Re: stress test on ppc

[Thomas Graichen <news-innominate.list.sgi.xfs@xxxxxxxxxxxxx>]

Russell Cattelan <cattelan@xxxxxxxxxxx> wrote:

> Ok this doesn't look good.
> Several fields looked bad already.
> Notably the rootino should be 128

> Was this output taken directly after a mkfs or after a mount?

[just a start info: all the stuff in this posting is on ppc]

directly after a mkfs - i think i remember it was 128 at some time
on xfs before - also xfs_repair or xfs_ch does not find anything
on that fs and it runs as far as i can see close to rock solid
on ppc so i assume the problem is in xfs_db

> We need to narrow down our problems.
> First we need to figure out if mkfs is correctly
> making the file system.

ok - lets see

ppc:~ # dd if=/dev/zero of=/tmp/testfile bs=1024k count=32
32+0 records in
32+0 records out
ppc:~ # mkfs -t xfs -f /tmp/testfile 
meta-data=/tmp/testfile          isize=256    agcount=2, agsize=4096 blks
data     =                       bsize=4096   blocks=8192, imaxpct=25
         =                       sunit=0      swidth=0 blks, unwritten=0
naming   =version 2              bsize=4096  
log      =internal log           bsize=4096   blocks=1200
realtime =none                   extsz=65536  blocks=0, rtextents=0
ppc:~ #
...
[root@orange /tmp]# uname -a
Linux orange.bln.innominate.de 2.4.0-XFS-test10 #2 SMP ... i686 unknown
[root@orange /tmp]# scp ppc:/tmp/testfile .
testfile             100% |*****************************| 32768 KB    00:17    
[root@orange /tmp]# xfs_db -r -c sb -c p testfile 
magicnum = 0x58465342
blocksize = 4096
dblocks = 8192
rblocks = 0
rextents = 0
uuid = a0a20e0a-7fcb-4fc8-a7de-e79e939245de
logstart = 4100
rootino = 128
rbmino = 129
rsumino = 130
rextsize = 16
agblocks = 4096
agcount = 2
rbmblocks = 0
logblocks = 1200
versionnum = 0x2084
sectsize = 512
inodesize = 256
inopblock = 16
fname = "\000\000\000\000\000\000"
fpack = "\000\000\000\000\000\000"
blocklog = 12
sectlog = 9
inodelog = 8
inopblog = 4
agblklog = 12
rextslog = 0
inprogress = 0
imax_pct = 25
icount = 64
ifree = 61
fdblocks = 6980
frextents = 0
uquotino = 0
pquotino = 0
qflags = 0
flags = 0
shared_vn = 0
inoalignmt = 2
unit = 0
width = 0
dirblklog = 0
[root@orange /tmp]# 

so its xfs_db i think (the other direction shows the same: an i386
created image also gives wrong results with the ppc xfs_db)

> Second is xfs_db correctly reporting the values
> on disk.

no - mkfs is ok - xfs_db is wrong and i assume the xfs in the
kernel is also ok - the stress tests only give bad results due
to the broken xfs_db

> Finally where in XFS are we trashing AG blocks.

keep in mind: the agf trashing was on the alpha! - so this has more
to do with 8k blocksize or 64bit but nothing with endianness - just
to keep that in mind

will do the other stuff below on the alpha in a minute in another
posting ...

t

> Ok lets do things in this order
> mkfs a file system
> od -c -N 8 -j 512 /dev/sdb1
> should result in
> 0001000   X   A   G   F  \0  \0  \0 001
> od -c -N 8 -j 1024 /dev/sdb1
> and
>  od -c -N 8 -j 1024 /dev/sdb1
> 0002000   X   A   G   I  \0  \0  \0 001

> Those are the magic #s

> now try
> xfs_db
> sb 0
> print
> agf 0
> print
> agi 0
> print

> save output

> Mount the file system
> run od commands and xfs_db command again
> save output

> Now do some FS activity
> run od command and xfs_sb commands again
> save output.

> Send all the output to us.

> This should tell us where we need to start fixing things first.

> I suspect we will need to find an alpha around someplace to actually
> get some of this stuff debugged.

-- 
thomas.graichen@xxxxxxxxxxxxxx
technical director                                       innominate AG
clustering & security                             the linux architects
tel: +49-30-308806-13   fax: -77             http://www.innominate.com