View Incident:
http://co-op.engr.sgi.com/BugWorks/code/bwxquery.cgi?search=Search&wlong=1&view_type=Bug&wi=800293
*Status : wontfix Priority : 2
Assigned Engineer : nathans Submitter : nathans
Opened Date : 08/27/00 *Closed Date : 09/04/00
*Fixed By : nathans *Fixed By Domain : engr
*Modified Date : 09/04/00 *Modified User : nathans
*Modified User Domain : engr *Fix Description :
==========================
ADDITIONAL INFORMATION (WONTFIX)
From: nathans@engr (BugWorks)
Date: Sep 04 2000 10:57:34PM
==========================
Oh my - this one's a corker...
I've finally figured out why test 031 is failing for me. It isn't
a libxfs, libsim, repair or mkfs bug ... all work just the way
they were intended.
The problem arises because at the point when we run this test
in our QA suite (from 1 thru 33 sequentially), we have previously
created a small - 3 allocation group - XFS filesystem on the scratch
device. Now, we _do_ actually do a mkfs in test 031, the problem
arises because we use the whole device. In my case, this means I
create an 8 allocation group filesystem with relatively large
allocation groups (the previous, small filesystem fits entirely
within my new first AG).
So, the test now blows away the primary superblock ... and runs
repair. This does the right thing - realises the primary superblock
is useless, and starts hunting for secondaries. It finds two valid
superblocks in short time which seem to agree with each other and
with the fact that there should be only three allocation groups.
Again repair does the right thing, and thinking it has a good SB
promotes the contents of the first secondary into the primary SB.
Of course, at this point we're completely hosed. Even if we did
proceed onward looking for other superblocks on the rest of the
device - there's not alot we could do with them... we'd just have
two valid, conflicting sets of superblocks.
We could possibly eventually decide which set was consistent with
the rest of the AG data structures, but it seems to me that this is
such an unlikely situation to ever happen in practice that its not
worth the effort, nor the pain of going through and initializing the
entire data device to something known to be good (at mkfs time).
Its also never bitten anyone before in IRIX, afaik, so I'm inclined
to simply WONTFIX this "bug".
cheers.
|