The talk proposes and demonstrates a methodology for monitoring and troubleshooting the performance of a typical web application by using Elasticsearch as an event analytics engine. It discusses ways of collecting performance metrics for the transaction data (response time, error codes, URL path, data base tables, etc.), the appropriate level of details captured, sampling techniques and tips for efficiently storing the data.
It then shows how to use Elasticsearch aggregations to provide the application performance specific metrics (response time percentiles, error rates) at a global level and also segmented over multiple dimensions (service, server, URL path, etc.). It shows how to use these techniques and visualisations to perform a top-down analysis through the data in order to identify a performance issue.
During the talk, the audience learns how Packetbeat, Yahoo Boomerang, Logstash, Elasticsearch, Kibana and Bonito work together for a complete application performance management solution that is more applicable and more flexible than the commercial offerings.