KubeCon + CloudNativeCon North America 2017

As the footprint of our K8 clusters increased across eBay data centers, many work loads were on boarded onto these clusters. With different work loads running across different clusters, cluster operations and insights into these applications has become an interesting problem. We wanted to know the footprint of these applications across different clusters in our data centers. One of the most obvious solutions is to query against different api servers and derive insights. However as the scale increased querying api servers has become a challenge and we could not get data in real time. To solve these problems, we developed an application called kube watch. Kube watch helps collects data for various resources from different clusters and persists into data stores. By building different dashboards using this data it helped us gain insights into the clusters in real time. During this talk we shall go through the kube watch system, its architecture and the problems it has helped us to solve.