3

I am experiencing a weird issue lately:

Sometimes (I cannot reproduce it on purpose), my system is using all its swap, despite there being more than enough free RAM. If this happens, the systems then becomes unresponsive for a couple of minutes, then the OOM killer kills either a "random" process which does not help much, or the X server. If it kills a "random" process, the system does not become responsive (there is still no swap but much free RAM); if it kills X, the swap is freed and the system becomes responsive again.

Output of free when it happens:

$ free -htl
              total        used        free      shared  buff/cache   available
Mem:           7.6G        1.4G         60M        5.7G        6.1G        257M
Low:           7.6G        7.5G         60M
High:            0B          0B          0B
Swap:          3.9G        3.9G          0B
Total:          11G        5.4G         60M

uname -a:

Linux fedora 4.4.7-300.fc23.x86_64 #1 SMP Wed Apr 13 02:52:52 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

Swapiness:

cat /proc/sys/vm/swappiness 
5

Relevant section in dmesg: http://pastebin.com/0P0TLfsC

tmpfs:

$ df -h -t tmpfs
Filesystem      Size  Used Avail Use% Mounted on
tmpfs           3.8G  1.5M  3.8G   1% /dev/shm
tmpfs           3.8G  1.7M  3.8G   1% /run
tmpfs           3.8G     0  3.8G   0% /sys/fs/cgroup
tmpfs           3.8G  452K  3.8G   1% /tmp
tmpfs           776M   16K  776M   1% /run/user/42
tmpfs           776M   32K  776M   1% /run/user/1000

Meminfo: http://pastebin.com/CRmitCiJ

top -o SHR -n 1
Tasks: 231 total,   1 running, 230 sleeping,   0 stopped,   0 zombie
%Cpu(s):  8.5 us,  3.0 sy,  0.3 ni, 86.9 id,  1.3 wa,  0.0 hi,  0.0 si,  0.0 st
KiB Mem :  7943020 total,   485368 free,   971096 used,  6486556 buff/cache
KiB Swap:  4095996 total,  1698992 free,  2397004 used.   989768 avail Mem 

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND                                             
 2066 mkamlei+  20   0 8342764 163908 145208 S   0.0  2.1   0:59.62 Xorg                                                
 2306 mkamlei+  20   0 1892816 138536  27168 S   0.0  1.7   1:25.47 gnome-shell                                         
 3118 mkamlei+  20   0  596392  21084  13152 S   0.0  0.3   0:04.86 gnome-terminal-                                     
 1646 gdm       20   0 1502632  60324  12976 S   0.0  0.8   0:01.91 gnome-shell                                         
 2269 mkamlei+  20   0 1322592  22440   8124 S   0.0  0.3   0:00.87 gnome-settings-                                     
  486 root      20   0   47048   8352   7656 S   0.0  0.1   0:00.80 systemd-journal                                     
 2277 mkamlei+   9 -11  570512  10080   6644 S   0.0  0.1   0:15.33 pulseaudio                                          
 2581 mkamlei+  20   0  525424  19272   5796 S   0.0  0.2   0:00.37 redshift-gtk                                        
 1036 root      20   0  619016   9204   5408 S   0.0  0.1   0:01.70 NetworkManager                                      
 1599 gdm       20   0 1035672  11820   5120 S   0.0  0.1   0:00.28 gnome-settings-                                     
 2386 mkamlei+  20   0  850856  24948   4944 S   0.0  0.3   0:05.84 goa-daemon                                          
 2597 mkamlei+  20   0 1138200  13104   4596 S   0.0  0.2   0:00.28 evolution-alarm                                     
 2369 mkamlei+  20   0 1133908  16472   4560 S   0.0  0.2   0:00.49 evolution-sourc                                     
 2529 mkamlei+  20   0  780088  54080   4380 S   0.0  0.7   0:01.14 gnome-software                                      
 2821 mkamlei+  20   0 1357820  44320   4308 S   0.0  0.6   0:00.23 evolution-calen                                     
 2588 mkamlei+  20   0 1671848  55744   4300 S   0.0  0.7   0:00.49 evolution-calen                                     
 2525 mkamlei+  20   0  613512   8928   4188 S   0.0  0.1   0:00.19 abrt-applet                                         

ipcs:

[mkamleithner@fedora ~]$ ipcs -m -t

------ Shared Memory Attach/Detach/Change Times --------
shmid      owner      attached             detached             changed             
294912     mkamleithn Apr 30 20:29:16      Not set              Apr 30 20:29:16     
393217     mkamleithn Apr 30 20:29:19      Apr 30 20:29:19      Apr 30 20:29:17     
491522     mkamleithn Apr 30 20:42:21      Apr 30 20:42:21      Apr 30 20:29:18     
524291     mkamleithn Apr 30 20:38:10      Apr 30 20:38:10      Apr 30 20:29:18     
786436     mkamleithn Apr 30 20:38:12      Not set              Apr 30 20:38:12     

[mkamleithner@fedora ~]$ ipcs

------ Message Queues --------
key        msqid      owner      perms      used-bytes   messages    

------ Shared Memory Segments --------
key        shmid      owner      perms      bytes      nattch     status      
0x00000000 294912     mkamleithn 600        524288     2          dest         
0x00000000 393217     mkamleithn 600        2576       2          dest         
0x00000000 491522     mkamleithn 600        4194304    2          dest         
0x00000000 524291     mkamleithn 600        524288     2          dest         
0x00000000 786436     mkamleithn 600        4194304    2          dest         

------ Semaphore Arrays --------
key        semid      owner      perms      nsems     

[mkamleithner@fedora ~]$ ipcs -m -t

------ Shared Memory Attach/Detach/Change Times --------
shmid      owner      attached             detached             changed             
294912     mkamleithn Apr 30 20:29:16      Not set              Apr 30 20:29:16     
393217     mkamleithn Apr 30 20:29:19      Apr 30 20:29:19      Apr 30 20:29:17     
491522     mkamleithn Apr 30 20:42:21      Apr 30 20:42:21      Apr 30 20:29:18     
524291     mkamleithn Apr 30 20:38:10      Apr 30 20:38:10      Apr 30 20:29:18     
786436     mkamleithn Apr 30 20:38:12      Not set              Apr 30 20:38:12     

[mkamleithner@fedora ~]$ sudo grep 786436 /proc/*/maps
/proc/2084/maps:7ff4a56cc000-7ff4a5acc000 rw-s 00000000 00:05 786436                     /SYSV00000000 (deleted)
/proc/3984/maps:7f4574d00000-7f4575100000 rw-s 00000000 00:05 786436                     /SYSV00000000 (deleted)
[mkamleithner@fedora ~]$ sudo grep 524291 /proc/*/maps
/proc/2084/maps:7ff4a4593000-7ff4a4613000 rw-s 00000000 00:05 524291                     /SYSV00000000 (deleted)
/proc/2321/maps:7fa9b8a67000-7fa9b8ae7000 rw-s 00000000 00:05 524291                     /SYSV00000000 (deleted)
[mkamleithner@fedora ~]$ sudo grep 491522 /proc/*/maps
/proc/2084/maps:7ff4a4ad3000-7ff4a4ed3000 rw-s 00000000 00:05 491522                     /SYSV00000000 (deleted)
/proc/2816/maps:7f2763ba1000-7f2763fa1000 rw-s 00000000 00:05 491522                     /SYSV00000000 (deleted)
[mkamleithner@fedora ~]$ sudo grep 393217 /proc/*/maps
/proc/2084/maps:7ff4b1a60000-7ff4b1a61000 rw-s 00000000 00:05 393217                     /SYSV00000000 (deleted)
/proc/2631/maps:7fb89be79000-7fb89be7a000 rw-s 00000000 00:05 393217                     /SYSV00000000 (deleted)
[mkamleithner@fedora ~]$ sudo grep 294912 /proc/*/maps
/proc/2084/maps:7ff4a5510000-7ff4a5590000 rw-s 00000000 00:05 294912                     /SYSV00000000 (deleted)
/proc/2582/maps:7f7902dd3000-7f7902e53000 rw-s 00000000 00:05 294912                     /SYSV00000000 (deleted)

getting the process names:

[mkamleithner@fedora ~]$ ps aux | grep 2084
mkamlei+  2084  5.1  2.0 8149580 159272 tty2   Sl+  20:29   1:10 /usr/libexec/Xorg vt2 -displayfd 3 -auth /run/user/1000/gdm/Xauthority -nolisten tcp -background none -noreset -keeptty -verbose 3
mkamlei+  5261  0.0  0.0 118476  2208 pts/0    S+   20:52   0:00 grep --color=auto 2084
[mkamleithner@fedora ~]$ ps aux | grep 3984
mkamlei+  3984 11.4  3.6 1355100 293240 tty2   Sl+  20:38   1:38 /usr/lib64/firefox/firefox
mkamlei+  5297  0.0  0.0 118472  2232 pts/0    S+   20:52   0:00 grep --color=auto 3984

Should I also post the results for the other shmids? I don't really know how to interpret the output.

How can I fix this?

Edit: Starting the game "Papers, Please" always seems to trigger this problem after some time. It also happens sometimes when this game is not started, though.

Edit2: Seems to be an X issue. On wayland this does not happen. Might be due to custom settings in xorg.conf.

Final Edit: For anyone experiencing the same problem: I was using DRI 2. Switching to DRI 3 also fixes the problem. this is my relevant section in the xorg.conf:

Section "Device"
    Identifier  "Intel Graphics"
    Driver      "intel"
    Option      "AccelMethod"     "sna" # 
    Option      "Backlight"       "intel_backlight"
    BusID       "PCI:0:2:0"
    Option      "DRI"             "3" #here
    Option      "TearFree"        "true"
EndSection

The relevant file on my system is in /usr/share/X11/xorg.conf.d/ .

  • You obviously have some process(es) consuming a lot of RAM. As it helps to kill X it is probably some X program(s) which also die when X gets killed. You could look with top and see which processes are using a lot of virtual memory. It is possible to sort on memory usage in top. You could avoid those programs consuming a lot of memory or you could add some more swap to your machine. – Henrik Carlqvist Apr 30 '16 at 18:15
  • One more thing that might be worth checking... If you have some tmpfs file system mounted contents below that directory could also consume your virtual memory. df -h | grep tmpfs – Henrik Carlqvist Apr 30 '16 at 18:21
  • But why is my RAM nearly empty, and my swap full? And why does the system not at least fill the RAM before it kills processes? My tmpfs only use few kB. – knaecke Apr 30 '16 at 18:29
  • 1
    Try ( as root ) echo 1 > /proc/sys/vm/drop_caches and see if that 6 gb of cache gets freed up. Also add the output of cat /proc/meminfo to your question. – psusi Apr 30 '16 at 18:44
  • 1
    Ok, we got the output of top (no big processes), the output of df (no big tmpfs). So what about sys v ipc? Could we also get the output of ipcs? Maybe you have some big shared memory. – Henrik Carlqvist Apr 30 '16 at 20:19
2

shared Memory used (mostly) by tmpfs (Shmem in /proc/meminfo, available on kernels 2.6.32, displayed as zero if not available)>

So the manpage definition of Shared is not as helpful as it could be :(. If the tmpfs use does not reflect this high value of Shared, then the value must represent some process(es) "who did mmap() with MAP_SHARED|MAP_ANONYMOUS" (or System V shared memory).

6G of shared memory on an 8G system is still a lot. Seriously, you don't want that, at least not on a desktop.

It's weird that it seems to contribute to "buff/cache" as well. But I did a quick test with python and that's just how it works.

To show the processes with the most shared memory, use top -o SHR -n 1.

System V shared memory

Finally it's possible you have some horrible legacy software that uses system V shared memory segments. If they get leaked, they won't show up in top :(.

You can list them with ipcs -m -t. Hopefully the most recently created one is still in use. Take the shmid number and e.g.

$ ipcs -m -t

------ Shared Memory Attach/Detach/Change Times --------
shmid      owner      attached             detached             changed             
3538944    alan       Apr 30 20:35:15      Apr 30 20:35:15      Apr 30 16:07:41     
3145729    alan       Apr 30 20:35:15      Apr 30 20:35:15      Apr 30 15:04:09     
4587522    alan       Apr 30 20:37:38      Not set              Apr 30 20:37:38     

# sudo grep 4587522 /proc/*/maps

-> then the numbers shown in the /proc paths are the pid of the processes that use the SHM. (So you could e.g. grep the output of ps for that pid number).

Apparent contradictions

  1. Xorg has 8G mapped. Even though you don't have separate video card RAM. It only has 150M resident. It's not that the rest is swapped out, because you don't have enough swap space.

  2. The SHM segments shown by ipcs are all attached to two processes. So none of them have leaked, and they should all show up in the SHR column of top (double-counted even). It's ok if the number of pages used is less than the size of the memory segment, that just means there are pages that haven't been used. But free says we have 6GB of allocated shared memory to account for, and we can't find that.

  • It seems that X is using a lot of virtual memory - I updated my question with the output of top. If grep the shmid's in /proc/*/maps, I always (for all entries) get something like this: /proc/2066/maps:7f8d446fa000-7f8d44afa000 rw-s 00000000 00:05 950273 /SYSV00000000 (deleted) Does that mean that the shm is already deleted? – knaecke Apr 30 '16 at 20:04
  • I think the shm must exist, or it wouldn't be listed. I see the same on my machine. I think the (deleted) is a way of saying that the filename doesn't really exist, I think it would look similar if you mmapped a file and then deleted the file (the mmapped memory would still work). – sourcejedi Apr 30 '16 at 20:16
  • 1
    I only use an intel integrated graphics card without dedicated video ram. I updated my question with the results your suggestions, but I don't really know how to interpret it, I have no idea what I am looking at... – knaecke Apr 30 '16 at 20:59
  • 2
    It just happened again, the virtual memory of Xorg grew to 10.1 GB before the oom killed it. – knaecke Apr 30 '16 at 21:17
  • 2
    So it seems to be Xorg which eats memory. The output from ipcs only showed a few megabytes of shared memory. Maybe this is a bug in your X.org, if so probably a memory leak. If you are lucky it might help to switch to another graphics driver like vesa in the Device section of your xorg.conf. However, switching graphics driver will probably give worse performance and/or lack of functionality. Maybe others have also seen this problem? Maybe there are some patches/updates available for your distribution that will fix the problem. – Henrik Carlqvist Apr 30 '16 at 22:56
2

shared Memory used (mostly) by tmpfs (Shmem in /proc/meminfo, available on kernels 2.6.32, displayed as zero if not available)

tmpfs is swappable. You have tmpfs filesystem(s) which are being filled beyond safe limits. For comparison, the system I'm typing this on has 200M shared. 6G is too much on an 8G system running a desktop, with stuff like dropbox and Steam at the same time.

You could use the normal tools to find what files are causing the problem. Though it is theoretically possible that the files go away when your X session dies.

$ df -h -t tmpfs
Filesystem      Size  Used Avail Use% Mounted on
tmpfs           1.9G  1.7M  1.9G   1% /dev/shm
tmpfs           1.9G  1.6M  1.9G   1% /run
tmpfs           1.9G     0  1.9G   0% /sys/fs/cgroup
tmpfs           1.9G   80K  1.9G   1% /tmp
tmpfs           376M   20K  376M   1% /run/user/42
tmpfs           376M   20K  376M   1% /run/user/1000

Limit your tmpfs mounts in order to survive your problems, gain the opportunity to analyze them, and maybe even trigger a helpful error message from the software which fills them.

By default, each tmpfs is limited to 1/2 of available RAM.

It is therefore desirable not to proliferate multiple tmpfs mounts with default limits. Distributions aren't quite as good at this as they should be, as you can see above, for my 4GB system.

Apparently it's possible to change the limits at runtime with mount -oremount,size=10% /tmp.

You can put your remount commands somewhere they will run at boot time, e.g. /etc/rc.local (may require systemctl enable rc-local). Note /run/user/* are likely mounted after your script runs; hopefully they have sensible limits already.

The default tmpfs mounts tend not to be listed in /etc/fstab. Under systemd, you can modify the /tmp mount with e.g. systemctl edit tmp.mount. Otherwise, you could grep through your system scripts to find where they mount /tmp; it may use a configuration file you can edit. Another valid option for /tmp would be to disable the tmpfs mount altogether (systemctl disable tmp.mount), just letting programs write to the root filesystem instead.

  • Thank you for your response, but it appears that my tmpfs are not causing my problems. I updated my question with to output of df. – knaecke Apr 30 '16 at 18:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.