template update.
- Add versioning for the CloudFormation template.
- Buildsystem updates (both on Makefile and Travis CI configuration).
- Change build dependencies: now building Lambda code in Docker, use wget instead of curl in order not to download data unnecessarily.
- Remove the Python Lambda wrapper, it is no longer needed.
- Start using go-bindata for shipping static files, instead of packaging them in the Lambda zip file.
- Introduce a configuration object for the main functionality, not in use yet.
- Documentation updates and better formatting.
- Test and fix support for EC2 Classic
- Fix corner case in handling of ephemeral storage
- Earlier spot request tagging
- Regional expansion for R3 and D2 instances
- Add support for the new Ohio AWS region
- Add support in all the regions for the newly released instance types: m4.16xlarge, p2.xlarge, p2.8xlarge, p2.16xlarge and x1.16xlarge
Before this file was created, change logs used to be posted as blog posts:
- recent changes as of October 2016
- in the initial phase of the project they were posted at the end of the first announcement blog post