Memory leak with linux-2.6.19-rc6-CITI_NFS4_ALL-1

[Le Rouzic]

 Hi,

  In fact we can get the problem without using kerberos environnement,
  by only running connectathon tests.

    ./runtests -N 10000 -a -t /mnt/nosec1/test_nfs4_nfs1

  We still see a strong size-512 buckets allocation not freed if we stop 
before the out of memory crash on the server.

 Result of 'slabtop -s c'
      OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
1693880 1693880 100%    0.50K 211735        8    846940K size-512

Best regards






>Le Rouzic a écrit :
>
>  
>
>>Bryce Harrington wrote:
>>
>> 
>>
>>    
>>
>>>The performance data for the iozone runs matches up with prior runs and
>>>show no performance decreases, however during the third iozone run
>>>(against krb5i), the nfsd server triggered the oom-killer, as can be
>>>seen here:
>>>
>>>  http://crucible.osdl.org/runs/3186/sysinfo/nfs02.console
>>>
>>>
>>>   
>>>
>>>      
>>>
>>Hi,
>>I got also an Out Of Memory issue with 
>>linux-2.6.19-rc6-CITI_NFS4_ALL-1.diff on both client and server.
>>I get a crash on the server side due to an Out of memory when under 
>>krb5i I have running several benchs
>>simultaneously such as iozone, connectathon,fsx, nfs_fsstress,fsb.
>>
>>On the server side, we can see with 'slabtop -s c' the number of 
>>size-512 buckets 
>>
>>    
>>
>   Hi,
>     I get the same issue in a 'krb5p' environnement running the
>     same benchmarks.
>     Still a lot of  size-512 buckets  showed by slabtop -s c:
>
>     1681416 1681398  99%    0.50K 210177        8    840708K size-512
>
>     Best regards
>
>
>  
>
>>increasing a lot:
>>
>> OBJS       ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
>>1686840 1686825    99%      0.50K     210855            8           
>>843420K    size-512
>>   226383   213350    94%      0.06K         3837           59         
>>  15348K    size-64
>>         842        792     94%      2.00K           421            
>>2               1684K    size-2048
>>     18760      8442     45%      0.05K           280           67     
>>        1120K    buffer_head
>>      6831       2637     38%       0.14K          253           27     
>>        1012K    dentry_cache
>>      1278         602     47%       0.58K          213             
>>6                 852K    ext3_inode_cache
>>        179        178      99%       4.00K         179             
>>1                716K     size-4096
>>
>>In /var/log/messages on the server side we get:
>>Nov 27 15:33:06 nfs4 kernel: nfsd invoked oom-killer: gfp_mask=0xd0, 
>>order=0, oomkilladj=0
>>Nov 27 15:33:06 nfs4 kernel:  [<c0104b31>] show_trace_log_lvl+0x1a/0x2f
>>Nov 27 15:33:06 nfs4 kernel:  [<c0104c2a>] show_trace+0x12/0x14
>>Nov 27 15:33:06 nfs4 kernel:  [<c010540a>] dump_stack+0x19/0x1b
>>Nov 27 15:33:06 nfs4 kernel:  [<c0151362>] out_of_memory+0x6b/0x1a1
>>Nov 27 15:33:06 nfs4 kernel:  [<c0152941>] __alloc_pages+0x216/0x29f
>>Nov 27 15:33:06 nfs4 kernel:  [<c03bf5ac>] svc_recv+0xc9/0x3a1
>>Nov 27 15:33:06 nfs4 kernel:  [<c01f1a69>] nfsd+0xd3/0x286
>>Nov 27 15:33:06 nfs4 kernel:  [<c0104a43>] kernel_thread_helper+0x7/0x10
>>
>>Nov 27 15:33:07 nfs4 kernel:  [<c0104b31>] show_trace_log_lvl+0x1a/0x2f
>>Nov 27 15:33:07 nfs4 kernel:  [<c0104c2a>] show_trace+0x12/0x14
>>Nov 27 15:33:07 nfs4 kernel:  [<c010540a>] dump_stack+0x19/0x1b
>>
>>
>>Issue is registered in bugzilla under #132
>>
>>------------------------------------------------------------- 
>>Bug: 
>> http://developer.osdl.org/dev/nfsv4/bugzilla/show_bug.cgi?id=132  
>>
>>
>>
>>Best regards
>>
>>
>>
>> 
>>
>>    
>>
>
>
>  
>


-- 
-----------------------------------------------------------------
Company : Bull, Architect of an Open World TM (www.bull.com)
Name    : Aime Le Rouzic 
Mail    : Bull - BP 208 - 38432 Echirolles Cedex - France
E-Mail  : aime.le-rouzic at bull.net
Phone   : 33 (4) 76.29.75.51
Fax     : 33 (4) 76.29.75.18
-----------------------------------------------------------------