Several weeks ago a customer launch his new WebSphere Portal after migrating from WPS 7.0. We change the database server from Db2 to Oracle RAC but…
WebSphere Portal has support for Oracle RAC. We ask the Oracle DBA team for the connection URL and he sent an URL like this one:
jdbc:oracle:thin:@(DESCRIPTION=(ADDRESS=(PROTOCOL=TCP)(HOST=HOSTNAME)(PORT=1521))(CONNECT_DATA=(SERVICE_NAME=SERVICENAME))
The url for Oracle RAC is like this one:
jdbc:oracle:thin:@(DESCRIPTION=(ADDRESS_LIST=(ADDRESS=(PROTOCOL=TCP)(HOST=PRIMARY_NODE_HOSTNAME)(PORT=1521))(ADDRESS=(PROTOCOL=TCP)(HOST=SECONDARY_NODE_HOSTNAME)(PORT=1521)))(CONNECT_DATA=(SERVICE_NAME=DATABASE_SERVICENAME))).
Using Oracle One Node we have only one active instance. From Oracle i found this description for Oracle RAC One Node
“The main difference between Oracle RAC One Node and Oracle RAC is that with an Oracle RAC One Node
Database, there is only one instance running at a time under normal operation. Should this one instance be
impacted by unplanned downtime, there is a short period of time during which no database instance is running.
The stack used for Oracle RAC One Node ensures that the database service will be resumed as fast as possible,
but technically and as with any other failover cluster solution, there is a moment of time, during which no instance of
the Oracle RAC One Node Database is available to perform any work. This would not be the case in a multi-node
Oracle RAC system, because multiple instances are already running concurrently, accessing the same database.
For planned downtime Oracle RAC One Node provides a solution superior to any other failover solution on the
market; “Online Database Relocation”. Online Database Relocation is an Oracle RAC One Node-specific feature
that enables an Oracle RAC One Node database instance to be relocated to another server in the cluster without
interrupting the database service. In an Oracle RAC database this feature is not required, as databases instances
are typically running on multiple servers in the cluster already. Relocating a running Oracle RAC database instance
is unnecessary.”
Until last week everithing was fine until the Oracle node fails and switch to the second instance. The switch took more than 20 minutes and the entire WebSphere Portal hung.
We must restart all cluster nodes on WPS. The 20 minutes time was not spected for the Oracle team and they found a bug and apply a fix provided by Oracle. The “normal” time = 3 minutes. This is more than the WPS can support.
We open an PMR asking IBM how the WPS can recover from this situation without a reboot and the answer was:
“Oracle RAC One Node” is not explicitly supported by Portal. However I suppose this is transparent for Portal in the sense
that it should operate as usual.”
Portal works well until a failure.
“To your question regarding recovery of Portal if the database is available again: If the database is available again, Portal would connect to the database. However, Portal makes heavy use of caching. If the database is not available for a longer time, it can be that during error handling of requests coming in during this time, some Portal caches come into a state where recovery is difficult or even impossible after the database is available again.
If you are Portal Cache Viewer portlet is available on your system and if this portlet is still available after the database is operational
again, you could try to clear all caches.”
And finally IBM told that this is an unsupported setup.
“Our general approach regarding unsupported environments is that if a customer reports problems when using Portal in an unsupported setup, we will try to fix this problem if the problem is not related to the unsupported setup. This is especially true if the problem can be reproduced by the customer on a supported environment. “
Conclusion: Oracle RAC is for HA and supported by IBM. Oracle RAC One Node is for FAIL OVER and is not supported by IBM.
