Find out which before you touch the limit. Plot container_memory_working_set_bytes over a few restart cycles:

• Sawtooth that plateaus below the limit, then gets killed on a spike → the limit is genuinely too low.
• Steady climb that never plateaus → a leak; raising the limit only lengthens the interval between restarts.

For Node specifically, also set the heap ceiling below the container limit:

1env:
2  - name: NODE_OPTIONS
3    value: "--max-old-space-size=384"

1env:
2  - name: NODE_OPTIONS
3    value: "--max-old-space-size=384"

Otherwise V8 happily grows past 512Mi and the kernel kills the process before the garbage collector ever feels pressure.

Worth checking requests too, not just limits. If requests are far below actual usage the scheduler overpacks the node, and you get evictions that look like OOM kills but are really node memory pressure.