The jts-distributed-crash-rec
quickstart uses JTS and Byteman to demonstrate distributed crash recovery across multiple application servers.
The jts-distributed-crash-rec
quickstart demonstrates a distributed crash recovery across multiple application servers in {productNameFull}.
Crash recovery is a key feature provided by an application server and ensures an application’s data consistency, even in the presence of failure. Providing reliable crash recovery helps qualify the pedigree of an application server, distributed crash recovery even more so.
This quickstart uses the application components from the jts
quickstart. It provides a Byteman rule to inject a halt into the application server at a crucial point in the phase2 commit of the transaction. Byteman is used solely to raise an artificial fault. An IDE could also provide this simulation, although it is more complex to use an IDE for this purpose.
Apart from that, this quickstart works the same as the jts
quickstart and if the Byteman rule is left out, this quickstart is the JTS quickstart.
As an overview, the sequence of events to expect:
-
Configure and start two {productName} servers.
-
Build and deploy the two application components.
-
Open a web browser and attempt to invoice two customers as with the
jts
quickstart. -
{productName} server 1 will run through a two-phase commit (2PC), preparing the resources in {productName} server 1 and {productName} server 2. {productName} server 1 will then crash before it can call commit.
-
The user is invited to inspect the content of the transaction objectstore.
-
{productName} server 1 should be restarted. It will then recover the "invoices" delivered to the MDBs, just as it does in the
jts
quickstart.
This quickstart uses Byteman to help demonstrate crash recovery. You can find more information about Byteman here: Configure Byteman for Use with the Quickstarts
Follow the instructions here to download and configure Byteman: Download and Configure Byteman
Developers should be familiar with the concepts introduced in the following quickstarts:
Important
|
This quickstart depends on the deployment of the jts quickstart for its test. Before running this quickstart, see the jts README file for details on how to deploy it.
|
You can verify the deployment of the jts
quickstart by accessing the following URL: http://localhost:8080/jts-application-component-1/.
-
If you have not yet done so, configure the two application servers and deploy the
jts
quickstart. Follow the instructions in the jts README file. -
Configure Byteman to halt {productName} server 1
-
Stop both {productName} servers.
-
Follow the instructions here to clear the transaction objectstore remaining from any previous tests: Clear the Transaction ObjectStore
-
The following 2 lines of text must be appended to the server configuration file for server 1 only using the instructions located here: Use Byteman to Halt the Application
JAVA_OPTS="-javaagent:/BYTEMAN_HOME/lib/byteman.jar=script:/QUICKSTART_HOME/jts-distributed-crash-rec/byteman-scripts/failAfterPrepare.btm ${JAVA_OPTS}" JAVA_OPTS="-Dorg.jboss.byteman.transform.all -Djboss.modules.system.pkgs=org.jboss.byteman -Dorg.jboss.byteman.verbose=true ${JAVA_OPTS}"
NoteFor Windows, append the following line.
JAVA_OPTS=%JAVA_OPTS% -javaagent:C:BYTEMAN_HOME\lib\byteman.jar=script:C:__QUICKSTART_HOME__\jts-distributed-crash-rec\byteman-scripts\failAfterPrepare.btm %JAVA_OPTS% JAVA_OPTS=%JAVA_OPTS% -Dorg.jboss.byteman.transform.all -Djboss.modules.system.pkgs=org.jboss.byteman -Dorg.jboss.byteman.verbose=true
-
-
Start both of the {productName} servers
$ {jbossHomeName}_1/bin/standalone.sh -c standalone-full.xml -Djboss.tx.node.id=UNIQUE_NODE_ID_1 $ {jbossHomeName}_2/bin/standalone.sh -c standalone-full.xml -Djboss.tx.node.id=UNIQUE_NODE_ID_2 -Djboss.socket.binding.port-offset=100
NoteFor Windows, use the {jbossHomeName}_1\bin\standalone.bat
and{jbossHomeName}_2\bin\standalone.bat
scripts. -
Access the application at the following URL: http://localhost:8080/jts-application-component-1/
-
When you enter a name and click to "add" that customer, you will see the following in the application server 1 console:
INFO [org.jboss.ejb.client] (default task-2) JBoss EJB Client version 2.1.4.Final-redhat-1 INFO [stdout] (default task-2) Rule.execute called for Fail 2PC after prepare_0 INFO [stdout] (default task-2) HelperManager.install for helper class org.jboss.byteman.rule.helper.Helper INFO [stdout] (default task-2) calling activated() for helper class org.jboss.byteman.rule.helper.Helper INFO [stdout] (default task-2) Default helper activated INFO [stdout] (default task-2) calling installed(Fail 2PC after prepare) for helper classorg.jboss.byteman.rule.helper.Helper INFO [stdout] (default task-2) Installed rule using default helper : Fail 2PC after prepare INFO [stdout] (default task-2) Fail 2PC after prepare execute INFO [stdout] (default task-2) rule.debug{Fail 2PC after prepare} : Prepare completed INFO [stdout] (default task-2) rule.debug{Fail 2PC after prepare} : !!!killing JVM!!!
-
-
At this point, Byteman halts or crashes server 1. You should be able to view the contents of the object store for this server by typing the following in the terminal for server 1. Make sure you replace
{jbossHomeName}_1
with the path to the first server.tree {jbossHomeName}_1/standalone/data/tx-object-store
This should display:
{jbossHomeName}_1/standalone/data/tx-object-store -- ShadowNoFileLockStore -- defaultStore |-- CosTransactions | -- XAResourceRecord | -- 0_ffffc0a8013c_38e104bd_4f280cdb_1d |-- Recovery | -- FactoryContact | |-- 0_ffffc0a8013c_38e104bd_4f280cdb_17 | |-- 0_ffffc0a8013c_-671009a_4f280e7e_17 | -- 0_ffffc0a8013c_6d5d82b5_4f280a16_f |-- RecoveryCoordinator | -- 0_ffff52e38d0c_c91_4140398c_0 -- StateManager -- BasicAction -- TwoPhaseCoordinator -- ArjunaTransactionImple -- 0_ffffc0a8013c_38e104bd_4f280cdb_19
View the contents of the object store for the second server by typing the following in the terminal for server 2. Make sure you replace
{jbossHomeName}_2
with the path to the second server.tree {jbossHomeName}_2/standalone/data/tx-object-store
This should display:
{jbossHomeName}_2/standalone/data/tx-object-store -- ShadowNoFileLockStore -- defaultStore |-- CosTransactions | -- XAResourceRecord | -- 0_ffffc0a8013c_-2eb1158b_4f280ce3_1e |-- Recovery | -- FactoryContact | |-- 0_ffffc0a8013c_-2eb1158b_4f280ce3_18 | -- 0_ffffc0a8013c_4f6459f0_4f280a24_f |-- RecoveryCoordinator | -- 0_ffff52e38d0c_c91_4140398c_0 -- StateManager -- BasicAction -- TwoPhaseCoordinator -- ArjunaTransactionImple -- ServerTransaction -- 0_ffffc0a8013c_-2eb1158b_4f280ce3_1a
-
Disable the Byteman script by restoring the backup configuration file for server 1.
-
Follow the steps above to restart server 1 and wait for recovery to complete.
ImportantBy default, the recovery process checks the transactional state every two minutes, therefore it can take a while for recovery to happen. Also recovery for each server will take place at its own recovery interval. -
You will know when recovery is complete for server 2 as you will see the following in application-server-2 console:
INFO [org.jboss.ejb.client] (RequestProcessor-10) JBoss EJB Client version 2.1.2.Final INFO [class org.jboss.as.quickstarts.cmt.jts.mdb.HelloWorldMDB] (Thread-3 (group:ActiveMQ-client-global-threads-649946595)) Received Message: Created invoice for customer named: Tom
-
You will also get several stack traces in {productName} server 1 console during recovery, these are to be expected as not all resources are available at all stages of recovery.
WARN [com.arjuna.ats.jts] (Periodic Recovery) ARJUNA022223: ExtendedResourceRecord.topLevelCommit caught exception: org.omg.CORBA.OBJECT_NOT_EXIST: ----------BEGIN server-side stack trace---------- org.omg.CORBA.OBJECT_NOT_EXIST: vmcid: SUN minor code: 1004 completed: No at com.sun.corba.se.impl.logging.POASystemException.nullServant(POASystemException.java:2040) at com.sun.corba.se.impl.logging.POASystemException.nullServant(POASystemException.java:2062) at com.sun.corba.se.impl.oa.poa.POAPolicyMediatorImpl_R_AOM.internalGetServant(POAPolicyMediatorImpl_R_AOM.java:68) at com.sun.corba.se.impl.oa.poa.POAPolicyMediatorBase.getInvocationServant(POAPolicyMediatorBase.java:121) at com.sun.corba.se.impl.oa.poa.POAImpl.getInvocationServant(POAImpl.java:1634) at com.sun.corba.se.impl.protocol.CorbaServerRequestDispatcherImpl.getServant(CorbaServerRequestDispatcherImpl.java:326) at com.sun.corba.se.impl.protocol.CorbaServerRequestDispatcherImpl.getServantWithPI(CorbaServerRequestDispatcherImpl.java:360) at com.sun.corba.se.impl.protocol.CorbaServerRequestDispatcherImpl.dispatch(CorbaServerRequestDispatcherImpl.java:202) at com.sun.corba.se.impl.protocol.CorbaMessageMediatorImpl.handleRequestRequest(CorbaMessageMediatorImpl.java:1700) at com.sun.corba.se.impl.protocol.SharedCDRClientRequestDispatcherImpl.marshalingComplete(SharedCDRClientRequestDispatcherImpl.java:180) at com.sun.corba.se.impl.protocol.CorbaClientDelegateImpl.invoke(CorbaClientDelegateImpl.java:148) at org.omg.CORBA.portable.ObjectImpl._invoke(ObjectImpl.java:475) at com.arjuna.ArjunaOTS._ArjunaSubtranAwareResourceStub.commit(_ArjunaSubtranAwareResourceStub.java:124) at com.arjuna.ats.internal.jts.resources.ExtendedResourceRecord.topLevelCommit(ExtendedResourceRecord.java:502) ...
-
The easiest way to check when {productName} server 1 is recovered is to look in the object store and check that all the records are now cleaned up. The records that should be cleared are the ones in the
defaultStore/CosTransactions/XAResourceRecord
anddefaultStore/StateManager/BasicAction/TwoPhaseCoordinator/ArjunaTransactionImple
. -
Records will remain in
defaultStore/Recovery/FactoryContact
anddefaultStore/RecoveryCoordinator
for server 1 and that is to be expected. Run:tree {jbossHomeName}_1/standalone/data/tx-object-store
You should see this output:
{jbossHomeName}_1/standalone/data/tx-object-store -- ShadowNoFileLockStore -- defaultStore |-- CosTransactions | -- XAResourceRecord |-- Recovery | -- FactoryContact | |-- 0_ffffc0a8013c_38e104bd_4f280cdb_17 | |-- 0_ffffc0a8013c_-671009a_4f280e7e_17 | -- 0_ffffc0a8013c_6d5d82b5_4f280a16_f |-- RecoveryCoordinator | -- 0_ffff52e38d0c_c91_4140398c_0 -- StateManager -- BasicAction -- TwoPhaseCoordinator -- ArjunaTransactionImple
View the contents of the object store for the second server by typing the following in the terminal for server 2. Make sure you replace
{jbossHomeName}_2
with the path to the second server.tree {jbossHomeName}_2/standalone/data/tx-object-store
This should display:
{jbossHomeName}_2/standalone/data/tx-object-store -- ShadowNoFileLockStore -- defaultStore |-- CosTransactions | -- XAResourceRecord |-- Recovery | -- FactoryContact | |-- 0_ffffc0a8013c_-2eb1158b_4f280ce3_18 | -- 0_ffffc0a8013c_4f6459f0_4f280a24_f |-- RecoveryCoordinator | -- 0_ffff52e38d0c_c91_4140398c_0 -- StateManager -- BasicAction -- TwoPhaseCoordinator -- ArjunaTransactionImple -- ServerTransaction
-
-
After recovery is complete, access the application URL http://localhost:8080/jts-application-component-1/customers.jsf. The user you created should now appear in the list.
-
Do not forget to Disable the Byteman script by restoring the backup server configuration file. The Byteman rule must be removed to ensure that your application server will be able to commit 2PC transactions!