8 Tips for Marathon Performance
summary
Eight concrete operational changes that cut outages and CPU waste: enable Marathon metrics and raise the polling to metrics interval 55s to drop metrics CPU from ~20% to ~2%; capture GC logs and analyze with GCViewer to fix heap and promotion issues; increase Akka mailbox batch size (from ~5 to ~20) when events backlog shows latency; keep ZooKeeper znodes small and raise max buffer or upgrade to Marathon 1.4+ to avoid 1MB write failures; move health checks to agents (switch to Mesos health checks) and use SSE instead of callbacks to reduce registration delays from ~7 minutes to under 40 seconds.
AI-written summary · about
chapters (6)
- 0:00Start
- 14:33UPDATE TO 1.3.13
- 20:48DO NOT USE THE EVENT BUS
- 24:52PREFER BATCHING
- 25:35SHARD YOUR MARATHON
- 26:52SUMMARY 1. Monitor - enable metrics 2. Tune JVM 3. Optimize Zookeeper
related talks
50mPerformance: Avoiding Pitfalls, Diagnosing Issues & Scaling for Growth
39mHow we learned to love the Data Center Operating System
29mGetting Started with Apache Mesos
41mHow to configure your streaming jobs like a pro
01h 05mThe art and methodology of troubleshooting
42mKeynote: The Performance Mindset
01h 02mThe performance loop—A practical guide to profiling and benchmarking
24mFrom VMs To Containers: A DevOps Journey