autotune
autotune copied to clipboard
kruize
Autonomous Performance Tuning for Kubernetes!
We need to investigate and find the best way to pass the non-container layer tunables. For example, OPENJ9_JAVA_OPTIONS takes precedence over JAVA_OPTIONS - https://www.eclipse.org/openj9/docs/cmdline_specifying/ ``` { "name": "update env", "spec":...
Parking it for Iteration design changes, Added EM Iteration Data abstraction as well (still needs some implementation, updated it to current level at my local as this can be used...
We are currently using 31 tunables in the openshift experiments as can be seen [here](https://github.com/kruize/autotune-results/blob/main/techempower/experiment-8/benchmark.yaml). They all need to be added to the respective layers.
The input JSON to the Experiment Manager will need to specify settings related to experiment trials as well as the deployment settings. See the [EM design discussion](https://github.com/kruize/autotune/discussions/227) for more details...
Current EM design doesn't have a mechanism to release the memory of it's internal components when it's idle, the components include process queues, stage processors, completed ETD's (which takes up...
Autotune searchSpace has duplicate tunables. Applied multiple container configs with the same tunable name (cpuRequest) with different values. Both these are picked as two different layers (container layer & my...
Autotune object id changes when deleted and created again. Steps to recreate: - Deploy autotune using the deploy.sh script from autotune repo in minikube - Deploy spring petclinic application from...
Querying /listApplications End point gives an error message: ["Error: Application null not found!"] Autotune doesn't detect multiple autotune objects ``` kubectl get autotune NAME AGE petclinic-autotune-0 21s petclinic-autotune-1 11s petclinic-autotune-2...