Marathon is a Mesos scheduler that can handle huge production installation. Using Marathon at Allegro, an eCommerce platform, Tomasz Janiszewski found ways to build upon Marathon’s assets in order to customize performance for Allegro’s needs. In this talk from MesosCon EU 2017 in Prague, Tomasz shares his eight tips for improving Marathon performance and preventing outages.
Tips include how to enable metrics support via /metrics endpoint and implementing a custom adaptor to use Graphite, Datadog, StatsD, and Prometheus for metrics support; how to best tune JVM; how to optimize Zookeeper; updating to 1.3.13; advice for using the event bus; using a custom executor; and doing batch performance. Each tip includes a summary with information about when and how it can decrease performance and how to avoid the danger of the outage.
About the speaker:
Tomasz is a software engineer passionate about distributed systems. He believes in free and open source philosophy and occasionally contributes to projects on GitHub. At Allegro he works as a Software Engineer working with Mesos and Marathon cluster.