Just a stupid error. Posting it so that someone else googling for the same thing can get a clue.
An ASM instance running with default parameters (no pfile, no spfile). Updated spfile for the instance with asmcmd spset command and bounced crs. After reboot also, it still wasn’t using spfile. Got puzzled and checked GPnP settings again. All looked good. Then in alert log came across this
ERROR: SPFile in diskgroup <> does not match the specified spfile +DATA/asm/asmparameterfile/registry.253.769187275
The problem was that while copying the spfile path the complete name didn’t get copied. The last character got missed. So the filename that it was looking for wasn’t there. Updating GPnP with correct filename and bouncing crs resolved the issue.
So this customer has an Exadata quarter rack and they have an IB listener configured on both DB nodes (for DB connections from a multi-racked Exalogic system). We were adding a new DB node to this rack. So just followed the standard procedure of creating users, directories etc on the new node, setting up ssh equivalence and running addNode.sh. All went fine but root.sh failed. Little looking into the logs revealed that it failed while running srvctl start listener –n <node_name>
If we manually run this command, it will immediately reveal what the problem is. It is not able to start IB listener on the new node as the IB VIP doesn’t yet exist. It could happen for any of the additional networks added.
There is a MOS note that describes this exact situation but the solution that it gives is to remove the additional listener, complete addNode.sh & root.sh and add the additional listener back. That wasn’t possible in this case. After little bit of googling I stumbled upon this post by Jeremy Schneider. His colleague solved this problem with a very simple and clever workaround. Before root.sh prepares to run srvctl start listener command, run the add VIP command from another Window . Additional network would have already got added when root.sh runs on the new node.
To be able to perform this trick, you have to have the hosts file updated with the new VIP name and IP and be ready with the command to add the VIP. While root.sh is running, it will show a message like “there is already an active cluster, restarting to join”, immediately start trying to run srvctl add vip command in another window. The moment CRS, comes up the command will succeed. Immediately after that root.sh is going to run srvctl start listener command, and this time it shouldn’t fail as the VIP is already added.
Another small mistake we made was not updating the cellip.ora on the new node before running root.sh. That caused the root.sh to fail as it couldn’t talk to ASM running on existing cell nodes. Updating cellip.ora with the existing storage node IPs fixed the problem.
So if you are filling an OEDA for Exadata deployment there are few things you should take care of. Most of the screens are self explanatory but there are some bits where one should focus little more. I am running the Aug version of it and the screenshots below are from that version.
- On the Define customer networks screen, the client network is the actual network where your data is going to flow. So typically it is going to be bonded (for high availability) and depending upon the network in your data center you have to select one out of 1/10 G copper and 10 G optical.
- If you are going to use trunk VLANs for your client network, remember to enabled it by clicking the Advanced button and then entering the relevant VLAN id.
Also if it is going to be an OVM configuration, you may want to have different VMs in different VLAN segments. It will allow you to change VLAN ids for individual VMs on the respective cluster screens like below
- If all the cores aren’t licensed remember to enable Capacity on Demand (COD) on the Identify Compute node OS screen.
- On the Define clusters screen make sure that you enter a unique (across your environment) cluster name.
- The cluster details screen captures some of the most important details like
- Whether you want to have flash cache in WriteBack mode instead of WriteThrough
- Whether you want to have a role separated install or want to install both GI and Oracle binaries with oracle user itself.
- GI & Database versions and home for binaries. Always good to leave it at the Oracle recommended values as that makes the future maintenance easy and less painful.
- Disk Group names, redundancy and the space allocation.
- Default database name and type (OLTP or DW).
Of course it is important to carefully fill the information in all the screens but the above ones are some of them which should be filled very carefully after capturing the required information from other teams, if needed.
Faced this error while querying v$asm_disk after adding new storage cell IPs to cellip.ora on DB nodes of an existing cluster on Exadata. Query ends with ORA-03113 end-of-file on communication channel and ORA-56841 is reported in $ORA_CRS_HOME/log/<hostname>/diskmon/diskmon.log. Reason in my case was that the new cell was using different subnet for IB. It was pingable from the db nodes but querying v$asm_disk wasn’t working. Changing the subnet for IB on new cell to the one on existing cells fixed the issue.
On a T5 Super Cluster (running 126.96.36.199) I was creating a cascaded standby from an already functional standby using RMAN DUPLICATE and it errored out with
ORA-01671: control file is a backup, cannot make a standby control file
A quick search reveals that it is bug 11715084 that affects most of the 11.x versions except 188.8.131.52. There is a one off patch available for most of the versions or one can install the bundle patch that includes the fix for this patch. I applied BP26 and it worked fine after that.