URGENT: NEED ORACLE & AIX EXPERT OPINION PLS

[KOG]

Hi Folks,

We are operating AIX 4.3 and Oracle 8.0.5 on RS/6000 box and one of our disk corrupted last week, we had it replaced now this disk is so full and we have installed new disk which is 4.5gb in size as opposed to the current disk 2 gb size.

Now our db is up and running, IBM AIX support say it is ok to keep the db up and running while migrating one of the filesystem and the datafiles within that filesystem from one disk to another without shutting the db down.

I am confused here, is it really possible to migrate the filesystem and its contents when oracle is constantly reading and writing to the disks?

Pls help me to answer this ques.

Many thanks and regards

Katherine

 

[KenCunningham]

Katherine,

I'm not sure about AIX, but it's certainly not something I'd do with our Solaris-based Oracle systems. Better to be safe than sorry is my opinion. Will your Oracle application still know where it's datafiles are following this migration? Of course, things might be different with AIX and others might be able to advise accordingly. Good luck.

 

[KOG]

AIX Support say the filesystem structure is the same and so is the location of the datafiles but the actual disk will be different .. I agree with you that it is better shut the db down to be on the safe side.

I am having a lot of probs with IBM and Oracle, it is SO hard to find an 'inner' person who knows BOTH AIX and Oracle to answer my questions.

Many thanks for your immediate response, much appreciated.

Regards

Katherine

 

[KenCunningham]

I agree with you it's difficult to obtain an overall picture sometimes. My problem with what they suggest is that even if you did decide to migrate whilst the database is active, I don't see how any synchronisation could take place between what's been written to disk A and what needs to be written to disk B, even if the filesystem structure is the same. Just seems quite an unneccessary risk to me! Good luck - let us know how you get on.

 

[KenCunningham]

Sorry, my

"what's been written to disk A and what needs to be written to disk B"

should have been:

"what's already been written to disk B and what still needs to be written to disk B"

Clear as mud, eh?

 

[apb99]

if you use lv mirroring you can keep the db up

add a copy to the logical volume then sync that lv, when the sync is finnished the data will exist on both physical disks, removing the copy from the original disk will have completed the move.

the db doesn't need to be down as the syncvg will ensure all logical partitions in the logical volume are sync'd.

performance of db can be an issue as the disks will be busy during the sync but i have done this on oracle systems that are 24x7

 

[KOG]

This is the commands I got from AIX Support,

1. Add hdisk12 to Oracle vg using this command

#extendvg oracle hdisk12

Then make a mirror image of hdisk7 containing /oracledata8 filesystem and
its contents

#mklvcopy oracledata8 2 hdisk12

#syncvg -v oraclevg

Check if I have two copies

#lslv oracledata8

Also check with lspv -p hdisk12 that you have the same number of physical
partitions allocated to oracledata on hdisk12 as hdisk7

What happens when I remove the original, will oracle automatically switch over to new disk or is that part of AIX configuration?

 

[apb99]

the lv will have two complete copies of the db one on each disk after the syncvg has finnished successfully.

if you run lsvg oraclevg -l this will list the lv's in the volume group, prior to mirroring the report will say the same number of lp's as pp's and that it is open/syncd,

after the mklvcopy has been run the lsvg oraclevg -l would show twice the number of pp's as lp's but a state of open/stale

after the syncvg, then the state should be open/syncd again

Note: all logical volumes in the volume group need to be done... these include jfslogs

 

[KOG]

But AIX Support say no as the filesystem /oracledata8 is still mounted and accessible ! Is that true?

Can you please explain to me what happens to Oracle datafiles when it is mirrored? Will oracle switch over so I can increase the filesystem on new disk?

I wonder why we bother paying so much money for AIX support if they can't get their facts right.

Excuse me for my language but I am just so frustrated at the moment and I really do find this web far more helpful and educational than any other so called 'reliable source'.

Regards and many thanks to you apb99.

Katherine

 

[apb99]

can you post the output from "lsvg oraclevg -l"

when using logical volumes the oracle db doesn't see the changes happen it will uese the lv and aix manages the mapping of the logical partitions to physical partitions.

i have done this on systems using oracle when we have lost mirroring of the logical volume due to disk failure and need to replace the disk without downing the system.

what you are doing is using mirroring to move data ... the overall performance of the db can be effected whilst the sync of data is carried out but the db will not fail ...unless you remove the original copy before the sync is complete

 

[KOG]

Hi there,

Here is the listings

$ lsvg oracle -l
oracle:
LV NAME TYPE LPs PPs PVs LV STATE MOUNT POINT
oracledata1 jfs 185 185 1 open/syncd /oracledata1
loglv00 jfslog 1 1 1 open/syncd N/A
oracledata jfs 190 190 1 open/syncd /oracledata
oracleindex jfs 236 236 1 open/syncd /oracleindex
oracledata8 jfs 539 539 2 open/syncd /oracledata8
oracletest jfs 236 236 1 open/syncd /oracletest
orabackup jfs 964 964 3 open/syncd /orabackup
oracledata2 jfs 123 123 2 open/syncd /oracledata2
backup jfs 153 153 2 open/syncd /backup

We have five disks on ORACLE vg, and all are now full. And filesystem /oracledata8 has 100% on it but it has some free space but do not want to take another risk as we had a corrupted disk last week and it was replaced (the one with /oracledata8 filesystem was on).

/dev/oracledata8 4415488 34360 100% 38 1% /oracledata8

Are you sure it is 100% safe to mirror the hdisk7 disk where filesystem /oracledata8 is held to new disk hdisk12?

Many thanks to you for all your help.

Regards

Katherine

 

[apb99]

oracledata8 is on 2 pv's thats 2 disks is a jfs file system with 539 partitions

run a "lv oracledata8 -l"

this will show the physical disks in use

then "lspv hdisk??" this will show pp's in use on each disk

then "lspv hdisk?? -l" to show what logical volumes are on that disk

post the results

 

[KOG]

Tried extending oracle vg, got error message

I can't extend the oracle vg, got this error message

# extendvg oracle hdisk12
0516-1162 extendvg: Warning, The Physical Partition Size of 4 requires the
creation of 1075 partitions for hdisk12. The limitation for volume grou
p
oracle is 1016 physical partitions per physical volume. Use chvg comman
d
with -t option to attempt to change the maximum Physical Partitions per
Physical volume for this volume group.
0516-792 extendvg: Unable to extend volume group.


What shall I do?

 

[bjverzal]

The "PP Size" of the volume group is too small to support the new bigger disk. You can use the "chvg -t" to change the maximum allowable # of PP's per disk. Perhaps try a "chvg -t 2" (by default, t=1) and this will double the # of PP's per disk currently allocateable. There is a catch - you will only be allowed half the number of disks in the volume group after you do this.

Bill.

 

[apb99]

As Bill has said the defined pp size for oraclevg is 4mb that and the fact you can only have 1016 partitions on a disk ( ibm default number of partitions on disk i beleive ) meant it wouldn't extend ... the chvg -t2 will double the number of partitions allowed.

my concern from your earlier post was that the oraceldata8 is on 2 PV's thats 2 physical disks

cheers

alan

 

[Yegolev]

You will have to change the "factor" of the vg so that you can address the larger disk. Like so:

chvg -t 2 oracle

You can then have disks with twice the number of PP but only half the total number of disks in the vg. So you will end up with 2032 possible PP per disk with a max 16 disks in that vg.

As for the mirroring thing, I do it all the time. Oracle has no knowledge of what AIX is doing underneath the filesystem/file level (it would be awful if Oracle wrote directly to storage). Oracle will just keep doing what it always does while the LVM handles the mirroring. The command I usually use is "migratepv" because of our layout, but all the migration commands do pretty much the same thing: the LPs are mirrored, synced, then the mirror is broken. There is no "Oracle switch" since to any application, including Oracle, it all looks the same.

I migrated dozens of our production servers during normal business hours from SCSI-connected EMC DASD to fibre-connected EMC DASD and no one noticed... aside from the increased IO speed. =) No outage, no nothing.

 

[KenCunningham]

Thanks for the education today, folks!