Skip to content
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
194 lines (145 sloc) 13.1 KB

jta-crash-rec: Example of JTA Crash Recovery

The jta-crash-rec quickstart uses JTA and Byteman to show how to code distributed (XA) transactions in order to preserve ACID properties on server crash.

What is it?

The jta-crash-rec quickstart demonstrates how to code distributed or XA (eXtended Architecture) transactions so that the ACID properties are preserved across participating resources deployed to {productNameFull} after a server crash. An XA transaction is one in which multiple resources, such as MDBs and databases, participate within the same transaction. It ensures all operations are performed as a single entity of work. ACID is a set of 4 properties that guarantee the resources are processed in the following manner:

  • Atomic - if any part of the transaction fails, all resources remain unchanged.

  • Consistent - the state will be consistent across resources after a commit

  • Isolated - the execution of the transaction for each resource is isolated from each others

  • Durable - the data will persist after the transaction is committed

This quickstart shows how to atomically update multiple resources within one transaction. It updates a relational database table using JPA and sends a message using JMS. This type of paired updates to two different resources are called XA transactions and are defined by the Jakarta EE JTA specification JSR-907.

The relational database table in this example contains two columns that represent a key / value pair. The application presents an HTML form containing two input text boxes and allows you to create, update, delete or list these pairs. When you add or update a key / value pair, the quickstart starts a transaction, updates the database table, produces a JMS message containing the update, and then commits the transaction. If all goes well, eventually the consumer gets the message and generates a database update, setting the value corresponding to the key to something that indicates it was changed by the message consumer.

In this example, you halt the {productName} server in the middle of an XA transaction after the database modification has been committed, but before the JMS producer is committed. You can verify that the transaction was started, then restart the {productName} server to complete the transaction. You then verify that everything is in a consistent state.

{productName} ships with H2, an in-memory database written in Java. In this example, we use H2 for the database. Although H2 XA support is not recommended for production systems, the example does illustrate the general steps you need to perform for any datasource vendor. This example provides its own H2 XA datasource configuration. It is defined in the jta-crash-rec-ds.xml file in the WEB-INF folder of the WAR archive.

Download and Configure Byteman

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

Configure the Server

NOTE: The Byteman scripts only work in JTA mode. They do not work in JTS mode. If you have configured the server for a quickstart that uses JTS, you must follow the quickstart instructions to remove the JTS configuration from the {productName} server before making the following changes. Otherwise Byteman will not halt the server.

Access the Application

The application will be running at the following URL: http://localhost:8080/{artifactId}/XA.

Test the Application

  1. When you access the application, you will find a web page containing two html input boxes for adding key / value pairs to a database. Instructions for using the application are shown at the top of the application web page.

  2. When you add a new key / value pair, the change is committed to the database and a JMS message sent. The message consumer then updates the newly inserted row by appending the text updated via JMS to the value. Since the consumer updates the row asynchronously, you may need to click Refresh Table to see the text added to the key / value pair you previously entered.

  3. When an XA transaction is committed, the application server completes the transaction in two phases.

    • In phase 1 each of the resources, in this example the database and the JMS message producer, are asked to prepare to commit any changes made during the transaction.

    • If all resources vote to commit then the application server starts phase 2 in which it tells each resource to commit those changes.

    • The added complexity is to cope with failures, especially failures that occur during phase 2. Some failure modes require cooperation between the application server and the resources in order to guarantee that any pending changes are recovered.

  4. To demonstrate XA recovery, you must enable the Byteman tool to terminate the application server while phase 2 is running as follows:

    • Stop the {productName} server.

    • Follow the instructions here to clear the transaction objectstore remaining from any previous tests: Clear the Transaction ObjectStore

    • The following line of text must be appended to the server configuration file using the instructions located here: Use Byteman to Halt the Application

      JAVA_OPTS="-javaagent:/BYTEMAN_HOME/lib/byteman.jar=script:/QUICKSTART_HOME/jta-crash-rec/src/main/scripts/xa.btm ${JAVA_OPTS}"

      For Windows, append the following line.

      JAVA_OPTS=%JAVA_OPTS% -javaagent:C:BYTEMAN_HOME\lib\byteman.jar=script:C:__QUICKSTART_HOME__\jta-crash-rec\src\main\scripts\xa.btm %JAVA_OPTS%`
    • Start the {productName} server with the standalone full profile as described above.

  5. Once you complete step 4, you are ready to create a recovery record. Go to the application URL http://localhost:8080/{artifactId}/XA and insert another row into the database. At this point, Byteman halts the application server.

  6. If you want to verify the database insert was committed but that message delivery is still pending, you can use an SQL client such as the H2 database console tool. Issue a query to show that the value is present but does not contain the message added by the consumer (updated via JMS). Here is how you can do it using H2:

    • Start the H2 console by typing:

      $ java -cp {jbossHomeName}/modules/system/layers/base/com/h2database/h2/main/h2*.jar
    • Log in:

      Database URL: jdbc:h2:file:~/jta-crash-rec-quickstart
      User name:    sa
      Password:     sa
    • The console is available at the url http://localhost:8082. If you receive an error such as Exception opening port "8082" it is most likely because some other application has that port open. You will need to find which application is using the port and close it.

    • Once you are logged in enter the following query to see that the pair you entered is present but does not contain "updated via JMS".

      select * from kvpair
    • Log out of the H2 console and make sure you close the terminal. H2 is limited to one connection and the application will need it from this point forward.

    • If you are using the default file based transaction logging store, there will be a record in the file system corresponding to the pending transaction.

      • Open a terminal and navigate to the {jbossHomeName} directory

      • List the contents of the following directory:

        $ ls {jbossHomeName}/standalone/data/tx-object-store/ShadowNoFileLockStore/defaultStore/StateManager/BasicAction/TwoPhaseCoordinator/AtomicAction/
      • An example of a logging record file name is:

      • After recovery, log records are normally deleted automatically. However, logs may remain in the case where the Transaction Manager ™ commit request was received and acted upon by a resource, but the TM crashed before it had time to clean up the logs of that resource.

  7. To observe XA recovery

    • Stop the H2 console and exit the terminal to close the database connections. Otherwise, you may see messages like the following when you start your server:

      Database may be already in use: "Locked by another process"
    • Disable the Byteman script by restoring the backup server configuration file.

    • Start the {productName} server with the standalone full profile as described above.

    • Load the web interface to the application

    • By the time the {productName} server is ready, the transaction should have recovered.

    • A message is printed on the {productName} server console when the consumer has completed the update. Look for a line that reads:

      JTA Crash Record Quickstart: key value pair updated via JMS
    • Check that the row you inserted in step 4 now contains the text updated via JMS, showing that the JMS message was recovered successfully. Use the application URL to perform this check.

    • You will most likely see the following messages in the server log.

      WARN  [com.arjuna.ats.jta] (Periodic Recovery) ARJUNA016037: Could not find new XAResource to use for recovering non-serializable XAResource XAResourceRecord < resource:null, txid:< formatId=131077, gtrid_length=29, bqual_length=36, tx_uid=0:ffff7f000001:1040a11d:534ede43:1c, node_name=1, branch_uid=0:ffff7f000001:1040a11d:534ede43:20, subordinatenodename=null, eis_name=java:jboss/datasources/JTACrashRecQuickstartDS >, heuristic: TwoPhaseOutcome.FINISH_OK, product: H2/1.3.168-redhat-2 (2012-07-13), jndiName: java:jboss/datasources/JTACrashRecQuickstartDS com.arjuna.ats.internal.jta.resources.arjunacore.XAResourceRecord@788f0ec1 >
      WARN  [com.arjuna.ats.jta] (Periodic Recovery) ARJUNA016038: No XAResource to recover < formatId=131077, gtrid_length=29, bqual_length=36, tx_uid=0:ffff7f000001:1040a11d:534ede43:1c, node_name=1, branch_uid=0:ffff7f000001:1040a11d:534ede43:20, subordinatenodename=null, eis_name=java:jboss/datasources/JTACrashRecQuickstartDS >

      This is normal. What actually happened is that the first resource, JTACrashRecQuickstartDS, committed before the {productName} server was halted to insert the recovery record. The transaction logs are only updated/deleted after the outcome of the transaction is determined. If the transaction manager did update the log as each participant (database and JMS queue) completed then throughput would suffer. Notice you do not get a similar message for the JMS resource since that is the resource that recovered and the log record was updated to reflect this change. You need to manually remove the record for the first participant if you know which one is which or, if you are using the community version of the ${} server, then you can also inspect the transaction logs using a JMX browser. For the demo it is simplest to delete the records from the file system, however, be wary of doing this on a production system.

  8. 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!

Server Log: Expected Warnings and Errors

You will see the following warnings in the server log. You can ignore these warnings.

WFLYJCA0091: -ds.xml file deployments are deprecated. Support may be removed in a future version.
HHH000431: Unable to determine H2 database version, certain features may not work
Within {JBDSProductName}, make sure you define a server runtime environment that uses the standalone-full.xml configuration file.
You can’t perform that action at this time.