Work out oom risk estimate instantly with clear inputs, formula shown and shareable results.
The OOM killer fires when the kernel cannot reclaim enough memory, which happens once demand exceeds RAM plus swap. With overcommit_memory set to 2, the kernel refuses allocations beyond swap plus a share of RAM instead, converting a random process kill into a clean allocation failure — which is usually preferable for predictable behaviour.
OOM thresholds
physical capacity = RAM + swap; CommitLimit = swap + RAM x overcommit ratio; OOM when demand exceeds physical capacity
Yes, via oom_score_adj per process. Lowering it protects a critical database; raising it makes a cache or batch worker the preferred victim.
It delays it, at the cost of severe latency once thrashing starts. Many operators prefer a small swap so the system fails fast rather than becoming unusably slow.