TestI fails on Alpine Linux (musl libc): did not see badness 1000 or 999 in output

[jirutka]

=== RUN   TestI
    testsuite_cli_test.go:248: did not see badness 1000 or 999 in output
    testsuite_cli_test.go:249: dry-running kill_largest_process()...
        pid     2: badness   0 vm_rss       0  
        pid     3: badness   0 vm_rss       0  
        ...
        pid  2212: badness   0 vm_rss       0  
        pid  2487: badness 666 vm_rss      60 uid    0 oom_score_adj    0 "syslogd" <--- new victim
        pid  2830: badness 666 vm_rss      48   
        pid  2858: badness 666 vm_rss     792 uid    0 oom_score_adj    0 "crond" <--- new victim
        pid  2920: badness 666 vm_rss      48   
        pid  3012: badness 666 vm_rss    2504 uid  100 oom_score_adj    0 "nrpe" <--- new victim
        pid  3039: badness 666 vm_rss    1764   
        pid  3065: badness 666 vm_rss    2004   
        pid  3066: badness 666 vm_rss    1916   
        pid  3072: badness   0 
        pid  3079: badness 666 vm_rss     868   
        pid  3080: badness 666 vm_rss     828   
        pid  3081: badness 666 vm_rss     828   
        pid  3082: badness 666 vm_rss     860   
        pid  3083: badness 666 vm_rss     884   
        pid  3084: badness 666 vm_rss     844   
        pid  9030: badness 666 vm_rss    3704 uid    0 oom_score_adj    0 "sshd" <--- new victim
        pid  9032: badness 666 vm_rss    2576   
        pid  9033: badness 666 vm_rss    1464   
        pid  9075: badness   0 
        pid  9379: badness   0 
        pid 13544: badness   0 
        pid 13897: badness 1332 vm_rss       4 uid 1000 oom_score_adj  999 "sleep" <--- new victim
        pid 14174: badness   0 
        pid 14186: badness   0 
        pid 14210: badness 666 
        pid 14235: badness 666 
        pid 14372: badness 666 
        pid 14429: badness 1332 vm_rss       4   
        pid 14430: badness 666 
        selecting victim took 2.444 ms
        mem avail: 31693 of 32123 MiB (98.66%), swap free: 2047 of 2047 MiB (100.00%)

$ sleep 600 &
$ pid=$!
$ echo 888 > /proc/$pid/oom_score_adj

$ cat /proc/$pid/oom_adj
15
$ cat /proc/$pid/oom_score
1258
$ cat /proc/$pid/oom_score_adj
888

$ ./earlyoom -d
...
pid 14755: badness 1258 vm_rss       4 uid 1000 oom_score_adj  888 "sleep" <--- new victim
...

---

Environment:

• x86_64 / QEMU/KVM VM
• Linux 5.10.9
• Alpine Linux Edge
• musl 1.2.2
• gcc 10.2.1 (with downstream patches)
• earlyoom 1.6.2

[rfjakob]

Thanks for the report. Kernel 5.9 changed the default oom_score from 0 to 666, which means that the -i option does not work properly anymore.

[jirutka]

Oh, really 666? I cannot wait for brainless numerologists starting freaking out from it! 🤣

[rfjakob]

This is pretty messy. Now writing 1000 to oom_score_adj raises oom_score by 666 points (instead of 1000 as before):

$ grep . /proc/$$/oom_score*
/proc/62520/oom_score:666
/proc/62520/oom_score_adj:0

$ echo 1000 > /proc/$$/oom_score_adj
$ grep . /proc/$$/oom_score*
/proc/62520/oom_score:1333
/proc/62520/oom_score_adj:1000

I think I will drop the -i option altogether (objections?).

[hakavlad]

This is pretty messy.

It's easy to understand:

666 is 1000 * 2/3.

If oom_score_adj=300 than 300 * 2/3 (= 200) will be added to oom_score.

It's easy to fix -i option to work with new (Linux 5.9+) kernels: add oom_score_adj * 2/3 to oom_score instead of adding oom_score_adj. And describe this change in the changelog instead of removing -i option completely. Of course, with older kernels, the option will work less efficiently in this case.