Message ID | 20190903144512.9374-1-mhocko@kernel.org (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [RFC] mm, oom: disable dump_tasks by default | expand |
On Tue, 2019-09-03 at 16:45 +0200, Michal Hocko wrote: > From: Michal Hocko <mhocko@suse.com> > > dump_tasks has been introduced by quite some time ago fef1bdd68c81 > ("oom: add sysctl to enable task memory dump"). It's primary purpose is > to help analyse oom victim selection decision. This has been certainly > useful at times when the heuristic to chose a victim was much more > volatile. Since a63d83f427fb ("oom: badness heuristic rewrite") > situation became much more stable (mostly because the only selection > criterion is the memory usage) and reports about a wrong process to > be shot down have become effectively non-existent. Well, I still see OOM sometimes kills wrong processes like ssh, systemd processes while LTP OOM tests with staight-forward allocation patterns. I just have not had a chance to debug them fully. The situation could be worse with more complex allocations like random stress or fuzzy testing. > > dump_tasks can generate a lot of output to the kernel log. It is not > uncommon that even relative small system has hundreds of tasks running. > Generating a lot of output to the kernel log both makes the oom report > less convenient to process and also induces a higher load on the printk > subsystem which can lead to other problems (e.g. longer stalls to flush > all the data to consoles). It is only generate output for the victim process where I tested on those large NUMA machines and the output is fairly manageable. > > Therefore change the default of oom_dump_tasks to not print the task > list by default. The sysctl remains in place for anybody who might need > to get this additional information. The oom report still provides an > information about the allocation context and the state of the MM > subsystem which should be sufficient to analyse most of the oom > situations. > > Signed-off-by: Michal Hocko <mhocko@suse.com> > --- > mm/oom_kill.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/oom_kill.c b/mm/oom_kill.c > index eda2e2a0bdc6..d0353705c6e6 100644 > --- a/mm/oom_kill.c > +++ b/mm/oom_kill.c > @@ -52,7 +52,7 @@ > > int sysctl_panic_on_oom; > int sysctl_oom_kill_allocating_task; > -int sysctl_oom_dump_tasks = 1; > +int sysctl_oom_dump_tasks; > > /* > * Serializes oom killer invocations (out_of_memory()) from all contexts to
On Tue 03-09-19 11:02:46, Qian Cai wrote: > On Tue, 2019-09-03 at 16:45 +0200, Michal Hocko wrote: > > From: Michal Hocko <mhocko@suse.com> > > > > dump_tasks has been introduced by quite some time ago fef1bdd68c81 > > ("oom: add sysctl to enable task memory dump"). It's primary purpose is > > to help analyse oom victim selection decision. This has been certainly > > useful at times when the heuristic to chose a victim was much more > > volatile. Since a63d83f427fb ("oom: badness heuristic rewrite") > > situation became much more stable (mostly because the only selection > > criterion is the memory usage) and reports about a wrong process to > > be shot down have become effectively non-existent. > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > processes while LTP OOM tests with staight-forward allocation patterns. Please report those. Most cases I have seen so far just turned out to work as expected and memory hogs just used oom_score_adj or similar. > I just > have not had a chance to debug them fully. The situation could be worse with > more complex allocations like random stress or fuzzy testing. Nothing really prevents enabling the sysctl when doing OOM oriented testing. > > dump_tasks can generate a lot of output to the kernel log. It is not > > uncommon that even relative small system has hundreds of tasks running. > > Generating a lot of output to the kernel log both makes the oom report > > less convenient to process and also induces a higher load on the printk > > subsystem which can lead to other problems (e.g. longer stalls to flush > > all the data to consoles). > > It is only generate output for the victim process where I tested on those large > NUMA machines and the output is fairly manageable. The main question here is whether that information is useful by _default_ because it is certainly not free. It takes both time to crawl all processes and cpu cycles to get that information to the console because printk is not free either. So if it more of "nice to have" than necessary for oom analysis then it should be disabled by default IMHO.
On Tue, 2019-09-03 at 17:13 +0200, Michal Hocko wrote: > On Tue 03-09-19 11:02:46, Qian Cai wrote: > > On Tue, 2019-09-03 at 16:45 +0200, Michal Hocko wrote: > > > From: Michal Hocko <mhocko@suse.com> > > > > > > dump_tasks has been introduced by quite some time ago fef1bdd68c81 > > > ("oom: add sysctl to enable task memory dump"). It's primary purpose is > > > to help analyse oom victim selection decision. This has been certainly > > > useful at times when the heuristic to chose a victim was much more > > > volatile. Since a63d83f427fb ("oom: badness heuristic rewrite") > > > situation became much more stable (mostly because the only selection > > > criterion is the memory usage) and reports about a wrong process to > > > be shot down have become effectively non-existent. > > > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > > processes while LTP OOM tests with staight-forward allocation patterns. > > Please report those. Most cases I have seen so far just turned out to > work as expected and memory hogs just used oom_score_adj or similar. > > > I just > > have not had a chance to debug them fully. The situation could be worse with > > more complex allocations like random stress or fuzzy testing. > > Nothing really prevents enabling the sysctl when doing OOM oriented > testing. > > > > dump_tasks can generate a lot of output to the kernel log. It is not > > > uncommon that even relative small system has hundreds of tasks running. > > > Generating a lot of output to the kernel log both makes the oom report > > > less convenient to process and also induces a higher load on the printk > > > subsystem which can lead to other problems (e.g. longer stalls to flush > > > all the data to consoles). > > > > It is only generate output for the victim process where I tested on those > > large > > NUMA machines and the output is fairly manageable. > > The main question here is whether that information is useful by > _default_ because it is certainly not free. It takes both time to crawl > all processes and cpu cycles to get that information to the console > because printk is not free either. So if it more of "nice to have" than > necessary for oom analysis then it should be disabled by default IMHO. It also feels like more a band-aid micro-optimization with the side-effect that affecting debuggability, as there could be loads of console output anyway during a kernel OOM event including failed allocation warnings. I suppose if you want to change the default behavior, the bar is high with more data and justification.
On Tue 03-09-19 11:32:58, Qian Cai wrote: > On Tue, 2019-09-03 at 17:13 +0200, Michal Hocko wrote: > > On Tue 03-09-19 11:02:46, Qian Cai wrote: > > > On Tue, 2019-09-03 at 16:45 +0200, Michal Hocko wrote: > > > > From: Michal Hocko <mhocko@suse.com> > > > > > > > > dump_tasks has been introduced by quite some time ago fef1bdd68c81 > > > > ("oom: add sysctl to enable task memory dump"). It's primary purpose is > > > > to help analyse oom victim selection decision. This has been certainly > > > > useful at times when the heuristic to chose a victim was much more > > > > volatile. Since a63d83f427fb ("oom: badness heuristic rewrite") > > > > situation became much more stable (mostly because the only selection > > > > criterion is the memory usage) and reports about a wrong process to > > > > be shot down have become effectively non-existent. > > > > > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > > > processes while LTP OOM tests with staight-forward allocation patterns. > > > > Please report those. Most cases I have seen so far just turned out to > > work as expected and memory hogs just used oom_score_adj or similar. > > > > > I just > > > have not had a chance to debug them fully. The situation could be worse with > > > more complex allocations like random stress or fuzzy testing. > > > > Nothing really prevents enabling the sysctl when doing OOM oriented > > testing. > > > > > > dump_tasks can generate a lot of output to the kernel log. It is not > > > > uncommon that even relative small system has hundreds of tasks running. > > > > Generating a lot of output to the kernel log both makes the oom report > > > > less convenient to process and also induces a higher load on the printk > > > > subsystem which can lead to other problems (e.g. longer stalls to flush > > > > all the data to consoles). > > > > > > It is only generate output for the victim process where I tested on those > > > large > > > NUMA machines and the output is fairly manageable. > > > > The main question here is whether that information is useful by > > _default_ because it is certainly not free. It takes both time to crawl > > all processes and cpu cycles to get that information to the console > > because printk is not free either. So if it more of "nice to have" than > > necessary for oom analysis then it should be disabled by default IMHO. > > It also feels like more a band-aid micro-optimization with the side-effect that > affecting debuggability, as there could be loads of console output anyway during > a kernel OOM event including failed allocation warnings. I suppose if you want > to change the default behavior, the bar is high with more data and > justification. Any specific idea what that justification should be? Because this is not something that you could measure very easily. It is very subjective and far from black and white. And I am fully aware of that. Hence RFC. That is why we should apply some common sense and cost/benefit evaluation. The cost of an additional output should be quite clear. Now we can argue about the benefit. I argue that for an absolute majority of oom report I have seen throughout past many years the task list was the least useful information of the report. Sure I could go there and double check that the victim was selected as designed. In minority cases I could use that the task list to confirm that expected-to-be-victim had OOM_SCORE_ADJ_MIN for some reason and that was why something esle has been selected. Those configuration issues tend to be reproducible and are easier to debug with sysctl enabled. All that being said, arguments tend to weigh more towards IMO.
On Wed 04-09-19 05:52:43, Tetsuo Handa wrote: > On 2019/09/03 23:45, Michal Hocko wrote: > > It's primary purpose is > > to help analyse oom victim selection decision. > > I disagree, for I use the process list for understanding what / how many > processes are consuming what kind of memory (without crashing the system) > for anomaly detection purpose. Although we can't dump memory consumed by > e.g. file descriptors, disabling dump_tasks() loose that clue, and is > problematic for me. Does anything really prevent you from enabling this by sysctl though? Or do you claim that this is a general usage pattern and therefore the default change is not acceptable or do you want a changelog to be updated?
On Wed, 4 Sep 2019, Michal Hocko wrote: > > > It's primary purpose is > > > to help analyse oom victim selection decision. > > > > I disagree, for I use the process list for understanding what / how many > > processes are consuming what kind of memory (without crashing the system) > > for anomaly detection purpose. Although we can't dump memory consumed by > > e.g. file descriptors, disabling dump_tasks() loose that clue, and is > > problematic for me. > > Does anything really prevent you from enabling this by sysctl though? Or > do you claim that this is a general usage pattern and therefore the > default change is not acceptable or do you want a changelog to be > updated? > I think the motivation is that users don't want to need to reproduce an oom kill to figure out why: they want to be able to figure out which process had higher than normal memory usage. If oom is the normal case then they ceratinly have the ability to disable it by disabling the sysctl, but that seems like something better to opt-out of rather than need to opt-in to and reproduce.
On Thu 05-09-19 22:39:47, Tetsuo Handa wrote: [...] > There is nothing that prevents users from enabling oom_dump_tasks by sysctl. > But that requires a solution for OOM stalling problem. You can hardly remove stalling if you are not reducing the amount of output or get it into a different context. Whether the later is reasonable is another question but you are essentially losing "at the OOM event state". > Since I know how > difficult to avoid problems caused by printk() flooding, I insist that > we need "mm,oom: Defer dump_tasks() output." patch. insisting is not a way to cooperate.
On Tue, 2019-09-03 at 17:13 +0200, Michal Hocko wrote: > On Tue 03-09-19 11:02:46, Qian Cai wrote: > > On Tue, 2019-09-03 at 16:45 +0200, Michal Hocko wrote: > > > From: Michal Hocko <mhocko@suse.com> > > > > > > dump_tasks has been introduced by quite some time ago fef1bdd68c81 > > > ("oom: add sysctl to enable task memory dump"). It's primary purpose is > > > to help analyse oom victim selection decision. This has been certainly > > > useful at times when the heuristic to chose a victim was much more > > > volatile. Since a63d83f427fb ("oom: badness heuristic rewrite") > > > situation became much more stable (mostly because the only selection > > > criterion is the memory usage) and reports about a wrong process to > > > be shot down have become effectively non-existent. > > > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > > processes while LTP OOM tests with staight-forward allocation patterns. > > Please report those. Most cases I have seen so far just turned out to > work as expected and memory hogs just used oom_score_adj or similar. Here is the one where oom01 should be one to be killed. [92598.855697][ T2588] Swap cache stats: add 105240923, delete 105250445, find 42196/101577 [92598.893970][ T2588] Free swap = 16383612kB [92598.913482][ T2588] Total swap = 16465916kB [92598.932938][ T2588] 7275091 pages RAM [92598.950212][ T2588] 0 pages HighMem/MovableOnly [92598.971539][ T2588] 1315554 pages reserved [92598.990698][ T2588] 16384 pages cma reserved [92599.010760][ T2588] Tasks state (memory values in pages): [92599.036265][ T2588] [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name [92599.080129][ T2588] [ 1662] 0 1662 29511 1034 290816 244 0 systemd- journal [92599.126163][ T2588] [ 2586] 998 2586 508086 0 368640 1838 0 polkitd [92599.168706][ T2588] [ 2587] 0 2587 52786 0 421888 500 0 sssd [92599.210082][ T2588] [ 2588] 0 2588 31223 0 139264 195 0 irqbalance [92599.255606][ T2588] [ 2589] 81 2589 18381 0 167936 217 -900 dbus- daemon [92599.303678][ T2588] [ 2590] 0 2590 97260 193 372736 573 0 NetworkManager [92599.348957][ T2588] [ 2594] 0 2594 95350 1 229376 758 0 rngd [92599.390216][ T2588] [ 2598] 995 2598 7364 0 94208 103 0 chronyd [92599.432447][ T2588] [ 2629] 0 2629 106234 399 442368 3836 0 tuned [92599.473950][ T2588] [ 2638] 0 2638 23604 0 212992 240 -1000 sshd [92599.515158][ T2588] [ 2642] 0 2642 10392 0 102400 138 0 rhsmcertd [92599.560435][ T2588] [ 2691] 0 2691 21877 0 208896 277 0 systemd- logind [92599.605035][ T2588] [ 2700] 0 2700 3916 0 69632 45 0 agetty [92599.646750][ T2588] [ 2705] 0 2705 23370 0 225280 393 0 systemd [92599.688063][ T2588] [ 2730] 0 2730 37063 0 294912 667 0 (sd-pam) [92599.729028][ T2588] [ 2922] 0 2922 9020 0 98304 232 0 crond [92599.769130][ T2588] [ 3036] 0 3036 37797 1 307200 305 0 sshd [92599.813768][ T2588] [ 3057] 0 3057 37797 0 303104 335 0 sshd [92599.853450][ T2588] [ 3065] 0 3065 6343 1 86016 163 0 bash [92599.892899][ T2588] [ 38249] 0 38249 58330 293 221184 246 0 rsyslogd [92599.934457][ T2588] [ 11329] 0 11329 55131 73 454656 396 0 sssd_nss [92599.976240][ T2588] [ 11331] 0 11331 54424 1 434176 610 0 sssd_be [92600.017106][ T2588] [ 25247] 0 25247 25746 1 212992 300 -1000 systemd-udevd [92600.060539][ T2588] [ 25391] 0 25391 2184 0 65536 32 0 oom01 [92600.100648][ T2588] [ 25392] 0 25392 2184 0 65536 39 0 oom01 [92600.143516][ T2588] oom- kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0- 1,global_oom,task_memcg=/system.slice/tuned.service,task=tuned,pid=2629,uid=0 [92600.213724][ T2588] Out of memory: Killed process 2629 (tuned) total- vm:424936kB, anon-rss:328kB, file-rss:1268kB, shmem-rss:0kB, UID:0 pgtables:442368kB oom_score_adj:0 [92600.297832][ T305] oom_reaper: reaped process 2629 (tuned), now anon- rss:0kB, file-rss:0kB, shmem-rss:0kB > > > I just > > have not had a chance to debug them fully. The situation could be worse with > > more complex allocations like random stress or fuzzy testing.
On Fri, 2019-09-06 at 05:59 +0900, Tetsuo Handa wrote: > On 2019/09/06 1:10, Qian Cai wrote: > > On Tue, 2019-09-03 at 17:13 +0200, Michal Hocko wrote: > > > On Tue 03-09-19 11:02:46, Qian Cai wrote: > > > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > > > > processes while LTP OOM tests with staight-forward allocation patterns. > > > > > > Please report those. Most cases I have seen so far just turned out to > > > work as expected and memory hogs just used oom_score_adj or similar. > > > > Here is the one where oom01 should be one to be killed. > > I assume that there are previous OOM killer events before > > > > > [92598.855697][ T2588] Swap cache stats: add 105240923, delete 105250445, find > > 42196/101577 > > line. Please be sure to include. Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=1048576kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=2048kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: 104 total pagecache pages Sep 5 12:00:52 hp-xl420gen9-01 kernel: 72 pages in swap cache Sep 5 12:00:52 hp-xl420gen9-01 kernel: Swap cache stats: add 105228915, delete 105238918, find 41766/100491 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Free swap = 16382644kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Total swap = 16465916kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: 7275091 pages RAM Sep 5 12:00:52 hp-xl420gen9-01 kernel: 0 pages HighMem/MovableOnly Sep 5 12:00:52 hp-xl420gen9-01 kernel: 1315554 pages reserved Sep 5 12:00:52 hp-xl420gen9-01 kernel: 16384 pages cma reserved Sep 5 12:00:52 hp-xl420gen9-01 kernel: Tasks state (memory values in pages): Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 1662] 0 1662 29511 0 290816 296 0 systemd- journal Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2586] 998 2586 508086 0 368640 1838 0 polkitd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2587] 0 2587 52786 0 421888 500 0 sssd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2588] 0 2588 31223 0 139264 207 0 irqbalance Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2589] 81 2589 18381 0 167936 217 -900 dbus- daemon Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2590] 0 2590 97260 0 372736 621 0 NetworkManager Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2594] 0 2594 95350 1 229376 758 0 rngd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2598] 995 2598 7364 0 94208 102 0 chronyd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2629] 0 2629 106234 0 442368 3959 0 tuned Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2638] 0 2638 23604 0 212992 240 -1000 sshd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2642] 0 2642 10392 0 102400 138 0 rhsmcertd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2691] 0 2691 21877 0 208896 277 0 systemd- logind Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2700] 0 2700 3916 0 69632 45 0 agetty Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2705] 0 2705 23370 0 225280 393 0 systemd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2730] 0 2730 37063 0 294912 667 0 (sd-pam) Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2922] 0 2922 9020 0 98304 232 0 crond Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 3036] 0 3036 37797 1 307200 305 0 sshd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 3057] 0 3057 37797 0 303104 335 0 sshd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 3065] 0 3065 6343 1 86016 163 0 bash Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 38249] 0 38249 58330 0 221184 293 0 rsyslogd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 11329] 0 11329 55131 0 454656 427 0 sssd_nss Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 11331] 0 11331 54424 0 434176 637 0 sssd_be Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 25247] 0 25247 25746 1 212992 300 -1000 systemd-udevd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 25391] 0 25391 2184 0 65536 32 0 oom01 25392 2184 0 65536 39 0 oom01 Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 25507] 0 25507 1581195 1411594 11395072 48 0 oom01 Sep 5 12:00:52 hp-xl420gen9-01 kernel: oom- kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0- 1,global_oom,task_memcg=/user.slice,task=oom01,pid=25507,uid=0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Out of memory: Killed process 25507 (oom01) total-vm:6324780kB, anon-rss:5647168kB, file-rss:0kB, shmem-rss:0kB, UID:0 pgtables:11395072kB oom_score_adj:0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: oom_reaper: reaped process 25507 (oom01), now anon-rss:5647452kB, file-rss:0kB, shmem-rss:0kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: irqbalance invoked oom-killer: gfp_mask=0x100cca(GFP_HIGHUSER_MOVABLE), order=0, oom_score_adj=0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: CPU: 40 PID: 2588 Comm: irqbalance Not tainted 5.3.0-rc7-next-20190904+ #5 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Hardware name: HP ProLiant XL420 Gen9/ProLiant XL420 Gen9, BIOS U19 12/27/2015 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Call Trace: Sep 5 12:00:52 hp-xl420gen9-01 kernel: dump_stack+0x62/0x9a Sep 5 12:00:52 hp-xl420gen9-01 kernel: dump_header+0xf4/0x610 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? ___ratelimit+0x4a/0x1a0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? find_lock_task_mm+0x110/0x110 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? ___ratelimit+0xfa/0x1a0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: oom_kill_process+0x136/0x1b0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: out_of_memory+0x1fb/0x960 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? oom_killer_disable+0x230/0x230 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? mutex_trylock+0x17d/0x1a0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: __alloc_pages_nodemask+0x1475/0x1bb0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? gfp_pfmemalloc_allowed+0xc0/0xc0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __swp_swapcount+0xbf/0x160 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? rwlock_bug.part.0+0x60/0x60 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __swp_swapcount+0x14a/0x160 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __swp_swapcount+0x109/0x160 Sep 5 12:00:52 hp-xl420gen9-01 kernel: alloc_pages_vma+0x18f/0x200 Sep 5 12:00:52 hp-xl420gen9-01 kernel: __read_swap_cache_async+0x3ba/0x790 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? lookup_swap_cache+0x3c0/0x3c0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: read_swap_cache_async+0x69/0xd0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __read_swap_cache_async+0x790/0x790 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? si_swapinfo+0xc0/0x150 Sep 5 12:00:52 hp-xl420gen9-01 kernel: swap_cluster_readahead+0x2a4/0x640 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? read_swap_cache_async+0xd0/0xd0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __kasan_check_read+0x11/0x20 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? check_chain_key+0x1df/0x2e0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? lookup_swap_cache+0xd3/0x3c0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: swapin_readahead+0xb9/0x83a Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? exit_swap_address_space+0x160/0x160 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? lookup_swap_cache+0x124/0x3c0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? swp_swap_info+0x8e/0xe0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? swapcache_prepare+0x20/0x20 Sep 5 12:00:52 hp-xl420gen9-01 kernel: do_swap_page+0x64e/0x1410 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? lock_downgrade+0x390/0x390 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? unmap_mapping_range+0x30/0x30 Sep 5 12:00:52 hp-xl420gen9-01 kernel: __handle_mm_fault+0xe0c/0x1a50 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? vmf_insert_mixed_mkwrite+0x20/0x20 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __kasan_check_read+0x11/0x20 Sep 5 12:00:52 hp-xl420gen9-01 kernel: ? __count_memcg_events+0x56/0x1d0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: handle_mm_fault+0x17f/0x37e Sep 5 12:00:52 hp-xl420gen9-01 kernel: __do_page_fault+0x369/0x630 Sep 5 12:00:52 hp-xl420gen9-01 kernel: do_page_fault+0x50/0x2d3 Sep 5 12:00:52 hp-xl420gen9-01 kernel: page_fault+0x2f/0x40 Sep 5 12:00:52 hp-xl420gen9-01 kernel: RIP: 0033:0x7f89fb1a4b14 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Code: Bad RIP value. Sep 5 12:00:52 hp-xl420gen9-01 kernel: RSP: 002b:00007ffcf7a88148 EFLAGS: 00010206 Sep 5 12:00:52 hp-xl420gen9-01 kernel: RAX: 0000000000000000 RBX: 0000556844dd5db0 RCX: 00007f89fa4bf211 Sep 5 12:00:52 hp-xl420gen9-01 kernel: RDX: 00000000000026f3 RSI: 0000000000000002 RDI: 0000000000000000 Sep 5 12:00:52 hp-xl420gen9-01 kernel: RBP: 0000000000000002 R08: 0000000000000000 R09: 0000556844dc37e0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: R10: 0000556844dc4f20 R11: 0000000000000000 R12: 0000556844d9c720 Sep 5 12:00:52 hp-xl420gen9-01 kernel: R13: 0000000000000000 R14: 00000000000026f3 R15: 0000000000000002 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Mem-Info: Sep 5 12:00:52 hp-xl420gen9-01 kernel: active_anon:112189 inactive_anon:307421 isolated_anon:0#012 active_file:0 inactive_file:1797 isolated_file:0#012 unevictable:984241 dirty:0 writeback:511 unstable:0#012 slab_reclaimable:14317 slab_unreclaimable:498931#012 mapped:27 shmem:0 pagetables:4114 bounce:0#012 free:36874 free_pcp:549 free_cma:0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 active_anon:2328kB inactive_anon:434856kB active_file:0kB inactive_file:20kB unevictable:414264kB isolated(anon):0kB isolated(file):0kB mapped:20kB dirty:0kB writeback:0kB shmem:0kB shmem_thp: 0kB shmem_pmdmapped: 0kB anon_thp: 0kB writeback_tmp:0kB unstable:0kB all_unreclaimable? yes Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 active_anon:448804kB inactive_anon:1107116kB active_file:40kB inactive_file:8440kB unevictable:3200440kB isolated(anon):0kB isolated(file):0kB mapped:2076kB dirty:0kB writeback:2304kB shmem:0kB shmem_thp: 0kB shmem_pmdmapped: 0kB anon_thp: 2353152kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 DMA free:15864kB min:12kB low:24kB high:36kB active_anon:0kB inactive_anon:0kB active_file:0kB inactive_file:0kB unevictable:0kB writepending:0kB present:15948kB managed:15864kB mlocked:0kB kernel_stack:0kB pagetables:0kB bounce:0kB free_pcp:0kB local_pcp:0kB free_cma:0kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: lowmem_reserve[]: 0 1273 7711 7711 7711 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 DMA32 free:26876kB min:1064kB low:2368kB high:3672kB active_anon:0kB inactive_anon:246988kB active_file:4kB inactive_file:4kB unevictable:172kB writepending:0kB present:1821440kB managed:1304560kB mlocked:184kB kernel_stack:288kB pagetables:480kB bounce:0kB free_pcp:28kB local_pcp:0kB free_cma:0kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: lowmem_reserve[]: 0 0 6437 6437 6437 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 Normal free:5792kB min:5392kB low:11980kB high:18568kB active_anon:2572kB inactive_anon:601564kB active_file:0kB inactive_file:4kB unevictable:244kB writepending:0kB present:10485760kB managed:6591660kB mlocked:4kB kernel_stack:7616kB pagetables:3304kB bounce:0kB free_pcp:292kB local_pcp:0kB free_cma:0kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: lowmem_reserve[]: 0 0 0 0 0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 Normal free:94492kB min:13024kB low:28948kB high:44872kB active_anon:450828kB inactive_anon:4304996kB active_file:52kB inactive_file:10544kB unevictable:268kB writepending:12kB present:16777216kB managed:15926064kB mlocked:60kB kernel_stack:7744kB pagetables:12672kB bounce:0kB free_pcp:6376kB local_pcp:0kB free_cma:0kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: lowmem_reserve[]: 0 0 0 0 0 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 DMA: 0*4kB 1*8kB (U) 1*16kB (U) 1*32kB (U) 1*64kB (U) 1*128kB (U) 1*256kB (U) 0*512kB 1*1024kB (U) 1*2048kB (M) 3*4096kB (M) = 15864kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 DMA32: 13*4kB (U) 17*8kB (U) 584*16kB (UMEH) 202*32kB (UMEH) 96*64kB (UMEH) 21*128kB (UMEH) 4*256kB (UM) 0*512kB 1*1024kB (H) 0*2048kB 0*4096kB = 26876kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 Normal: 0*4kB 0*8kB 18*16kB (UME) 176*32kB (UM) 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 5920kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 Normal: 0*4kB 0*8kB 10*16kB (UMEH) 10*32kB (MH) 7*64kB (UMEH) 600*128kB (M) 2082*256kB (UM) 0*512kB 0*1024kB 0*2048kB 0*4096kB = 610720kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=1048576kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 0 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=2048kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=1048576kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Node 1 hugepages_total=0 hugepages_free=0 hugepages_surp=0 hugepages_size=2048kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: 3936 total pagecache pages Sep 5 12:00:52 hp-xl420gen9-01 kernel: 541 pages in swap cache Sep 5 12:00:52 hp-xl420gen9-01 kernel: Swap cache stats: add 105240923, delete 105250445, find 42196/101577 Sep 5 12:00:52 hp-xl420gen9-01 kernel: Free swap = 16383612kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: Total swap = 16465916kB Sep 5 12:00:52 hp-xl420gen9-01 kernel: 7275091 pages RAM Sep 5 12:00:52 hp-xl420gen9-01 kernel: 0 pages HighMem/MovableOnly Sep 5 12:00:52 hp-xl420gen9-01 kernel: 1315554 pages reserved Sep 5 12:00:52 hp-xl420gen9-01 kernel: 16384 pages cma reserved Sep 5 12:00:52 hp-xl420gen9-01 kernel: Tasks state (memory values in pages): Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 1662] 0 1662 29511 1034 290816 244 0 systemd- journal Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2586] 998 2586 508086 0 368640 1838 0 polkitd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2587] 0 2587 52786 0 421888 500 0 sssd Sep 5 12:00:52 hp-xl420gen9-01 kernel: [ 2588] 0 2588 31223 0 139264 195 0 irqbalance Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2589] 81 2589 18381 0 167936 217 -900 dbus- daemon Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2590] 0 2590 97260 193 372736 573 0 NetworkManager Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2594] 0 2594 95350 1 229376 758 0 rngd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2598] 995 2598 7364 0 94208 103 0 chronyd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2629] 0 2629 106234 399 442368 3836 0 tuned Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2638] 0 2638 23604 0 212992 240 -1000 sshd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2642] 0 2642 10392 0 102400 138 0 rhsmcertd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2691] 0 2691 21877 0 208896 277 0 systemd- logind Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2700] 0 2700 3916 0 69632 45 0 agetty Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2705] 0 2705 23370 0 225280 393 0 systemd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2730] 0 2730 37063 0 294912 667 0 (sd-pam) Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 2922] 0 2922 9020 0 98304 232 0 crond Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 3036] 0 3036 37797 1 307200 305 0 sshd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 3057] 0 3057 37797 0 303104 335 0 sshd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 3065] 0 3065 6343 1 86016 163 0 bash Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 38249] 0 38249 58330 293 221184 246 0 rsyslogd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 11329] 0 11329 55131 73 454656 396 0 sssd_nss Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 11331] 0 11331 54424 1 434176 610 0 sssd_be Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 25247] 0 25247 25746 1 212992 300 -1000 systemd-udevd Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 25391] 0 25391 2184 0 65536 32 0 oom01 Sep 5 12:00:53 hp-xl420gen9-01 kernel: [ 25392] 0 25392 2184 0 65536 39 0 oom01 Sep 5 12:00:53 hp-xl420gen9-01 kernel: oom- kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0- 1,global_oom,task_memcg=/system.slice/tuned.service,task=tuned,pid=2629,uid=0 Sep 5 12:00:54 hp-xl420gen9-01 kernel: Out of memory: Killed process 2629 (tuned) total-vm:424936kB, anon-rss:328kB, file-rss:1268kB, shmem-rss:0kB, UID:0 pgtables:442368kB oom_score_adj:0 Sep 5 12:00:54 hp-xl420gen9-01 kernel: oom_reaper: reaped process 2629 (tuned), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB Sep 5 12:00:54 hp-xl420gen9-01 systemd[1]: tuned.service: Main process exited, code=killed, status=9/KILL Sep 5 12:00:54 hp-xl420gen9-01 systemd[1]: tuned.service: Failed with result 'signal'. Sep 5 12:06:32 hp-xl420gen9-01 kernel: oom01 invoked oom-killer: gfp_mask=0x400dc0(GFP_KERNEL_ACCOUNT|__GFP_ZERO), order=0, oom_score_adj=0 Sep 5 12:10:06 hp-xl420gen9-01 kernel: CPU: 35 PID: 25560 Comm: oom01 Not tainted 5.3.0-rc7-next-20190904+ #5 Sep 5 12:10:06 hp-xl420gen9-01 kernel: Hardware name: HP ProLiant XL420 Gen9/ProLiant XL420 Gen9, BIOS U19 12/27/2015 > > Any chance it is https://lkml.org/lkml/2017/7/28/317 ? I am not sure.
On Fri 06-09-19 19:46:10, Tetsuo Handa wrote: > On 2019/09/05 23:08, Michal Hocko wrote: > > On Thu 05-09-19 22:39:47, Tetsuo Handa wrote: > > [...] > >> There is nothing that prevents users from enabling oom_dump_tasks by sysctl. > >> But that requires a solution for OOM stalling problem. > > > > You can hardly remove stalling if you are not reducing the amount of > > output or get it into a different context. Whether the later is > > reasonable is another question but you are essentially losing "at the > > OOM event state". > > > > I am not losing "at the OOM event state". Please find "struct oom_task_info" > (for now) embedded into "struct task_struct" which holds "at the OOM event state". > > And my patch moves "printk() from dump_tasks()" from OOM context to WQ context. Workers might be blocked for unbound amount of time and so this information might be printed late.
On Fri, 2019-09-06 at 19:32 +0900, Tetsuo Handa wrote: > On 2019/09/06 6:21, Qian Cai wrote: > > On Fri, 2019-09-06 at 05:59 +0900, Tetsuo Handa wrote: > > > On 2019/09/06 1:10, Qian Cai wrote: > > > > On Tue, 2019-09-03 at 17:13 +0200, Michal Hocko wrote: > > > > > On Tue 03-09-19 11:02:46, Qian Cai wrote: > > > > > > Well, I still see OOM sometimes kills wrong processes like ssh, systemd > > > > > > processes while LTP OOM tests with staight-forward allocation patterns. > > > > > > > > > > Please report those. Most cases I have seen so far just turned out to > > > > > work as expected and memory hogs just used oom_score_adj or similar. > > > > > > > > Here is the one where oom01 should be one to be killed. > > > > > > I assume that there are previous OOM killer events before > > > > > > > > > > > [92598.855697][ T2588] Swap cache stats: add 105240923, delete 105250445, find > > > > 42196/101577 > > > > > > line. Please be sure to include. > > > > 12:00:52 oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0-1,global_oom,task_memcg=/user.slice,task=oom01,pid=25507,uid=0 > > 12:00:52 Out of memory: Killed process 25507(oom01) total-vm:6324780kB, anon-rss:5647168kB, file-rss:0kB, shmem-rss:0kB,UID:0 pgtables:11395072kB oom_score_adj:0 > > 12:00:52 oom_reaper: reaped process 25507(oom01), now anon-rss:5647452kB, file-rss:0kB, shmem-rss:0kB > > 12:00:52 irqbalance invoked oom-killer: gfp_mask=0x100cca(GFP_HIGHUSER_MOVABLE), order=0, oom_score_adj=0 > > (...snipped...) > > 12:00:53 [ 25391] 0 25391 2184 0 65536 32 0 oom01 > > 12:00:53 [ 25392] 0 25392 2184 0 65536 39 0 oom01 > > 12:00:53 oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0-1,global_oom,task_memcg=/system.slice/tuned.service,task=tuned,pid=2629,uid=0 > > 12:00:54 Out of memory: Killed process 2629(tuned) total-vm:424936kB, anon-rss:328kB, file-rss:1268kB, shmem-rss:0kB, UID:0 pgtables:442368kB oom_score_adj:0 > > 12:00:54 oom_reaper: reaped process 2629 (tuned), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB > > OK. anon-rss did not decrease when oom_reaper gave up. > I think this is same with https://lkml.org/lkml/2017/7/28/317 case. > > The OOM killer does not wait for OOM victims until existing OOM victims release > memory by calling exit_mmap(). The OOM killer selects next OOM victim as soon as > the OOM reaper sets MMF_OOM_SKIP. As a result, when the OOM reaper failed to > reclaim memory due to e.g. mlocked pages, the OOM killer immediately selects next > OOM victim. But since 25391 and 25392 are consuming little memory (maybe these are > already reaped OOM victims), neither 25391 nor 25392 was selected as next OOM victim. > Yes, mlocked is troublesome. I have other incidents where crond and systemd- udevd were killed by mistake, and it even tried to kill kworker/0. https://cailca.github.io/files/dmesg.txt
On Fri, 2019-09-06 at 22:41 +0900, Tetsuo Handa wrote: > On 2019/09/06 22:08, Qian Cai wrote: > > Yes, mlocked is troublesome. I have other incidents where crond and systemd- > > udevd were killed by mistake, > > Yes. How to mitigate this regression is a controversial topic. > Michal thinks that we should make mlocked pages reclaimable, but > we haven't reached there. I think that we can monitor whether > counters decay over time (with timeout), but Michal refuses any > timeout based approach. We are deadlocked there. > > > and it even tried to kill kworker/0. > > No. The OOM killer never tries to kill kernel threads like kworker/0. I meant, [40040.401575] kworker/0:1 invoked oom-killer: gfp_mask=0x40cc0(GFP_KERNEL|__GFP_COMP), order=2, oom_score_adj=0 > > > > > https://cailca.github.io/files/dmesg.txt > >
On Fri 06-09-19 22:41:39, Tetsuo Handa wrote: > On 2019/09/06 22:08, Qian Cai wrote: > > Yes, mlocked is troublesome. I have other incidents where crond and systemd- > > udevd were killed by mistake, > > Yes. How to mitigate this regression is a controversial topic. > Michal thinks that we should make mlocked pages reclaimable, but > we haven't reached there. I think that we can monitor whether > counters decay over time (with timeout), but Michal refuses any > timeout based approach. We are deadlocked there. There is really nothing controversial here. It just needs somebody to do the actual work. Besides that I haven't seen any real workload to mlock so much memory that there won't be anything reapable. LTP oom test simply doesn't represent any real world workload. So while not optimal I do not really think this is something to lose sleep over.
diff --git a/mm/oom_kill.c b/mm/oom_kill.c index eda2e2a0bdc6..d0353705c6e6 100644 --- a/mm/oom_kill.c +++ b/mm/oom_kill.c @@ -52,7 +52,7 @@ int sysctl_panic_on_oom; int sysctl_oom_kill_allocating_task; -int sysctl_oom_dump_tasks = 1; +int sysctl_oom_dump_tasks; /* * Serializes oom killer invocations (out_of_memory()) from all contexts to