kube-insight is a Kubernetes observability stack, intended to integrate
several best-of-breed cluster monitoring products with the intent of creating
a comprehensive observability solution to help administrators track overall
cluster and micro-service health, and to better understand system dynamics.
This repository contains the Kubernetes manifests necessary to deploy the
kube-insight stack. As such, it merely references included docker images. The
actual code of the software components of the stack are kept in separate repos.
kube-insight is intended to provide the following key functionalities:
- topology-awareness: automatically discover the service dependency graph by detecting the communication flow between containers.
- metrics collection: collect container metrics for all deployed services.
- log collection: collect container logs for troubleshooting
- event collection: collect Kubernetes events to form a system change history to, for example, be able to determine what changes introduced a performance degradation.
- visualization: provide a single UI for visualizing the system state in a topology-centric manner, allowing drill-down to watch details (metrics, logs, events) pertaining to a certain deployment/pod/container.
- time-travel: by continusously saving state (topology, metrics, logs, events) in a backing Cassandra cluster, the system state is tracked over time to produce a system audit trail.
We believe that kube-insight can ease the life of Kubernetes cluster operators
by providing contextualized insight into the system. Todays microservice-based
architectures involve a lot of moving parts with complex inter-dependencies. The
combination of topology-awareness and tracking per-container data such as logs
and metrics should simplify the task of root-cause-analysis, by quickly allowing
the operator to zoom in on the relevant services and then being able to drill
down to troubleshoot individual pods (for example by looking at their metrics,
logs).
The project is based on modern, cloud-native components, and adds some glue-code where necessary to integrate third-party software. For more details, refer to the architecture section below.
This project is very much a work in progress. Additional manifests will be added as components are added to the stack.
For the time being there isn't a single installer script but the manifests need to be deployed one-by-one. See the deploy section for additional details.
The kube-insight observability stack is "Kubernetes-native" in that it both
monitors a Kubernetes cluster and runs in a Kubernetes cluster. The monitored
cluster can, and probably should, be a different cluster from the one where
kube-insight is deployed (at least for production settings: the times when
your cluster is experiencing problems are the times when you don't want your
monitoring to break -- so keep those error domains separate!).
TODO: architecture sketch
The manifests are separated into two categories: manifests intended to be deployed onto the monitored cluster ("agents") and manifests intended to be deployed on the insight cluster ("servers"). The server-side components receive data sent by the agents and stores them in Cassandra.
- agents: intended to be deployed on the monitored cluster to collect data and send to receivers on the insight cluster.
- servers: intended to be deployed on the insight cluster to receive (and store) data sent from the agents on the monitored cluster.
Note that agents and servers can be deployed onto the same cluster, but are recommended to run on separate clusters (at least in production).
A single installer script is not available at this time. For more details and for installation instructions, go to each manifests subfolder.
To install the components helm is required to be install on the kubernetes cluster. This makes it easy to upgrade the deployment if this repo is updated