clustered DB and RAC

Hi,

OS AIX 5L
DB Oracle 9.2.0.1

Im a new hired dba for a banking company. They have two nodes,
the PROD server and the DEV server which are clustered. They also
have a 3rd party apps which is SAP financials. The SAP team
developers where the one who set-up their apps and the ORACLE
9.2.0.1 database on the two nodes. I presumed they clustered it .
As the new dba Im i little bit cunfused about clustered database. As what I understood it is the same as the RAC right? Two instances on
two oracle homes running against 1 shared database? But the set-
up here is quite different. The PROD server it has its own PROD database and PROD instance. and in DEV server has its own DEV
database and instance too, just like stand alone set-up. But the two
are clustered. Is this kind of clustering valid?

My problem now is...im task to install the patch 9.2.0.5 on both
nodes (oracle homes). So i installed first on the DEV node (not
knowing at first that they are clustered...i thot at first they are a
stand alone set-up). I successfully patched the DEV oracle home.
Then I transferred to the PROD node and try to install the patch
again. To my surprise I cant install the patch because is
says "Theres no patch that needs to be installed".
I tried to run sqlplus on DEV node and the version is now 9.2.0.5
while the sqlplus on PROD is still 9.2.0.1.

Can you help me pls how to patch the PROD node? Im afraid to
experiment other options coz its a PRODUCTION db and I might mess
it up. What do u think did i miss here. I read also that if u are to
patch a cluster u have to run the root.sh to all the nodes. So in my
case since i patched first the DEV node, do i need to copy/ftp
the root.sh found on the DEV ORACLE_HOME to the PROD
ORACLE_HOME and run it? Will running the root.sh alone update or
upgrade the sqlplus version on PROD to 9.2.0.5?

Thanks a lot

As a new DBA you should have been provided with a
Acceptance environment where you'd do this work(which has same setup, although nodes named as DEV and PROD is very dubious).
Anyways apparently you haven't been provided with the same, I presume.

It's a weird scenario anyways you need to investigate if it's the OS which is clustered. I'm guessing that's it's just some TAF kind of setup(from the SAP app dev's point of view) where if the DEV db fails it'll find the DEV schema on the PROD box and vice versa.

But you do need to investigate properly, ask someone for a schematic diagram of what goes where. Beginning right down from the HW and then working your way up to the SW.

Thanks,

So its a valid config to have two separate database on a clustered set-up
oracle_home.

I ask the HW engineer and he said that the setup was like what u said. If one node is down then the disk on that node will be transferred to the active node, with some intervention and the ORACLE_HOME of the active node will adopt it, The two nodes are
clustered but not a full-pledged one. Somewhat like half-done HACMP design. The engineer doesnt know oracle
so he cant comment on the clustered db. I ask the SAP dev team and they
said they just run their out-of-the-box db installer and cant comment on
the what to do wth the clustering. They thought also that they have stand alone PROD and DEV set-up too.
But with this clustered ORACLE_HOME, i dont know how to patch the other
home w/o encountering the "Theres no patch need to be applied" error. I
even shutdown the DEV server hoping the PROD will not see that cluster,
but to no avail. Do you have any idea how to turn this cluster off(at least on the oracle side, not the HW side), so that
i can patch the PROD server?

Thanks again

You got it right with your RAC understanding. However, since u have confirmed that there are both a DEV and PRD instance running on each ibm box. And your seniors told u that they are ibm clustered.

Therefore, most likely, its the PRD instance that is going to fail over to DEV if PRD hardware has some issues. IBM HACMP is a hardware clustering failure. Why DEV is there for the fail over is so that the box will not be left idling just waiting for a fail over to take place.

yes, i can understand that. So our database is a cluster which has each own instance and database separately. So how can i patch each node pls?

sounds like failsafe to me and not rac - u sure it is rac?

Hi friend,

Thanks...maybe is it a failsafe. I dont really understand much the difference about cluster, failsafe, and RAC. I just know they
are synonymous dierivatives of each other. All i know is RAC shared
one database and have more that one instance on separate nodes.

If I have a cluster database with 2 nodes PROD AND DEV,
is there an order to which node is to be patch first? Is there a term
"primary node" and "secondary node"? How can I tell which one
is primary and which one is secondary?

The RAC Patch install procedure says...

1. If u are installing the patch set in a cluster invironment, The
cluster node selection screen will appear.Click each participating
node and install. then run root.sh to all the participating nodes.

Please tell me if my understanding in cluster patch install is right.

1. I can choose 1 and only 1 node to where i can install the patch.
From this node, i will install the patch to all the participating
nodes in a cluster. Once I have chosen that node, I can not go
to other nodes to resume installing the same patch.

It does not tell me which node to install first. So i can assume any
node will do.

...So in may case, since i have patch the DEV node first, then I
will run the patch from DEV node again and click the PROD node
from there? and install the patch? then it will generate a root.sh
on the PROD sever home? even though im installing from another
server (DEV). Will this solve my problem "Theres no patches that
need to be applied"?

How can i detach the a node from the cluster and make it
stand alone?

How can i re-attach it?


Thanks a lot

failsafe is only for windows

really :confused: lol :D ...i thot stmontgo doesnt sound like Mr H.
Anyways,,,did i made myself clear? can anyone help pls.

Quote:

---

Originally Posted by yxez

really :confused: lol :D ...i thot stmontgo doesnt sound like Mr H.
Anyways,,,did i made myself clear? can anyone help pls.

---

you really need to do some reading on dataguard and rac, your questions are well, odd. What are you talking about when you say
"How can i detach the a node from the cluster and make it
stand alone"

please more than ever read the rac docs

Quote:

---

Originally Posted by pando

failsafe is only for windows

---

yeah whoops - missed that - thanks

Quote:

---

Originally Posted by stmontgo

you really need to do some reading on dataguard and rac, your questions are well, odd. What are you talking about when you say
"How can i detach the a node from the cluster and make it
stand alone"

---

I mean, if you have a cluster of servers(nodes) with separate ORACLE_HOME in each also in a cluster. How can i make one
of the cluster member behave like a stand alone server with
respect to oracle clustering.

Thanks

Again, you haven't provided with a detailed setup at your shop. So let me (hmm there we go again) try to *imagine* how your SAP app is setup.
Scenario 1:
------------
DEV Box has
1. SAPProd application server(?)
2. ORACLE_HOMEDEV(dev and prod schema)

PROD Box has
1. SAPDEV application server(?)
2. ORACLE_HOMEPROD(dev and prod schema's)

And here the "SAPProd application server" talks to the PROD box and
"SAPDEV application server" talks to the DEV box. Should PROD box Oracle fail, both app servers talk to the DEV box and vice versa.

Scenario 2
-----------

SAP Box1
Oracle_home2

SAP Box 2
Oracle_home1

Should SAP Box 1 fail to connect to Oracle_home1(preferred db), it reverts back to oracle_home2(which is sync'ed via log shipping, I guess) and should SAP Box1 fail(app server), SAP Box 2 is woken up(it's merely a HTTP/Container) and it continues to talk to oracle_home1(preferred db).

I hope I'm not confusing you. Go back to your HW engineer. Or you are not understanding well or the HW guy didn't really set it up himself(probably done by some external shop).

Let us know if this helps.

Tarry

Thanks again Tarry,

But i must say the people here doesnt know much of the set-up and so with Oracle clustering. The servers was intended to be (maybe a RAC) but it was not done so. Instead they have installed independent database for each server.

Scenario 3
-----------

SAP Box1(dev)
Oracle_home1
with dev instance
with dev database

SAP Box 2(prod)
Oracle_home2
with prod instance
with prod database

The two nodes are clustered. and both oracle_home1 and
oracle_home2 have same version. So in case PROD fails, the
result would be...


SAP BOX 1(DEV SERVER)
Oracle_home1
dev instance
dev database
prod instance
prod database


Im not really interested on the set-up. they can be anything u want
to imagine. What my problem is...I can not install the patch on the
PROD server. But i have it installed on the DEV server. Can u help
me how pls.

thanks

You might want to open a TAR and asking Oracle Support for possible remedies. I personally don't see why(two seperated instances) can have any issues against being patched up.

The reason why you'd want to be interested in the setup is
--That you're not left in quandry(problems) should a situation like this occur again (there is already a patch 9206 BTW, If I'm not mistaken).
-- I would have(if I were you), in this time of soul and web searching, quickly setup similar scenario on an available server with a VMWare or an identical setup physically (make the damn bosses pay for some cheap servers or even pc's) and tried the whole patching scenario there.
--All that eventually helps when (should you not succeed in solving the problem yourself) you want to relate the problem to the support staff.

Also patching up clustered environments does indeed require a good understanding of the cluster itself. You don't have a RAC, had that been the case, your patch(correct me if I'm wrong) would have propogated among all the nodes.

So your oracle environemnt was not cluster aware(not RAC) and that's why the cluster *resources* went out of sync which is not a problem until you really experience a fail over scenario.

Long story short:
--Ask Oracle Support.
-- Please do set up a similar test environment.


Cheers

Tarry

There were many other threads from my last post. I just briefly read through. Anyway, its very possible that 2 oracle instances can have one oracle home.

Your Query is that when PRD fails over to DEV, it will be using DEV binaries or so called home?

He wants to patch his envionment to 9205. That's *all* he wants.

yes, ..... I tried installing a new oracle_home(test) using the base 9.2.0.1 and create "test" DB against it on the PROD server. I didnt invoke clustering during the oracle install, though it can see the DEV server cluster
during install. Then I patched the test oracle_home(test) it worked fine.

So my question is....If the SAP oracle_home on PROD server in clustered
with SAP oracle_home on DEV server, on which server should i run the
patch 9.2.0.5 installation first?

Thanks a lot

how do you know that 'it can see the dev server cluster' be specific now, where in the OUI do you see this? and for what its worthm i always patch my rac or ops clusters 'node by node'. That way I eliminate and kludge factor with the binaries being pushed and updated across the nodes.

Thanks stmongo, i hope Im on the right track now....I just get
confused with this cluster set-up. Let me ask about ur rac cluster set-up. when u say ....
"and for what its worthm i always patch my rac or ops clusters 'node by node'. That way I eliminate and kludge factor with the binaries being pushed and updated across the nodes."

1. Is there other way of patching aside from node by node?
2. Do u mean mean that you go to each node physically, load the,
patchset, run OUI and install the patch?

Thanks

Quote:

---

Originally Posted by yxez

Thanks stmongo, i hope Im on the right track now....I just get
confused with this cluster set-up. Let me ask about ur rac cluster set-up. when u say ....
"and for what its worthm i always patch my rac or ops clusters 'node by node'. That way I eliminate and kludge factor with the binaries being pushed and updated across the nodes."

1. Is there other way of patching aside from node by node?
2. Do u mean mean that you go to each node physically, load the,
patchset, run OUI and install the patch?

Thanks

---

1. if you are running RAC when you invoke the OUI it will list all available nodes and give you the option to patch them all
2. yep yep

thanks,

When I invoke OUI in DEV server i saw two nodes the dev and the prod.
so I patch the node firsts...and it was successful.

But in #2 above....u said that I have to go to each server physically to
patch it...so i login to the PROD server and run OUI...i choose the prod
oracle home ...OUI also tells me that prod home was a cluster set-up
but i didnt see the dev node...just the prod node. Unlike on the DEV server
i can see both nodes in OUI. So i continue to run the patch on thr prod
node but it is here where i encounter the error "There is no patch the needs
to be applied.

Dont you think i should run the patch on the DEV server to patch both the
prod and dev node? This means i am physically on 1 server login pathcing
all nodes/server.

Thanks a lot