The Box Keeps OOM-Killing. Just Add a Huge Swapfile?
A latency-sensitive service keeps getting OOM-killed and a teammate wants to add 16GB of swap and crank swappiness. Defend what you actually do instead.
Linux
Observabilitythe decision you defend
A latency-sensitive service keeps getting OOM-killed on its VM, and a teammate says just add 16GB of swap and crank vm.swappiness so the kernel stops killing it. What do you do, and why?
the situation
A low-latency service running on its own VM keeps getting terminated by the kernel OOM killer under load. Each kill lines up with a spike of 5xx errors while the process restarts, so users see periodic failures during busy periods. dmesg shows the familiar "Out of memory: Killed process" lines, and the service's memory footprint is clearly climbing toward the ceiling before each event.
context
The VM has no swap configured and a fixed amount of RAM. A teammate wants to add a 16GB swapfile and raise vm.swappiness so the kernel pages the process out to disk instead of killing it, arguing that this will stop the kills and end the 5xx spikes. You have not yet determined whether the growth is a genuine working-set that exceeds RAM, a memory leak, or a transient load spike.
How this challenge works
Take a position on the decision above and defend it. A senior-engineer AI will push back over up to 4 rounds. When you are done, you are scored against a verified rubric so you can see exactly what a complete answer covers - these are learning prompts, not gotchas.