This file contains information about the systems that watch for and respond to out-of-memory (OOM) events.
When the system runs out of memory and the kernel OOM thread is running, you should see a series of log messages like:
OOM: 5915.8M free (+0B) / 8072.4M total OOM: oom_lowmem(shortfall_bytes=524288) called OOM: Process mapped committed bytes: OOM: proc 1043 397M 'bin/devmgr' OOM: proc 2107 88M 'devhost:pci#1:8086:1916' OOM: proc 1297 12M 'virtual-console' OOM: proc 3496 17M 'netstack' OOM: proc 4157 170M 'flutter:userpicker_device_shell' OOM: proc 28708 353M 'flutter:armadillo_user_... (+3)' OOM: proc 31584 9M 'dart:weather_agent' OOM: proc 32093 14M 'dart:mi_dashboard.dartx' OOM: Finding a job to kill... OOM: (skip) job 57930 'story-8cf82cb9f742d9ecc77f1d449' OOM: (skip) job 37434 'story-10293ae401bc0358b3ce52d2a' OOM: *KILL* job 29254 'agent' OOM: + proc 32093 run 'dart:mi_dashboard.dartx' OOM: = 1 running procs (1 total), 0 jobs OOM: (next) job 29247 'agent' OOM: (next) job 29240 'agent' OOM: (next) job 29233 'agent'
The first line shows the current state of system memory:
OOM: 45.8M free (-12.4M) / 8072.4M total
The next section prints a list of processes that are consuming large amounts of memory, in no particular order:
OOM: Process mapped committed bytes: OOM: proc 1043 397M 'bin/devmgr' OOM: proc 2107 88M 'devhost:pci#1:8086:1916' OOM: proc 1297 12M 'virtual-console' OOM: ... ^koid ^mem
The next section shows the walk through the ranked job list, printing skipped jobs (which don't have killable process descendants), the job that will be killed, and the next jobs on the chopping block:
OOM: Finding a job to kill... OOM: (skip) job 57930 'story-8cf82cb9f742d9ecc77f1d449' OOM: (skip) job 37434 'story-10293ae401bc0358b3ce52d2a' OOM: *KILL* job 29254 'agent' OOM: + proc 32093 run 'dart:mi_dashboard.dartx' OOM: = 1 running procs (1 total), 0 jobs OOM: (next) job 29247 'agent' OOM: (next) job 29240 'agent' OOM: (next) job 29233 'agent' ^koid ^name
The *KILL*
entry will also show all process descendants of the to-be-killed job.
A kernel thread that periodically checks the amount of free memory in the system, and kills a job if the free amount is too low (below the “redline”).
Use k oom info
to see the state of the OOM thread (on the kernel console):
$ k oom info OOM info: running: true printing: false simulating lowmem: false sleep duration: 1000ms redline: 50M (52428800 bytes)
The redline, sleep duration, and auto-start values are controlled by kernel.oom.*
kernel commandline flags.
The thread can be started with k oom start
and stopped with k oom stop
.
k oom print
will toggle a flag that prints the current free and total memory every time the thread wakes up.
k oom lowmem
will trigger a false low-memory event the next time the thread wakes up, potentially killing a job.
TODO(dbort/maniscalco): Implement and document.