peer-aware-groupcache

This is a proof of concept, showing how the Kubernetes watch API can be used to keep an up-to-date list of all peer pods.

The repo has two parts:

the peerwatch package, which is a library that knows how to monitor Kubernetes for pod changes
the peer-aware-groupcache example application, which is a toy HTTP API computing prime factors, caching the results with groupcache. It uses the peerwatch package to ensure any added or removed pods are also removed from groupcache's peer list.

TODO: document how to use peerwatch library. and/or generate godoc for it

Running

$ helm install -n peer-aware-groupcache helm-chart/
$ kubectl scale --replicas=3 deployment/peer-aware-groupcache
$ export NODE_PORT=$(kubectl get --namespace default -o jsonpath="{.spec.ports[0].nodePort}" services peer-aware-groupcache)
$ export NODE_IP=$(kubectl get nodes --namespace default -o jsonpath="{.items[0].status.addresses[0].address}")
$ for i in `seq 0 9`; do echo 1234$i; curl http://$NODE_IP:$NODE_PORT/factors?n=1234$i; done

Development

Notes to self about how to publish new versions of this.

Building image

$ docker build -t peer-aware-groupcache .
$ docker tag peer-aware-groupcache robwil/peer-aware-groupcache:1.2.2
$ docker push robwil/peer-aware-groupcache:1.2.2

Running locally

Note that running locally doesn't do anything with Kube peer awareness, since it isn't inside a Kube cluster.

$ docker build -t peer-aware-groupcache .
$ docker run -p 5000:5000 -it peer-aware-groupcache