[dm-devel] Re: [Bugme-new] [Bug 7158] New: Out of memory happen in snapshot

[Andrew Morton]

I've switched this to the mailing list - please send all replies via email
(not the bugzilla web interface) and please ensure that all cc's are
retained.


On Wed, 13 Sep 2006 20:20:00 -0700
bugme-daemon at bugzilla.kernel.org wrote:

> http://bugzilla.kernel.org/show_bug.cgi?id=7158
> 
>            Summary: Out of memory happen in snapshot
>     Kernel Version: 2.6.17.13
>             Status: NEW
>           Severity: blocking
>              Owner: agk at redhat.com
>          Submitter: kevin_cheng at thecus.com
> 
> 
> Most recent kernel where this bug did not occur:
> Distribution:
> Hardware Environment:
>   celoron M 1,5GHz
>   memory :256MB
>   swap memory have 1GB
> 
> Software Environment:
>   My volumn be created on RAID0
>   kernel is 2.6.17.13
>   Filesyetm is ext3
>   
> Problem Description:
>    We create the 8 snapshots in one volumn ,and copy a 4G file to the volumn ,
> and system happen "out of memory" and oom-killer kill my processes. 
> Someone have happen this problem ,have any way to solve it ? Thanks!!
> 
> Steps to reproduce:
>   - use three hdd to create RAID0
>   - create vg on RAID0 is called vg0
>   - Create lv on vg0 is called lv0
>   - mke2fsck -j /dev/vg0/lv0 , format to ext3 filesystem
>   - mount /dev/vg0/lv0 to /raid/data
>   - create a folder pub on /raid/data
>   - create 8 snapshots base on lv0 ,snapshot lv sizes have 22GB,every snapshot 
> are the same.
>   - use dd command to create 4G file to lv0
>     dd if=/dev/zero of=/raid/data/4G.bin bs=1M count=4096
>   - system have some thing happen 
>     sometime system will crash ,not any more message.
>     sometime have "out of memory" in dmesg
> 


The oom-killer info which you've included there is ambiguous.  A large
amount f memory is in slab, which might indicate a slab leak.  But there is
also a large amount of memory on the page LRU, which one would expect to
have been reclaimed before declaration of OOM.  

Could you please capture the contents of /proc/slabinfo after the
oom-killing and send that?