Work out memory requirement instantly with clear inputs, formula shown and shareable results.
Memory need is a fixed base plus a per-request allocation multiplied by concurrency, and headroom matters more than the arithmetic suggests: garbage-collected runtimes need spare space to avoid constant collection, and a container without headroom is killed by the OOM killer rather than merely slowing down. Cloud instance memory comes in powers of two, so the answer rounds up.
Memory sizing
working set = base memory + concurrency x per-request memory; required = working set x (1 + headroom); round up to the next instance size
Typically 25 to 50 percent above the live set. Below that, collection frequency rises sharply and pause times become the dominant latency source.
The kernel kills it immediately with no graceful shutdown. Unlike CPU, memory limits are not throttled — they are enforced by termination.