dysnix / pancakeswap-ifo-scaler

pancakeswap-ifo-scaler

Auto-scaler for Pancakeswap Blockchain Nodes Infrastructure.

Principle algorithm

1. Event of new Pancakeswap IFO
2. Create k8s resource KEDA scaleobject
3. Resize blockchain cluster by scaling pod
4. Send notification to Telegram (by @predictkube_bot)
5. Cleanup k8s resources

Config environment variables

• TARGET_NAME - Kubernetes target controller for scaling (required)
• TARGET_NAMESPACE - Kubernetes namespace with controller (required)
• K8S_REPLICAS_COUNT - Replicas count for IFO period (required)
• TELEGRAM_TOKEN - Token for notification (required)
• TARGET_API_VERSION - Kubernetes target controller API version (default is apps.kruise.io/v1alpha1)
• TARGET_KIND - Kubernetes target controller Kind (default is CloneSet)
• NODE_URL - BSC JsonRPC Endpoint. Public endpoint by default

Run

Please add @predictkube_bot to any telegram group(s) for receive notifications.

Locally

export TELEGRAM_TOKEN="changeme"
pip install -r requirements.txt
python main.py

Docker

docker build -t ifo-scaler .
docker run -v $HOME/.kube:/root/.kube -e TELEGRAM_TOKEN="changeme" ifo-scaler

pancakeswap-ifo-scaler's Issues

Multiple ScaledObject with one scaleTargetRef are not supported

It seems that multiple ScaledObject referencing single scaleTargetRef are not supported with current Keda implementation.
Keda creates independent HPA objects for each ScaledObject, which leads to HPA wars to scale single resource.

k get hpa
NAME                                                          REFERENCE          TARGETS                                    MINPODS   MAXPODS   REPLICAS   AGE
keda-hpa-bsc-so                                               CloneSet/bsc-rpc   891/1200 (avg), 500m/1 (avg) + 1 more...   2         30        2          244d
keda-hpa-ifo-0x80828b68d126741eea62b4712ebf7ceb9d54cbaa-mgp   CloneSet/bsc-rpc   -9223372036854775808m/1 (avg)              1         100       2          3m38s

Proposed fix is to patch existing ScaledObject instead of creating a new one.