ltrinks/linkerd-scaler

Linkerd Scaler

About

Linkerd Scaler collects metrics from the Kubernetes metrics server and Linkerd proxy prometheus metrics server. An HPA style scaling algorithm can be used to scale pods.

Usage

Start by running:

./start.sh

Keep this script running for as long as you want the cluster alive, ^C to exit.

With the cluster running, in another terminal, kubectl and other commands can be executed on the cluster.

Outputs of the watcher are found in metrics/, these persist but are deleted on each new watcher start.

Run settings are found at the top of watcher/src/main.py. To configure Kubernetes HPA either comment out or run the nodevoto-hpa.yaml line in start.sh. Kubernetes HPA and the ACTIVE setting in main.py should not both be enabled.

To restart Linkerd Scaler without restarting the cluster run:

./start-watcher.sh

While Linkerd Scaler is running, follow its logs using:

./follow-logs.sh

To watch pod counts for Nodevoto as well as the bots, run:

./watch-pods.sh

To get stats on a run use:

python watcher/src/calc_responsiveness.py some/path/to/a/run.json

Samples

See the runs folder

Nodevoto

Nodevoto can be found here. Nodevoto is used as a test application for Linkerd Scaler to effect. Web, Voting, and Gif are placed in the nodevoto namespace. VoteBots are used as traffic generators and placed in the nodevoto-bot namespace.