General

Telemetry Aware Scheduling

• update GA workflows to use the latest package versions available
• update extender, TAS, E2E tests module to use K8s 1.28.4, golang v1.21
• add Healthcheck to TAS container

GPU Aware Scheduling

• update GAS modules, libraries to latest available
• update docs/README.md

Configurator

• update configurator library version and go version to 1.21
• backups are no longer left to the filesystem if there aren't any changes to the configuration files. This then allows to run the configurator also in a cronjob periodically.

General

• [FEATURE] The PAS configurator has been added to the Platform Aware Scheduling project. The configurator is a tool for setting up Telemetry Aware Scheduling (TAS) or GPU Aware Scheduling (GAS) for a Kubernetes scheduler installation. For more details please go to https://github.com/intel/platform-aware-scheduling/tree/master/configurator#pas-configurator
• [CI] The project's CI will now run static analysis for the GPU Aware Scheduling module e72fc46

Telemetry Aware Scheduling

• Bump extender, Telemetry Aware Scheduling code and E2E module dependencies to use K8s v1.27 and go 1.20 a9133c2, b765026 and 0649248
• Update GithubAction workflow dependencies dde1fc9
• Update the Telemetry Aware Scheduling health-metric demo to work with K8s descheduler v1.27by adding the evictLocalStoragePods property 5ef9fe6
• Relocate the Telemetry Aware Scheduling main and related demo pods from "default" into their own namespaces 0990644
• Add gofuzz testing for extender's Filter and Prioritize methods a2dfbd0 and 3144ce7
• Added small follow-up validations found through fuzz testing b4daf15, 5e49d4d and 0ae01c1

GPU Aware Scheduling

• Bump GAS module dependencies to use K8s 1.27.4, go 1.20 1142ca2
• Rename GAS dockerfile 154b8e3
• Fix GAS linting issues 85aa651

General

• Updated the issue templates for the repo: customers now will be asked to provide K8s default scheduler and Telemetry Aware Scheduling logs when creating an issue 644acdb

Telemetry Aware Scheduling

[Important] This current version of TAS (v0.5.2) does not have a corresponding image on Docker hub (the current image available is v0.5.0.). To make use of the TAS v0.5.2 source code please use link, or more specifically:

Note: If you want to create the build and the image you can still do it by running make build && make image . This will build locally the image tasextender. Once created you may replace it into the deployment file.

This release contains the following changes:

• Compacting health-metric demo files 5e100a7 c355108
• Address GithubActions set-ouput function deprecation 4ffb0d7
• TAS Configuration script refactoring and bug fix 6da8c79 fac4722
• Power demo & docs refresh fda8e05
• Updating demos (health-metric, labeling) with the supported version of the Descheduler project 88d6cd4

GPU Aware Scheduling

• Fix GAS xe link label usage f61f135
• Version updates and lint fixes 0342090
• Add update-baseimage step to makefile targets 0342090
• Added 5 examples to demonstrate GPU aware scheduling features 380d857

General

• Address CVE-2022-27664, CVE-2021-43565 found by dependabot in both TAS and GAS modules
• Extender now takes into account values for QPS and burst when configuring the go KubeClient
• Updated the issue templates for of repo. Customers now be asked ask for more inputs & further details when raising an issue

Telemetry Aware Scheduling

• Added more E2E tests for the TAS module. They will not run in the intel/platform-aware-scheduling repo
• Made small security updates to the health-metric demo and the current GithubActions workflows
• TAS deployment will now use the newly released 0.5.0 image

GPU Aware Scheduling

• Informer interval updated with new value
• GAS module uses extender package version v0.5.0
• GAS module can be configured with burst and QPS values
• Deployment sets higher burst and QPS values by default

General

• Updated dependencies with Kubernetes 1.25 and Go 1.19
• Improvement to the descheduling & labelling strategies
• Security improvements to the TAS and custom-metrics-api containers and of the health-metric demo
• Address issues with Dockerfile and code license and copyright
• Documentation updates
• Linting improvements

Telemetry Aware Scheduling

• Release of telemetry-aware-scheduling/v0.4.0 aligned to the container image
• Improvements to the descheduling and labelling strategies:
• • Refactored descheduling strategy logic: https://github.com/intel/platform-aware-scheduling/commit/533c754904235de79e70593c5fa67f8a8b23ce39 and https://github.com/intel/platform-aware-scheduling/commit/6106d3fb3294c1a2975764bb98abc7926169a98c
• • Strategies are enforced sequentially to reduce number of spawned goroutines: https://github.com/intel/platform-aware-scheduling/commit/75cf83ec137b5e059d675248dc128842150e1a4c
• • Update burst and QPS parameters for K8s go client to avoid throttling: https://github.com/intel/platform-aware-scheduling/commit/75cf83ec137b5e059d675248dc128842150e1a4c
• Bump to K0s v1.25 and go 1.19: https://github.com/intel/platform-aware-scheduling/commit/d00722e52da31a951b775aa933c1dcbd154867a1
• Improved security of the TAS and custom-metrics-api containers and of the health metric demo: https://github.com/intel/platform-aware-scheduling/commit/692d5ffb7ce1a09f038003f57aad2411d2f93474
• Address issues with Dockerfile and code license and copyright: https://github.com/intel/platform-aware-scheduling/commit/5661d468f08718f96f7b8306a5d4c05ed87223d4, https://github.com/intel/platform-aware-scheduling/commit/6f6883fa73c2b8b6b942fd7e189b12fef2607aa8

GPU Aware Scheduling

• GPU allocations can be targeted to Xe Linked cards
• Support single-numa GPU allocations, when multiple GPUs are allocated.
• Bugfixes, especially around split (long) label handling

General

• Updated dependencies with Kubernetes 1.24 and Go 1.18
• Documentation updates
• Security improvements
• Refactor
• Lint fixes

Telemetry Aware Scheduling

• Release of telemetry-aware-scheduling/v0.3.0 aligned to the container image
• Documentation updates 25a646ece15aaf4c549d8152c4ffbbfc61f8a009
• Bump to k8s 1.24 and Go 1.18 0eb3b7275eea495875707386dc9b2680c40b8525
• Update of extender/v0.3.0 2b65538f0bc092d29d199ae9657986d5c8bc630e
• Bug fix on a race condition 303e65614472a8670c5def88b8524ac9e7a6309e
• Security improvements 56c45732b4861f8ac0f7adf452b93ae38ccdf303
• Refactoring 56c45732b4861f8ac0f7adf452b93ae38ccdf303
• Adding more unit tests d5a90c8af737811a4d29ee3e412f513dde1535ae
• Lint fixes by golangci-lint d5a90c8af737811a4d29ee3e412f513dde1535ae

GPU Aware Scheduling

• Release of gpu-aware-scheduling/v0.4.0 aligned to the container image
• New feature: 'gas-same-gpu' Pod annotation support to ensure same GPU device is allocated to listed containers within Pod
• Bump to k8s 1.24 and Go 1.18
• Version updates

General

• Updated dependencies with Kubernetes 1.23 and Go 1.17
• Documentation updates
• Security improvements
• Lint fixes

Telemetry Aware Scheduling

• New Logical Operator for multimetric rules
• Release of telemetry-aware-scheduling/v0.2.0 aligned to the container image
• Update of extender/v0.2.0
• Fixes for issues #60 #78
• New test cases for e2e
• Guide example for the Labeling strategy

GPU Aware Scheduling

• Release of gpu-aware-scheduling/v0.3.0 aligned to the container image
• Support for GPU tile resource and related telemetry originated node labels
• Refactoring

Telemetry Aware Scheduling

• New Labeling Strategy feature
• Release of telemetry-aware-scheduling/v0.1.0 (also tagged as tas_v0.1) aligned to the container image
• Documentation updates
• Security and error handling improvements
• Lint fixes

GPU Aware Scheduling

• Release of gpu-aware-scheduling/v0.2.0 aligned to the container image
• Telemetry based label support for finer-grained gpu resource controls, support for gpu allowlist and denylist
• Documentation updates
• Security capability improvements
• Lint fixes

Extender

• Release of extender/v0.1.0 as the first module version
• Lint fixes

Module versioning

This release adds go module version git tags in the format described in https://go.dev/doc/modules/version-numbers for the folders telemetry-aware-scheduling, gpu-aware-scheduling, and extender

• Configure TAS set-up to use kube-scheduler Configuration APIs for K8s v22 onwards
• Updated dependencies with Kubernetes 1.22 and Go 1.16
• Updating TAS policy CRD to build with APIExtensions v1
• Updating Helm scripts to use the v1 of the apiregistration and rbac.authorization APIs
• Fix bugs related to issues #29 and #56
• Documentation update
• Improving actions workflow

• Create GPU Aware Scheduling under PAS
• Create GAS v0.1