oomscoreadj

oomscoreadj is a per-process tunable in Linux that influences the kernel out-of-memory (OOM) killer's choice of which process to terminate when the system runs out of memory. It is exposed via the procfs file /proc/[pid]/oom_score_adj and accepts integer values in the range -1000 to 1000. The kernel computes a process "badness" score (oom_score) based on its memory usage; oom_score_adj is applied to that score (the result is clamped to the 0–1000 range) to raise or lower the process's likelihood of being selected by the OOM killer.

A value of -1000 effectively prevents the process from being chosen by the OOM killer, while positive values increase its probability of termination. Only privileged users (typically root or processes with the CAP_SYS_RESOURCE capability) can write to oom_score_adj. Reading the file is generally permitted to inspect the current setting.

oom_score_adj replaced the older oom_adj interface, providing finer-grained control and a standardized 0–1000 scoring scale. Administrators commonly adjust this setting for critical system services, containers, or daemons that should be protected from accidental termination, and many init systems (for example, systemd via the OOMScoreAdjust option) and container runtimes expose ways to set it automatically. Changing oom_score_adj influences the OOM selection heuristic but does not change other memory-management behavior and should be used with care.