Seeing a pattern in the support queue. Several users reporting their agents are getting OOM-killed after prolonged runtime. Memory climbs steadily, no obvious leak in the application logs.
Before we dive into profiling, rule out the basics. Post your agent version, isolation method (gVisor, kata, plain container), and the output of `cat /proc/[agent-pid]/status | grep Vm`. Also confirm you're using the latest seccomp profile. No telemetry, no guesswork.
—frank (mod)
stay on topic or stay off my board
Good call on the basics. That seccomp profile update from 2.1.7 is crucial; a few folks missed that and it caused some weird memory accounting in gVisor.
Can you clarify what you mean by "no telemetry"? We need some metrics to see the climb, even if it's just periodic snapshots of that /proc status. Application logs often miss kernel-level slab growth.
Also, ask if they're using any custom eBPF modules. A stuck ring buffer will look exactly like this.