REISER4 - A simple test with a filesystem and Oracle

REISER4 - A simple test with a filesystem and Oracle

[Paolo Correnti]

Hallo,

I decided to test some days ago Kernel 2.6 with
Reiser4
on a laptop 3,8 Gb partition.
I installed Oracle 9.2 test DB (size: about 2 Gb ) and
wrote a script to insert 10 millions rows on a
tablespace.

Linux Distribution: Slackware 9.1
Oracle 9.2.0.4

Oracle test tablespace size: 300 Mb

START

Filesystem           1K-blocks      Used Available
Use% Mounted on

/dev/hda6              3701704   1953024   1748680 
53% /u01

After about 7 millions of inserted rows Oracle has had
an abnormal shutdown due to problems writing log
files.

I was monitoring tablespace occupation and it was
almost 50% just before Oracle shutdown, so I was sure
that more than 1 Gb was free for Oracle operations but
.... I founded that partition /u01 was full (?!).

END

Filesystem           1K-blocks      Used Available
Use% Mounted on

/dev/hda6              3701704   3701644        60
100% /u01

After deleting 300 Mb Oracle datafile I obtained that
only 9776 Kb were free (?!)

So I decided to 

root@pamilo:/# umount /u01

root@pamilo:/# measurefs.reiser4 -f /dev/hda6
measurefs.reiser4 0.4.20
Copyright (C) 2001, 2002, 2003 by Hans Reiser,
licensing governed by reiser4progs/COPYING.
 
Packing statistics:
  Formatted nodes:    2593.04b (63.31%)
  Branch nodes:        396.00b (9.67%)
  Twig nodes:         2921.50b (71.33%)
  Leaf nodes:         4095.94b (100.00%)
 
Node statistics:
  Total nodes:         847610
  Formatted nodes:         23
  Unformatted nodes:   847587
  Branch nodes:             1
  Twig nodes:               8
  Leaf nodes:          847601
 
Item statistics:
  Total items:            448
  Nodeptr items:           22
  Statadata items:        210
  Direntry items:           8
  Tail items:               4
  Extent items:           204

root@pamilo:/# fsck.reiser4 --fix /dev/hda6

Fscking the /dev/hda6 block device.
Will fix minor corruptions of the Reiser4 SuperBblock.
Will fix minor corruptions of the Reiser4 FileSystem.
Continue?
(Yes/No): Yes
***** Openning the fs.
Reiser4 fs was detected on the /dev/hda6.
Master super block (16):
magic:          R4Sb
blksize:        4096
format:         0x0 (format40)
uuid:           2fee362a-e9c7-4a2b-ab4c-8f285624eba4
label:          <none>
 
Format super block (17):
plugin:         format40
description:    Disk-format for reiser4, ver. 0.4.20
magic:          R4Sb-Default
flushes:        0
mkfs id:        0x791886f1
blocks:         973932
free blocks:    126254
root block:     76235
tail policy:    0x4 (smart)
next oid:       0x100e0
file count:     209
tree height:    3
key policy:     LARGE
 
***** Tree Traverse Pass: scanning the reiser4
internal tree.
        Read nodes 23
        Nodes left in the tree 23
                Leaves of them 14, Twigs of them 8
        Time interval: Fri Jan 16 13:24:54 2004 - Fri
Jan 16 13:24:55 2004
***** TwigScan Pass: checking extent pointers of all
twigs.
        Read twigs 8
        Time interval: Fri Jan 16 13:24:55 2004 - Fri
Jan 16 13:24:55 2004
***** Semantic Traverse Pass: reiser4 semantic tree
checking.
Error: Node (76237), item (8): StatData of the file
[10004:1616662635f5041:10022] has the wrong bytes
(5634560), Fixed to
(5636096). Plugin (stat40).
.....
..... other similar messages 
.....
 (27754496), Fixed to
(26222592). Plugin (stat40).
Error: Node (423116), item (16): StatData of the file
[10008:78646230312e64:1002e] has the wrong bytes
(22511616), Fixed to
(20979712). Plugin (stat40).
        Time interval: Fri Jan 16 13:24:55 2004 - Fri
Jan 16 13:24:56 2004
***** fsck.reiser4 finished at Fri Jan 16 13:24:56
2004
Closing fs...done
 
No corruption found.

After mounting of /dev/hda6 I finally founded that
almost 300 Mb were available.

root@pamilo:/# df
Filesystem           1K-blocks      Used Available
Use% Mounted on

/dev/hda6              3701704   3390712    310992 
92% /u01

I'm just a DBA, not a kind of filesystem expert, so I
can just feel there's something wrong calculating
space occupation.

Hoping this kind of "hobby" test could be useful.

All the best

Paolo Correnti


__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: REISER4 - A simple test with a filesystem and Oracle

[Paolo Correnti]

I made another test with "only" 5 million records.
The result is OK !

real    58m22.468s
user    28m54.382s
sys     3m53.500s

on a P4 2,8 GHz with 512Mb RAM, to write about 1,5 Gb
of data.

So the first test problems where:

1) Oracle beeing in archivelog mode created something
like 2 Gb of redolog files -> disk full explained

2) The point 1 made a mess ...

3) Also deleting a 300mb datafile I wasn't able to
obtain a 300mb free space and I had to umount -> fsck
-> mount reiser4 partition

Best regards
 
 Paolo Correnti
 
 


__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: REISER4 - A simple test with a filesystem and Oracle

[Alex Zarochentsev]

On Fri, Jan 16, 2004 at 06:40:17AM -0800, Paolo Correnti wrote:
> I made another test with "only" 5 million records.
> The result is OK !
> 
> real    58m22.468s
> user    28m54.382s
> sys     3m53.500s
> 
> on a P4 2,8 GHz with 512Mb RAM, to write about 1,5 Gb
> of data.
> 
> So the first test problems where:
> 
> 1) Oracle beeing in archivelog mode created something
> like 2 Gb of redolog files -> disk full explained
> 
> 2) The point 1 made a mess ...
> 
> 3) Also deleting a 300mb datafile I wasn't able to
> obtain a 300mb free space and I had to umount -> fsck
> -> mount reiser4 partition

can you try sync(1) and see does it help to recover free space ?

> 
> Best regards
>  
>  Paolo Correnti
>  
>  
> 
> 
> __________________________________
> Do you Yahoo!?
> Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
> http://hotjobs.sweepstakes.yahoo.com/signingbonus

-- 
Alex.

Re: REISER4 - A simple test with a filesystem and Oracle

[Alex Zarochentsev]

On Fri, Jan 16, 2004 at 07:11:44AM -0800, Paolo Correnti wrote:
> "Alex Zarochentsev" wrote
> 
> > can you try sync(1) and see does it help to recover
> > free space ?
> 
> Of course, but how do you use sync (I mean with what
> option) ?


just /bin/sync, it should force commit of reiser4 transactions  and, thereby, 
free space.

> 
> Just to know, another "strange" data
> 
> When I use df:
> 
> /dev/hda6              3701704   3042088    659616 
> 83% /u01
> 
> Then 
> 
> cd /u01/oradata/pamodb
> du
> 
> 1261200 ./archivelogs
> 4386399 .
> 
> First data (1,2 Gb) is effective, but 4,3 Gb occupated
> on a 3,7 Gb partition is not real, of course.

du(1) counts blocks (default behaviour), not bytes. 

so if two small files share one block in reiser4 du(1) counts 2 blocks 
instead of 1.


> 
> Best regards
>   
> Paolo Correnti
> 
> > >  
> > >  
> > > 
> > > 
> > > __________________________________
> > > Do you Yahoo!?
> > > Yahoo! Hotjobs: Enter the "Signing Bonus"
> > Sweepstakes
> > > http://hotjobs.sweepstakes.yahoo.com/signingbonus
> > 
> > -- 
> > Alex.
> 
> 
> __________________________________
> Do you Yahoo!?
> Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
> http://hotjobs.sweepstakes.yahoo.com/signingbonus

-- 
Alex.

Re: REISER4 - A simple test with a filesystem and Oracle

[Paolo Correnti]

No results with sync (I mean no changes on free
space).

But these results are very strange ...

df

Filesystem           1K-blocks      Used Available
Use% Mounted on

/dev/hda6              3701704   3042028    659676 
83% /u01

du /u01/oradata

617649  /u01/oradata/archived
56774   /u01/oradata/pamodb/archivelogs
3192028 /u01/oradata/pamodb              <- !!
3809677 /u01/oradata                     <- !!

Regards

PC



__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: REISER4 - A simple test with a filesystem and Oracle

[Alex Zarochentsev]

On Fri, Jan 16, 2004 at 07:39:14AM -0800, Paolo Correnti wrote:
> No results with sync (I mean no changes on free
> space).

that means all trancations are committed already.  

anyway we will re-ckeck reiser4 for existance of space leaks, it is so easy to
add a bug of this kind to reiser4 again ...
  
> 
> But these results are very strange ...
> 
> df
> 
> Filesystem           1K-blocks      Used Available
> Use% Mounted on
> 
> /dev/hda6              3701704   3042028    659676 
> 83% /u01
> 
> du /u01/oradata

what does du -b report ? I hope there are no files with holes...

> 
> 617649  /u01/oradata/archived
> 56774   /u01/oradata/pamodb/archivelogs
> 3192028 /u01/oradata/pamodb              <- !!
> 3809677 /u01/oradata                     <- !!
> 
> Regards
> 
> PC
> 
> 
> 
> __________________________________
> Do you Yahoo!?
> Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
> http://hotjobs.sweepstakes.yahoo.com/signingbonus

-- 
Alex.

Re: REISER4 - A simple test with a filesystem and Oracle

[Paolo Correnti]

After about 5-6 minutes after sending last e-mail
I obtained a "real" df result.

oracle@pamilo:/u01/oradata$ du

1868581 ./archived
10245   ./pamodb/archivelogs
1230589 ./pamodb                 <- OK
3099170 .                        <- OK

It seems there is a kind of very large delay in
reiser4 filesystem info synchronization.

Hope to be not much boring ...

Paolo Correnti

__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: REISER4 - A simple test with a filesystem and Oracle

[Alex Zarochentsev]

On Fri, Jan 16, 2004 at 08:01:54AM -0800, Paolo Correnti wrote:
> After about 5-6 minutes after sending last e-mail
> I obtained a "real" df result.

"df" or "du" ?

> 
> oracle@pamilo:/u01/oradata$ du
> 
> 1868581 ./archived
> 10245   ./pamodb/archivelogs
> 1230589 ./pamodb                 <- OK
> 3099170 .                        <- OK

what processes were accessing the fs those 5-7 min?  
oracle was?

> 
> It seems there is a kind of very large delay in
> reiser4 filesystem info synchronization.
> 
> Hope to be not much boring ...

:) 

thanks for the report. 


-- 
Alex.

Re: REISER4 - A simple test with a filesystem and Oracle

[Paolo Correnti]

Sorry for delay in my answers (I'm at home now ...).

> "df" or "du" ?
 
 du (I made a typing mistake with df)

> what processes were accessing the fs those 5-7 min? 
> oracle was?

Oracle processes only, with no particular activities.
I made also a backup vith Oracle RMAN with data and
backup paths on the same partition and all went fine,
but after having seen good du result.

Next days I'll make the same test with another PC,
kernel 2.6.1 and  a larger reiser4 partition
 
Have a nice week-end

Paolo


__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: REISER4 - A simple test with a filesystem and Oracle

[Hans Reiser]

Alex Zarochentsev wrote:

>On Fri, Jan 16, 2004 at 07:11:44AM -0800, Paolo Correnti wrote:
>  
>
>>"Alex Zarochentsev" wrote
>>
>>    
>>
>>>can you try sync(1) and see does it help to recover
>>>free space ?
>>>      
>>>
>>Of course, but how do you use sync (I mean with what
>>option) ?
>>    
>>
>
>
>just /bin/sync, it should force commit of reiser4 transactions  and, thereby, 
>free space.
>  
>
Reiser4 should do this automatically when it runs out of disk space, I 
think I said this before, but folks were probably busy with other things 
when I said it....  Please correct it.


-- 
Hans

Re: REISER4 - A simple test with a filesystem and Oracle

[Hans Reiser]

Well, ignore my email just sent since sync had no effect.

Let us have grev or you experiment with doing things like file 
overwrites in near to out of disk space situations, and see if we are 
doing a proper job of forcing commits when we get low on disk space.

-- 
Hans