[Gluster-devel] Create directory error

Anand Avati avati at zresearch.com
Tue Aug 26 06:15:45 UTC 2008


Ben,
 this is a known issue with FUSE kernel module. One workaround is to keep
doing 'echo 2 > /proc/sys/vm/drop_caches' in your client regularly. This
will keep the flushing stale dcache. With newer FUSE versions filesystems
will have an API to forget inodes when this problem can be solved the 'right
way'.

avati

2008/8/26 Ben Mok <benmok at powerallnetworks.com>

> Hi All,
>
> I got some error when create directories. If create several directories
> from
> client A and then delete the directories from client B. The error message
> will be showed as below when the same directories are created again from
> client A, and the return value is 1.
>
> mkdir: cannot create directory `/pgfsfs/test/a98': Device or resource busy
>
>
>
> In fact, all directories can be created, just some creating got error
> message and return 1. If create and delete directories from same client, it
> have no problem.
>
> You can repeat the symptom easily by below script
>
>
>
> #!/bin/bash
>
> folder="/pgfs/test/"
>
> for ((i=0;i<=100;i+=1)); do
>
> mkdir $folder$1$i
>
> echo $?
>
> Done
>
>
>
> Testing procedure:
>
> 1.       run script from client A
>
> 2.       rm -rf * from client B
>
> 3.       run script from client A again
>
>
>
> Version of clients and servers are 1.3.10
>
> ----------
>
> Client
>
> ---------
>
> volume pgfs01
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.11
>
>     option remote-subvolume storage-ds-locks
>
>   end-volume
>
>
>
>   volume pgfs01-ns
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.11
>
>     option remote-subvolume storage-ns
>
>   end-volume
>
>
>
>   volume pgfs02
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.12
>
>     option remote-subvolume storage-ds-locks
>
>   end-volume
>
>
>
>   volume pgfs02-ns
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.12
>
>     option remote-subvolume storage-ns
>
>   end-volume
>
>
>
>   volume pgfs03
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.13
>
>     option remote-subvolume storage-ds-locks
>
>   end-volume
>
>
>
>
>
>   volume pgfs04
>
>     type protocol/client
>
>     option transport-type tcp/client
>
>     option remote-host 172.16.40.14
>
>     option remote-subvolume storage-ds-locks
>
>   end-volume
>
>
>
>   volume afr-ns
>
>        type cluster/afr
>
>        subvolumes pgfs01-ns pgfs02-ns
>
>   end-volume
>
>
>
>   volume afr01
>
>        type cluster/afr
>
>        subvolumes pgfs01 pgfs02
>
>   end-volume
>
>
>
>   volume afr02
>
>        type cluster/afr
>
>        subvolumes pgfs03 pgfs04
>
>   end-volume
>
>
>
>   volume storage-unify
>
>           type cluster/unify
>
>           subvolumes afr01 afr02
>
>           option namespace afr-ns
>
>           option scheduler rr
>
>           option rr.limits.min-free-disk 5%
>
>   end-volume
>
>
>
> ---------
>
> Server
>
> ----------
>
> volume storage-ds
>
>           type storage/posix
>
>           option directory /pgfs/storage
>
>   end-volume
>
>
>
>   volume storage-ns
>
>           type storage/posix
>
>           option directory /pgfs/storage-ns
>
>   end-volume
>
>
>
> volume storage-ds-locks
>
>  type features/posix-locks
>
>  subvolumes storage-ds
>
>  option mandatory on
>
> end-volume
>
>
>
>   volume server
>
>     type protocol/server
>
>     option transport-type tcp/server
>
>     subvolumes storage-ds-locks
>
>     option auth.ip.storage-ds-locks.allow 172.16.*
>
>     option auth.ip.storage-ns.allow 172.16.*
>
>   end-volume
>
>
>
>
>
> Thx a lot,
>
> Ben
>
>
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at nongnu.org
> http://lists.nongnu.org/mailman/listinfo/gluster-devel
>



-- 
If I traveled to the end of the rainbow
As Dame Fortune did intend,
Murphy would be there to tell me
The pot's at the other end.



More information about the Gluster-devel mailing list