On Jun 14, 2007, at 9:03 AM, Piers Geyman wrote:
Hi David,
More background info:
I can connect to RAC successfully using the oracle tranQL Resource
Adapter,
but this method does not allow the ability to use the more powerful
features
of the RAC environment, namely Fast Connection Failover (FCF). With
FCF the
Oracle Datasource Connection pool
(oracle.jdbc.pool.OracleDataSourceFactory)
is used to obtain and manage DataSources. The
OracleDataSourceFactory uses
the Oracle Notification Service (ONS) to subscribe to RAC notification
events to help it manage its pool of connections. When a RAC node
goes down
the RAC will send a notification via ONS to any subscribers to let
them know
that the Node is no longer available. The OracleDataSourceFactory will
receive this event and then remove any of the connections to that
Node in
its pool instantly, so that any new connections requested will be
valid.
This means that the application will get valid connections if a
node goes
down, rather than getting connections that are invalid and having
handle the
event - hence Fast Connection Failover.
The following document describes most of this:
http://www.oracle.com/technology/products/database/clustering/pdf/
twpracwkldmgmt.pdf
We have some projects that do not use JTA, do you have any examples
of how
to create this gbean?
Newer projects we are working on do require jta though, so a "proper"
solution would be better.
I'm somewhat mystified why Oracle doesn't expose or advertise how to
use this stuff in app servers other than their own. You might poke
around in their jars to see if they actually have a
ManagedConnectionFactory or ResourceAdapter hidden in there.
I think that the main functions hidden in the driver can be exposed
through a j2ca resource adapter pretty easily:
-- adapter has to track all the connections itself and keep track of
which ones go to which oracle cluster member. This should probably
be done in a ResourceAdapter instance, since ManagedConnectionFactory
instances might be created over and over.
-- on notification that a node is down, adapter sends a
ConnectionErrorOccurred notification to all the connections attached
to that cluster member. This causes the app server's connection
manager to remove them from the pool and destroy them right away.
It's likely to confuse applications that are currently using these
connections, but they would have gotten equally confused when the
connection stopped working anyway.
-- the service level information needs to get fed into the
ManagedConnectionFactory so when the ConnectionManager attempts to
match connections the MCF can select one from a low-load cluster member.
The only missing part I see compared to the advertised features is
automatic creation of idle connections to a low-load cluster member.
So I think it should be possible to extend the tranql oracle adapter
to deal with all these RAC features.
--------------------
My memories are slowly coming back to me.... when Jeremy Boynes
originally implemented the tranql oracle adapter we ran into this
"oracle wants to do the pooling" feature and added a no-pooling
option to our connection manager to support it. I'm not sure if/when
anyone has tried it recently but specifying pooling with <no-pool/>
should make our connection manager not have a pool, letting oracle
deal with pooling however it wants. I don't know if the tranql
adapter is using the correct oracle class to get the RAC-aware
datasources you need, but if not it should be a simple change. Again
I don't understand the configuration involved but if you need to set
up different config-properties this is pretty easy to do.
I think that at least short term this is the best option. I don't
know how you'd get xa access with this approach, but if you can
locate an XADataSource or XAConnection in the oracle classes this too
should be fairly simple.
--------------------
Finally if using the tranql connector with no pooling doesn't work I
think you can write a gbean something like this:
public class OracleFoo implements ConnectionFactorySource {
private Object ds;
public OracleFoo() {
//create the oracle datasource
ds = ???;
}
public Object $getResource() {
return ds;
}
public static final GBeanInfo GBEAN_INFO;
static {
GBeanInfoFactory infoFactory = GBeanInfoFactory.createStatic
(OracleFoo.class, "JCAManagedConnectionFactory");
GBEAN_INFO = infoFactory.createGBeanInfo();
}
public static GBeanInfo getGBeanInfo() {
return GBEAN_INFO;
}
}
The important parts are "implements ConnectionFactorySource" and
"JCAManagedConnectionFactory" which is also available as
NameFactory.JCA_MANAGED_CONNECTION_FACTORY
Hope this helps
david jencks
Regards
Piers
djencks wrote:
On Jun 14, 2007, at 7:29 AM, Piers Geyman wrote:
Hi,
I can get a connection running to Oracle just using the tranql
connector
that uses the tranQL/geronimo connection pooling, but I need to use
the
Oracle Connection Pooling as I am using Oracle RAC and oracle
needs to
manage the connections in the pool based on information it receives
from the
Oracle RAC.
Do I need to develop my own Resource Adapter to be able to have
database
connection served by the Oracle Connection Factory Pool
(oracle.jdbc.pool.OracleDataSourceFactory), or am I missing some
obvious
configuration of geronimo/tranQL that would allow me to specify the
Oracle
class (oracle.jdbc.pool.OracleDataSourceFactory) as my connection
pool?
At the moment there's no way to have oracle manage a connection pool,
and it doesn't fit into the idea of the connector architecture. I
looked for some RAC documentation and it looks like there's a way for
weblogic to do the pooling at least for oracle 10g
(bestpracticesforxaandrac.pdf, p.6). That makes me suspect that
there is some way to use RAC through a resource adapter. For
instance, does oracle supply an XADataSource?
Do you know what the oracle connection pooling does? Why is this
something that would be difficult for the connection manager to do?
If there's really no way to avoid Oracle managing the connections it
should still be fairly easy to write a modified ConnectionManager
that doesn't do pooling but delegates to oracle. If you don't need
jta transactions at all there's probably an even simpler way to do
this by writing a gbean that creates the datasource, and has a method
called $getResource that returns it. If you then name this gbean
correctly it should fit into our jndi lookup mechanism.
Is there publicly available documentation on how to use RAC in a java
enterprise environment, e.g. javadoc or some indication of what
classes are available and for what purposes?
thanks
david jencks
Thanks and regards
Piers
Lin Sun-2 wrote:
Hi, I tried this with Geronimo 1.1 w/ daytrader. Basically I
need to
copy the oracle tranql rar file and the oracle jar file into my
repository, and specify the dependency of the oracle jar file
somewhere
in the plan -
<dep:dependency>
<dep:groupId>oracle</dep:groupId>
<dep:artifactId>classes12</dep:artifactId>
<dep:version>10g</dep:version>
<dep:type>jar</dep:type>
</dep:dependency>
and develop a resource adapter plan like below (below is an
application
scoped connection pool plan, but it can also be a server wide
plan).
<ext-module>
<connector>TradeDataSource</connector>
<external-path>tranql/tranql-connector-oracle-xa/1.1/rar</external-
path>
<connector
xmlns="http://geronimo.apache.org/xml/ns/j2ee/connector-1.1";>
<resourceadapter>
<outbound-resourceadapter>
<connection-definition>
<connectionfactory-interface>javax.sql.DataSource</
connectionfactory-interface>
<connectiondefinition-instance>
<name>jdbc/TradeDataSource</name>
<config-property-setting
name="UserName">trade</config-property-setting>
<config-property-setting
name="Password">trade</config-property-setting>
<config-property-setting
name="DatabaseName">tradedb</config-property-setting>
<config-property-setting
name="DataSourceName">TradeDataSource</config-property-setting>
<config-property-setting
name="ServerName">linsun2</config-property-setting>
<config-property-setting
name="PortNumber">1521</config-property-setting>
<config-property-setting
name="DriverType">thin</config-property-setting>
<connectionmanager>
<xa-transaction>
<transaction-caching/>
</xa-transaction>
<single-pool>
<max-size>10</max-size>
<min-size>0</min-size>
<blocking-timeout-milliseconds>5000</blocking-timeout-milliseconds>
<idle-timeout-minutes>30</idle-timeout-minutes>
<match-one/>
</single-pool>
</connectionmanager>
</connectiondefinition-instance>
</connection-definition>
</outbound-resourceadapter>
</resourceadapter>
</connector>
</ext-module>
The oracle tranql rar file isn't in the geronimo assembly,
unfortunatelly. So you'll have to get it from the Tranql project
(http://tranql.codehaus.org/Download). You can download the binary
(that is a bit old than what I used) or build it yourself. The
ra.xml
inside of the rar file contains what configuration parameters
you can
specify in the connection pool plan.
HTH, Lin
Piers Geyman wrote:
Hi,
I am trying to migrate from tomcat to geronimo and I am having
trouble
being
able to configure my Oracle RAC datasource in geronimo to use the
Oracle
Driver's connection pooling rather than the default geonimo
pooling.
In tomcat I configured my datasource as follows, which used the
Oracle
Datasource connection pool rather than DBCP, but I cannot see how
I can
do
the same in Geronimo.
<Resource
auth="Container"
description="My Oracle Datasource"
name="jdbc/myOracleDataSource"
type="oracle.jdbc.pool.OracleDataSource"
factory="oracle.jdbc.pool.OracleDataSourceFactory"
driverClassName="oracle.jdbc.OracleDriver"
url="jdbc:oracle:thin:@(DESCRIPTION=(ADDRESS=(PROTOCOL=TCP)
(HOST=oraclehost1)(PORT=1529))(ADDRESS=(PROTOCOL=TCP)
(HOST=oraclehost2)(PORT=1529))(LOAD_BALANCE=yes)(CONNECT_DATA=
(SERVER=DEDICATED)(SERVICE_NAME=TESTDB)))"
user="devapp"
password="devapp"
maxIdle="20"
minIdle="10"
maxWait="5000"
maxActive="100"
connectionCachingEnabled="true"
fastConnectionFailoverEnabled="true"
connectionCacheName="myOracleDatasourceCache"
onsConfigStr="nodes=oraclehost1:6200,oraclehost2:6200"
connectionProperties="oracle.jdbc.ReadTimeout=30000"/>
I see that other connection pools to the databases seem to use
tranQL,
but
there seems to be no documentation for this product.
Any help would be appreciated.
Thanks and regards
Piers
--
View this message in context: http://www.nabble.com/How-to-
configure-a-datasource-to-use-OracleDataSource-for-Datasource-
pooling-tf3921766s134.html#a11121302
Sent from the Apache Geronimo - Users mailing list archive at
Nabble.com.
--
View this message in context: http://www.nabble.com/How-to-
configure-a-datasource-to-use-OracleDataSource-for-Datasource-
pooling-tf3921766s134.html#a11123189
Sent from the Apache Geronimo - Users mailing list archive at
Nabble.com.
