⭐ Drop a star to support OptScale ⭐

FinOps and MLOps platform to run ML/AI and regular cloud workloads with optimal performance and cost

OptScale is an open source FinOps and MLOps platform that provides cloud cost optimization for all types of organizations and MLOps capabilities like experiment tracking, model versioning, ML leaderboards.

OptScale schema

MLOps capabilities	FinOps and cloud cost optimization
ML Leaderboards with candidates and qualifications Dataset and model tracking and versioning Run metrics and experiment tracker Hypertuning integrated with Optuna Training launcher ML Model training profiler	Optimal utilization of Reserved Instances, Savings Plans, and Spot Instances Unused resource detection R&D resource power management and rightsizing S3 duplicate object finder Resource bottleneck identification Optimal instance type and family selection Databricks support S3 and Redshift instrumentation VM Power Schedules

You can check OptScale live demo to explore product features on a pre-generated demo organization.

Learn more about the Hystax OptScale platform and its capabilities at our website.

Demos

ML Tasks	ML Leaderboards

Experiment tracking	ML model profiling integration

Datasets	Hypertuning

Databricks connection	Cost and performance recommendations

Cost geo map	VM Power Schedules

Reserved Instances and Savings Plans	Cost breakdown by owner

OptScale components and architecture

Getting started

Minimum hardware requirements for OptScale cluster: CPU: 8+ cores, RAM: 16Gb, SSD: 150+ Gb.

NVMe SSD is recommended.
OS Required: Ubuntu 20.04.
The current installation process does not work on Ubuntu 22.04

Installing required packages

Run the following commands:

sudo apt update ; sudo apt install git python3-venv python3-dev sshpass

Pulling optscale-deploy scripts

Clone the repository

git clone https://github.com/hystax/optscale.git

Change current directory:

cd optscale/optscale-deploy

Preparing virtual environment

Run the following commands:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Kubernetes installation

Run the following command: comma after ip address is required

ansible-playbook -e "ansible_ssh_user=<user>" -k -K -i "<ip address>," ansible/k8s-master.yaml

where - actual username; - host ip address, ip address should be private address of the machine, you can check it with

ip a

If your deployment server is the service-host server, add "ansible_connection=local" to the ansible command.

Creating user overlay

Edit file with overlay - optscale-deploy/overlay/user_template.yml; see comments in overlay file for guidance.

Cluster installation

run the following command:

./runkube.py --with-elk  -o overlay/user_template.yml -- <deployment name> <version>

or if you want to use socket:

./runkube.py --use-socket --with-elk  -o overlay/user_template.yml -- <deployment name> <version>

deployment name must follow the RFC 1123: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/

version:

• Use hystax/optscale git tag (eg: 2023110701-public) if you use optscale public version.
• Use your own tag version if you build your optscale images (eg: latest).

please note: if you use key authentication, you should have the required key (id_rsa) on the machine

Cluster update

Run the following command:

./runkube.py --with-elk  --update-only -- <deployment name>  <version>

Get IP access http(s):

kubectl get services --field-selector metadata.name=ngingress-nginx-ingress-controller

Troubleshooting

In case of the following error:

fatal: [172.22.24.157]: FAILED! => {"changed": true, "cmd": "kubeadm init --config /tmp/kubeadm-init.conf --upload-certs > kube_init.log", "delta": "0:00:00.936514", "end": "2022-11-30 09:42:18.304928", "msg": "non-zero return code", "rc": 1, "start": "2022-11-30 09:42:17.368414", "stderr": "W1130 09:42:17.461362  334184 validation.go:28] Cannot validate kube-proxy config - no validator is available\nW1130 09:42:17.461709  334184 validation.go:28] Cannot validate kubelet config - no validator is available\n\t[WARNING IsDockerSystemdCheck]: detected \"cgroupfs\" as the Docker cgroup driver. The recommended driver is \"systemd\". Please follow the guide at https://kubernetes.io/docs/setup/cri/\nerror execution phase preflight: [preflight] Some fatal errors occurred:\n\t[ERROR Port-6443]: Port 6443 is in use\n\t[ERROR Port-10259]: Port 10259 is in use\n\t[ERROR Port-10257]: Port 10257 is in use\n\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-apiserver.yaml]: /etc/kubernetes/manifests/kube-apiserver.yaml already exists\n\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-controller-manager.yaml]: /etc/kubernetes/manifests/kube-controller-manager.yaml already exists\n\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-scheduler.yaml]: /etc/kubernetes/manifests/kube-scheduler.yaml already exists\n\t[ERROR FileAvailable--etc-kubernetes-manifests-etcd.yaml]: /etc/kubernetes/manifests/etcd.yaml already exists\n\t[ERROR Port-10250]: Port 10250 is in use\n\t[ERROR Port-2379]: Port 2379 is in use\n\t[ERROR Port-2380]: Port 2380 is in use\n\t[ERROR DirAvailable--var-lib-etcd]: /var/lib/etcd is not empty\n[preflight] If you know what you are doing, you can make a check non-fatal with `--ignore-preflight-errors=...`\nTo see the stack trace of this error execute with --v=5 or higher", "stderr_lines": ["W1130 09:42:17.461362  334184 validation.go:28] Cannot validate kube-proxy config - no validator is available", "W1130 09:42:17.461709  334184 validation.go:28] Cannot validate kubelet config - no validator is available", "\t[WARNING IsDockerSystemdCheck]: detected \"cgroupfs\" as the Docker cgroup driver. The recommended driver is \"systemd\". Please follow the guide at https://kubernetes.io/docs/setup/cri/", "error execution phase preflight: [preflight] Some fatal errors occurred:", "\t[ERROR Port-6443]: Port 6443 is in use", "\t[ERROR Port-10259]: Port 10259 is in use", "\t[ERROR Port-10257]: Port 10257 is in use", "\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-apiserver.yaml]: /etc/kubernetes/manifests/kube-apiserver.yaml already exists", "\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-controller-manager.yaml]: /etc/kubernetes/manifests/kube-controller-manager.yaml already exists", "\t[ERROR FileAvailable--etc-kubernetes-manifests-kube-scheduler.yaml]: /etc/kubernetes/manifests/kube-scheduler.yaml already exists", "\t[ERROR FileAvailable--etc-kubernetes-manifests-etcd.yaml]: /etc/kubernetes/manifests/etcd.yaml already exists", "\t[ERROR Port-10250]: Port 10250 is in use", "\t[ERROR Port-2379]: Port 2379 is in use", "\t[ERROR Port-2380]: Port 2380 is in use", "\t[ERROR DirAvailable--var-lib-etcd]: /var/lib/etcd is not empty", "[preflight] If you know what you are doing, you can make a check non-fatal with `--ignore-preflight-errors=...`", "To see the stack trace of this error execute with --v=5 or higher"], "stdout": "", "stdout_lines": []}

run the following command to reset k8s and retry the installation command:

sudo kubeadm reset -f
ansible-playbook -e "ansible_ssh_user=<user>" -k -K -i "<ip address>," ansible/k8s-master.yaml

In case of the following error during cluster initialization:

requests.exceptions.ConnectionError: HTTPConnectionPool(host='172.22.24.157', port=2376): Max retries exceeded with url: /v1.35/auth (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f73ca7c3340>: Failed to establish a new connection: [Errno 111] Connection refused'))

check the docker port is opened:

sudo netstat -plnt | grep 2376

and open port in docker service config:

sudo nano /etc/systemd/system/docker.service

add this line (do not dorget to close docker port after installing Optscale)

ExecStart=/usr/bin/dockerd -H fd:// -H tcp://0.0.0.0:2376

then reload config and restart docker

sudo systemctl daemon-reload
sudo service docker restart

Roadmap

• Cost plugin for MLflow, WanDB, and neptune.ai
• Integration with Optuna to optimize Reserved Instance and other hardware parameter usage
• Model versioning
• Better hardware selection recommendations based on usage patterns and algorithms

Documentation

Read the full OptScale documentation 📖

Contributing

Please read and accept our Contribution Agreement before submitting pull requests.

Community

Hystax drives FinOps & MLOps methodology and has crafted a community of FinOps-related people. The community discusses FinOps & MLOps best practices, our experts offer users how-tos and technical recommendations, and provide ongoing details and updates regarding the open-source updated OptScale solution.

You can check it out on FinOps and MLOps in practice website

Contacts

Feel free to reach out to us with questions, feedback, or ideas at [email protected]. You can check out the latest news from Hystax at:

• LinkedIn
• Twitter
• Facebook
• YouTube