A Background Job Was Crashing Our JVM Every Week - Until We Taught It to Stop

Some Java services don't fail because of traffic. They fail because background jobs don't know when to stop. A background job should never be able to take down your production system. Ours did. Every week. At SAP, we ran a multi-tenant service processing thousands of tenants. Every morning, a scheduled job kicked off - telemetry, cleanup, routine work. Nothing unusual. Until it overlapped with real traffic. Then, within minutes: Heap jumped from 60% → 94% GC went into panic mode Latency spiked 5