[Bug 280846] Low memory freezes / OOM: a thread waited too long to allocate a page

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=280846

--- Comment #61 from Mark Millard <marklmi26-fbsd@yahoo.com> ---
(In reply to Mark Millard from comment #60)

It got to the failure point:

vm_fault_allocate_oom: proc 1526 (firefox) failed to alloc page on fault,
starting OOM
vm_fault_allocate_oom: proc 1526 (firefox) failed to alloc page on fault,
starting OOM
vm_fault_allocate_oom: proc 1526 (firefox) failed to alloc page on fault,
starting OOM
vm_fault_allocate_oom: proc 1512 (firefox) failed to alloc page on fault,
starting OOM
Oct 25 13:43:07 aarch64-main-pkgs kernel: pid 1525 (firefox), jid 0, uid 0, was
killed: a thread waited too long to allocate a page

That was about 4 hrs ago as I type this. Now:

Mem: 122832Ki Active, 438552Ki Inact, . . ., 1240Mi Wired, . . ., 4218Mi Free,
4842Mi MaxObsActive, 1304Mi MaxObsWired, 6815Mi MaxObs(Act+Wir+Lndry)

The largest RES processes are:

(Note: 1512 was not one of the processes killed. 1524 and 1526
are still running as well. 1524, 1525, and 1526 look to have been
for the 3 Home Depot tabs, 1525 having been killed.)

 1512     0 root         77  20    0   4397Mi    1062Mi select   0 170:55  
2.19% firefox
 1526     0 root         29  28    0   3218Mi  399560Ki select   0 543:37  
8.57% /usr/local/lib/firefox/firefox -contentproc
{b91e352c-11e4-473a-bb58-eb4826e54ecf} 1512 8 tab
 1520     0 root         21  20    0   2836Mi  375868Ki select   0   1:23  
0.00% /usr/local/lib/firefox/firefox -contentproc
{d3bc151f-c6c2-4e85-8d13-1f3eda86142f} 1512 2 tab
 1524     0 root         29  20    0   3057Mi  320676Ki select   0  44:52  
1.18% /usr/local/lib/firefox/firefox -contentproc
{075167db-4164-41b8-b7df-4f99b4610ef9} 1512 6 tab

# sysctl vm.domain | grep stat | sort
you have mail
vm.domain.0.stats.active: 31862
vm.domain.0.stats.actpdpgs: 1904670576
vm.domain.0.stats.free_count: 1078502
vm.domain.0.stats.free_min: 12880
vm.domain.0.stats.free_reserved: 2713
vm.domain.0.stats.free_severe: 7796
vm.domain.0.stats.free_target: 43381
vm.domain.0.stats.inactive: 109856
vm.domain.0.stats.inactive_pps: 426537
vm.domain.0.stats.inactive_target: 65071
vm.domain.0.stats.inactpdpgs: 38305445
vm.domain.0.stats.laundpdpgs: 6964714
vm.domain.0.stats.laundry: 1082
vm.domain.0.stats.unswappable: 493803
vm.domain.0.stats.unswppdpgs: 0

Note the size of: vm.domain.0.stats.actpdpgs

# sysctl vm.stats.vm.v_laundry_count
vm.stats.vm.v_laundry_count: 1082

One of the 3 Home Depot tabs reports that the page crashed.
The other 2 are still operable in the firefox session. The
OOM did not leave behind a *.core file (as expected).

Sure looks to me like a modified-memory leak during the
home depot web-page-activity-handling (when the user is
idle on the system).

I've no clue what type of data was/is in the accumulated
leak pages, limiting what conclusions I can make.

-- 
You are receiving this mail because:
You are the assignee for the bug.