Lustre discuss - Jul 2011 - Should run_*.sh CLIENT_LOADS scripts fail during failover?

Hi,

During recovery_*_scale.sh tests those commands used in
CLIENT_LOADS surely will result in write error since the
lustre filesystem is no longer available.

But I saw a comment in recovery_mds_scale.sh says:
ERRORS_OK=""    # No application failures should occur during this
test.

If ERRORS_OK not set, the $END_RUN_FILE will then be created
and thus while-loop will break and application exit.

I do not understand how to keep application up until servers
recovered during the tests? Please enlighten me. :(

Thanks,
Jay

On 11-06-30 09:56 PM, Jay Lan wrote:
> Hi,
Hi,
> During recovery_*_scale.sh tests those commands used in
> CLIENT_LOADS surely will result in write error since the
> lustre filesystem is no longer available.
Not if the target is mounted in failover (vs. failout) mode.  One of the
basic principles of Lustre is that when a target fails (and it is
mounted in failover mode) I/O to/from it will block (not fail) until the
target is functional again.

The recovery_*_scale.sh tests are written to stress test that basic
principle.

b.

-- 
Brian J. Murrell
Senior Software Engineer
Whamcloud, Inc.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20110704/b1fd9774/attachment.bin