Oracle7 Server Distributed Systems Volume I: Distributed Data

Contents Index Home Previous Next

Testing Distributed Transaction Recovery Features

If you like, you can force the failure of a distributed transaction to observe RECO, automatically resolving the local portion of the transaction. Alternatively, you might be interested in forcing a distributed transaction to fail so that you can practice manually resolving in-doubt distributed transactions and observing the results.

The following sections describes the features available and the steps necessary to perform such operations.

Forcing a Distributed Transaction to Fail

Comments can be included in the COMMENT parameter of the COMMIT statement. To intentionally induce a failure during the prepare/commit phases of a distributed transaction, include the following comment in the COMMENT parameter:

COMMIT COMMENT 'ORA-2PC-CRASH-TEST-n';

where n is one of the following integers:

n Effect
1 Crash commit point site after collect
2 Crash non-commit point site after collect
3 Crash before prepare (non-commit point site)
4 Crash after prepare (non-commit point site)
5 Crash commit point site before commit
6 Crash commit point site after commit
7 Crash non-commit point site before commit
8 Crash non-commit point site after commit
9 Crash commit point site before forget
10 Crash non-commit point site before forget
Table 5 - 1. Failure Values for the Parameter COMMENT

For example, the following statement returns the following messages if the local commit point strength is greater than the remote commit point strength and both nodes are updated:

COMMIT COMMENT 'ORA-2PC-CRASH-TEST-7';
ORA-02054: transaction #.##.## in-doubt
ORA-02059: ORA-CRASH-TEST-n in commit comment

At this point, the in-doubt distributed transaction appears in the DBA_2PC_PENDING view. If enabled, RECO automatically resolves the transaction rather quickly.

Privileges Required to Induce Prepare/Commit Phase Failures

You can induce prepare/commit phase failures via the previous comments only if the local and remote sessions have the FORCE ANY TRANSACTION system privilege. Otherwise, an error is returned if you attempt to issue a COMMIT statement with a crash comment.

The Recoverer (RECO) Background Process

The RECO background process of an Oracle7 instance automatically resolves failures involving distributed transactions. At exponentially growing time intervals, the RECO background process of a node attempts to recover the local portion of an in-doubt distributed transaction.

RECO can use an existing connection or establish a new connection to other nodes involved in the failed transaction. When a connection is established, RECO automatically resolves all in-doubt transactions. Rows corresponding to any resolved in-doubt transactions are automatically removed from each database's pending transaction table.

Disabling and Enabling RECO

The recoverer background process, RECO, can be enabled and disabled using the ALTER SYSTEM command with the ENABLE/DISABLE DISTRIBUTED RECOVERY options, respectively. For example, you might want to temporarily disable RECO to force the failure of a prepare/commit and manually resolve the in-doubt transaction. The following statement disables RECO:

ALTER SYSTEM DISABLE DISTRIBUTED RECOVERY;

Alternatively, the following statement enables RECO so that in-doubt transactions are automatically resolved:

ALTER SYSTEM ENABLE DISTRIBUTED RECOVERY;

Note: Single-process instances (for example, a PC running MS-DOS) have no separate background processes, and therefore no RECO process. Therefore, when a single-process instance that participates in a distributed system is started, distributed recovery must be manually enabled using the statement above.

Additional Information: See your Oracle operating system-specific documentation for more information about distributed transaction recovery for single-process instances.

Contents Index Home Previous Next