v1.10.0

Changelog

✨ New Features

• Metrics can be configured using a URL.

🏗 Improvements

• Detection of failed application pods in readiness checks.

🐛 Bug Fixes

• Adjusted generated experiment parameter definitions to account for baseline values.
• Cluster roles from multiple generated experiments no longer conflict.

🛑 Breaking Changes

• The Experiment CRD for metric definitions has changed to allow configuring metrics using a URL instead of a Service selector. The local and pods metrics types have been consolidated to a generic kubernetes type; all other metric types now require a URL. Instances of v1alpha1/Experiment are automatically converted to the new v1beta1 schema, however the following previously supported behaviors no longer work:
  • The v1alpha1 specific port field no longer handles string values (previously named ports were allowed).
  • The v1alpha1 service selector will only consider a single matching service (previously the first matching service to capture the metric successfully was used).

v1.9.3

Changelog

🏗 Improvements

• Added error-rate and duration objectives to the application.

v1.9.2

Changelog

🏗 Improvements

• Add a default TTL for server generated trials
• Validate parameter baseline configuration prior to experiment creation
• Do not fail baseline trials due to metric constraints
• Do not require patch target name for trial job patches
• Do not extend completion time for Prometheus metrics unless data is missing
• Do not duplicate readiness checks for patched resources
• Stop using dgrijalva/jwt-go due to known vulnerabilities
• Improve shell completion for multiple choice flags

⏳ Deprecated

• The following features related to metric definitions will be removed in a subsequent release:
  • Named ports
  • Matching multiple Service resources by label

🧨 Known Issues

• The controller will not start when using namespaced role bindings. See #195.

v1.9.1

Changelog

🏗 Improvements

• Generated utilization queries collect information for pods that terminate during the trial
• Built with Go 1.15

🐛 Bug Fixes

• Running redskyctl results caused a panic
• Setup tasks fail due to missing waitFn
• Locust scenarios can only be configured using nanoseconds

🧨 Known Issues

• The controller will not start when using namespaced role bindings. See #195.

v1.9.0

Changelog

📰 Announcements

Red Sky Ops is now Application Optimization within The StormForge Platform! To learn more, check out the announcement. There will be additional changes in upcoming releases as we switch to the new branding.

✨ New Features

• There is a new "application" type for declaratively describing your application
• Application state for a specific trial can now be exported using redskyctl export
• Experiments can be generated from an application descriptor using redskyctl generate experiment

🏗 Improvements

• Trial start and end times are now sent to the API

⏳ Deprecated

• The --name flag on the login and revoke commands has been deprecated in favor of the global --context flag, it will be removed in a subsequent release
• The REDSKY_* environment variables have been replaced by STORMFORGE_* equivalents
• The redskyctl results command will be removed entirely in a subsequent release

🧨 Known Issues

• The controller will not start when using namespaced role bindings. See #195.

🛑 Breaking Changes

• The redskyctl results command is no longer self-contained, instead it will open the default browser pointing at https://app.stormforge.io/
• The Docker images have moved from the redskyops organization to thestormforge
• The configuration file has moved from ~/.config/redsky/config to ~/.config/stormforge/config, additionally some configuration field names have changed. Using this version will automatically migrate your configuration file and older versions of the tool will not be able to use the configuration file
• The documentation has been moved to https://docs.stormforge.io/

v1.8.1

📰 Announcements

Red Sky Ops is now Application Optimization within The StormForge Platform! To learn more, check out the announcement. There will be additional changes in upcoming releases as we switch to the new branding.

Changelog

🏗 Improvements

• Shell completion can suggest experiment and trial names.

🐛 Bug Fixes

• Resetting the controller no longer fails if you are not logged in.
• Discourage zombie trial jobs from awakening.
• Fixed an issue where Pushgateway metrics were not collected.
• Setup tasks were not honoring skipCreate or skipDelete.

v1.8.0

Changelog

✨ New Features

• The baseline values can now be specified in the parameter definitions; the baseline will only be recorded if all parameters use the baseline field.
• Metrics can be constrained using min and max fields, observed values outside the allowed range will cause the trial to be marked a failure.
• Metrics can be captured but not optimized by setting the optimize field to false.

🏗 Improvements

• Generating RBAC for an experiment will consider CRD names from the current cluster.

🐛 Bug Fixes

• Fixed an issue which caused the version number to be reported incorrectly.

🧨 Known Issues

• The controller will not start when using namespaced role bindings. See #195.

🛑 Breaking Changes

• The redskyctl check server command has been removed.

⏳ Deprecated

• The redskyctl results command will be removed in a subsequent release, the functionality will be available on the application website.

v1.7.1

Changelog

✨ New Features

• The redskyctl ping command can be used to check API connectivity

🏗 Improvements

• Reduced the amount of metrics ingested by the built-in Prometheus
• Included the ConfigMap reloader with the built-in Prometheus
• The Push Gateway URL is set as an environment variable on trial job pods
• Display a more useful message when openning a web browser fails
• Included failure reasons in trial CSV output
• UI axis sliders are hidden for string valued parameters

v1.7.0

Changelog

✨ New Features

• Support for string valued parameters
• Trial plots support adjusting the axis ranges

🏗 Improvements

• Enhanced the Experiment status with conditions
• Included a push gateway for trial jobs to store metrics
• Trial failures reasons are now reported to remote servers
• Installation manifests can be written to a directory
• Individual trials can be identified on trial plots

🐛 Bug Fixes

• Restrict utilization queries to the trial namespace
• Utilization query labels are cleaned to match Prometheus
• Redeploy the controller for changes to the authorization secret
• Explicit sort orders for experiments and trials are honored

🧨 Known Issues

• The controller will not start when using namespaced role bindings. See #195.

⏳ Deprecated

• The redskyctl results command will be removed in a subsequent release, the functionality will be available on the application website.
• The redskyctl check server command will be removed in the next release.

v1.6.5

Changelog

🏗 Improvements

• Readiness checks can inspect a string valued field in the status of the target object.