Dirty connections from pool? via ORA-03113

Question

hi,
i've a fairly standard 3 tier setup with CMP beans and an Oracle server.
Its WL6.1 SP4 talking to Oracle 9.0.1 using jdriver. As part of a
failover test scenario I do the following;
startup DB & weblogic
run test to create some data via CMP - the test also uses some non bean
code to do direct SQL based on a connection retrieved form the WL pool
(new InitialContext, JNDI lookup etc)
test passes and all is good
shutdown and restart db leaving WL in place
run the test again and code that only uses EJB and CMP sucessfully
writes to the DB, but the app level SQL code that retrieves the
connection form the pool manages to create a conn and a statement but
the executeQuery throws the following;
java.sql.SQLException: ORA-03113: end-of-file on communication channel
at weblogic.db.oci.OciCursor.getCDAException(OciCursor.java:240)
at weblogic.jdbc.oci.Statement.private_parse(Statement.java:826)
at
weblogic.jdbc.oci.Connection.prepareStatement(Connection.java:733)
at
weblogic.jdbc.pool.Connection.prepareStatement(Connection.java:305)
at
weblogic.jdbc.rmi.internal.ConnectionImpl.prepareStatement(ConnectionImpl.java:135)
at
weblogic.jdbc.rmi.internal.ConnectionImpl_WLSkel.invoke(Unknown Source)
at
weblogic.rmi.internal.BasicServerRef.invoke(BasicServerRef.java:306)
at
weblogic.rmi.internal.BasicServerRef.handleRequest(BasicServerRef.java:275)
at
weblogic.rmi.internal.BasicExecuteRequest.execute(BasicExecuteRequest.java:22)
at weblogic.kernel.ExecuteThread.execute(ExecuteThread.java:139)
at weblogic.kernel.ExecuteThread.run(ExecuteThread.java:120)
I'm guessing that the socket in the pooled connection is bad due to the
DB restart. The only reason I can see the EJB code working is if the CMP
code traps the error and retries somehow. This feels really bad as that
means the responsibility to manage the connections state is above the
pool level rendering it useless to me.
To recap I dont understand why the EJB/CMP code works and the raw SQL
over a connection retrieved from the pool fails.
Any help much appreicated.
-s

Answer 1

Manish Vazirani wrote:
Joe,
Thats what i have observed too. However, i had written a test standalone
java client which created global DS & Conn objects in its
init/constructor... and then in an infinite while loop, i was using a
PreparedStatement to execute a "select * from dual" dummy query every 60
seconds using the "cached" global connection object. In between i : a)
Recycled/Bounced the DBMS and b) Invoked the weblogic.Admin CLI command
to explicitly reset the conn pool. And this DID work ! even though the
conn was held by app all along and NOT released back to the pool during the
recycling. I had the test table, test on reserve, and prepared statement
cache settings enabled on the pool. Does this mean that the PS cache was
sorta .. fooling me and it wasnt a fresh DB lookup each time after we
recycled ?I believe the refresh doesn't change the reserved-unreserved status of the
pooled connections, but it should reset the statement cache of the connection.
This is interesting. If the prepared statement you're using was not cached, it
would have been closed, but I'm wondering if your long-running (holds
connections across client-WLS calls) may cause a reference to a defunct
statement to be retained...
Joe
>
>
thanks,
manish
"Joseph Weinstein" <[email protected]> wrote in message
news:[email protected]...
Stephen Mc Namara wrote:
hi,
i've a fairly standard 3 tier setup with CMP beans and an Oracle server.
Its WL6.1 SP4 talking to Oracle 9.0.1 using jdriver. As part of a
failover test scenario I do the following;Hi. Is your pool set up to test connections at reserve time? Are theconnections/
statements that failed held by the application during the cycling of theDBMS?
The pool should replace bad connections and their cached preparedstatements
when the connection is tested at reserve time, but the pool will neverintercede
in the condition of a connection while it is in application hands.
Joe
startup DB & weblogic
run test to create some data via CMP - the test also uses some non bean
code to do direct SQL based on a connection retrieved form the WL pool
(new InitialContext, JNDI lookup etc)
test passes and all is good
shutdown and restart db leaving WL in place
run the test again and code that only uses EJB and CMP sucessfully
writes to the DB, but the app level SQL code that retrieves the
connection form the pool manages to create a conn and a statement but
the executeQuery throws the following;
java.sql.SQLException: ORA-03113: end-of-file on communication channel
at
weblogic.db.oci.OciCursor.getCDAException(OciCursor.java:240)
atweblogic.jdbc.oci.Statement.private_parse(Statement.java:826)
at
weblogic.jdbc.oci.Connection.prepareStatement(Connection.java:733)
at
weblogic.jdbc.pool.Connection.prepareStatement(Connection.java:305)
at
weblogic.jdbc.rmi.internal.ConnectionImpl.prepareStatement(ConnectionImpl.ja
va:135)
at
weblogic.jdbc.rmi.internal.ConnectionImpl_WLSkel.invoke(Unknown Source)
at
weblogic.rmi.internal.BasicServerRef.invoke(BasicServerRef.java:306)
at
weblogic.rmi.internal.BasicServerRef.handleRequest(BasicServerRef.java:275)
at
weblogic.rmi.internal.BasicExecuteRequest.execute(BasicExecuteRequest.java:2
2)
atweblogic.kernel.ExecuteThread.execute(ExecuteThread.java:139)
at weblogic.kernel.ExecuteThread.run(ExecuteThread.java:120)
I'm guessing that the socket in the pooled connection is bad due to the
DB restart. The only reason I can see the EJB code working is if the CMP
code traps the error and retries somehow. This feels really bad as that
means the responsibility to manage the connections state is above the
pool level rendering it useless to me.
To recap I dont understand why the EJB/CMP code works and the raw SQL
over a connection retrieved from the pool fails.
Any help much appreicated.
-s