Chunk dropping problem::

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

agarwalkaditya@gmail.com - 26 Mar 2007 10:03 GMT

Hi All Following is the result of tbstat -d result::

# tbstat -d

RSAM Version 5.20.UC1 -- On-Line -- Up 01:10:42 -- 4004 Kbytes

Dbspaces
address number flags fchunk nchunks flags owner name
1000d300 1 1 1 3 N informix rootdbs
1000d330 2 1 2 2 N informix
netplandbs
1000d360 3 1 3 3 N informix logsdbs
3 active, 38 total

Chunks
address chk/dbs offset size free bpages flags pathname
1000a2e8 1 1 0 16000 158 PO- /dev/
rinformix/roo
tdbs
1000a42c 2 2 0 16000 0 PO- /dev/
rinformix/net
plandbs
1000a570 3 3 0 16000 672 PO- /dev/
rinformix/log
sdbs
1000a6b4 4 2 0 100000 32302 PO- /
local_data/NetPla
nDB/netplandbs1
1000a7f8 5 3 0 200000 167497 PO- /dev/
rinformix/log
sdbs1
1000a93c 6 3 0 200000 199997 PO- /dev/
rinformix/roo
tdbs1
1000aa80 7 1 0 150000 149997 PO- /dev/
rinformix/roo
tdbs2
1000abc4 8 1 0 150000 149997 PO- /
local_data/NetPla
nDB/rootdbs2
8 active, 38 total
I was adding the chunk to the rootdbs but I mistakenly I added two
different chunks indicating
to same files::

1000aa80 7 1 0 150000 149997 PO- /dev/
rinformix/rootdbs2

1000abc4 8 1 0 150000 149997 PO- /
local_data/NetPlanDB/rootdbs2

These two chunks are using same physical file.Now I want to drop
anyone of these two chunks but not able to do so.Could you please tell
me the syntax??I t is having Version 5.20.UC1 -- On-Line version.I
tried following command
tbspaces -d rootdbs -p /dev/rinformix/rootdbs2 -o 0
but it did not help.Please help me.

Reply to this Message

Art S. Kagel - 26 Mar 2007 16:47 GMT

> Hi All Following is the result of tbstat -d result::
>
> # tbstat -d
>
> RSAM Version 5.20.UC1 -- On-Line -- Up 01:10:42 -- 4004 Kbytes

<SNIP>> I was adding the chunk to the rootdbs but I mistakenly I added two

> different chunks indicating
> to same files::
[quoted text clipped - 11 lines]
> tbspaces -d rootdbs -p /dev/rinformix/rootdbs2 -o 0
> but it did not help.Please help me.

You're hosed trying to fix this yourself. OL5 does NOT, as you have found
out, follow links to determine if a new chunk path is identical to another
chunk in anyway but the actual pathname, offset, and length. Been there,
done that. To fix this yourself is VERY complex, I won't describe it in
detail here. However, the easy fix is to call tech support for a dial-in
and they have tools to truncate the chunk table and patch up the dbspace header.

The problem right now is that the chunk header page doesn't match it's chunk
table entry.

Q&D description of fixing this? Shutdown the engine. Get the first page
from another chunk 7 of dbspace 1 and from a chunk 8 of dbspace 1 from
another instance. Move one of the links to point to different disk space.
Now, using dd, overwrite the first page of chunk 7 with the page from chunk7
from another instance, and similarly for the chunk 8. Then you will be able
to restart the instance and drop both chunks (and I would definitely drop
BOTH - you can readd a good one later). That's just a sketch of the process.

Like I said, complicated and not for the faint hearted. I did it once when
Informix couldn't dial into a particular machine for some reason I no longer
recall so I had no other choice. You risk all of your data - I had a
parallel back up server to fall back on. Instead just call IBM.

Art S. Kagel

Reply to this Message

Superboer - 26 Mar 2007 19:06 GMT

Hello Art,

sounds about right for V > 5
(that is if the engine did not figure out that some mucked up and
marked the chunk/dbspace down then you have more work...)
Anyways: Version 5 does not allow to drop chunks as far as i can
remember.

So

phone TS and ask for help
you may be facing an unload reinit etc
WARNING if dbexport /import is used then be aware of 2 GB limit of
files
be aware that dbexport does not bring in lock modes and extent
sizes!!!!
goto the catalog and grab that info before reloading!!!
or restore from archive.(test on a testbox may be first??????)

Superboer.

Reply to this Message

Superboer - 26 Mar 2007 19:19 GMT

> Hello Art,
>
[quoted text clipped - 16 lines]
>
> Superboer.

Hmmm to trigger happy, one should get page 0 and page 2 ( is the chunk
free list page) from that chunk.
Anyways PHONE IBM TS

Reply to this Message