3 Streams Conflict Resolution

Avoid Uniqueness Conflicts in a Streams Environment

You can avoid uniqueness conflicts by ensuring that each database uses unique identifiers for shared data. There are three ways to ensure unique identifiers at all databases in a Streams environment.

One way is to construct a unique identifier by executing the following select statement:

SELECT SYS_GUID() OID FROM DUAL;

This SQL operator returns a 16-byte globally unique identifier. This value is based on an algorithm that uses time, date, and the computer identifier to generate a globally unique identifier. The globally unique identifier appears in a format similar to the following:

A741C791252B3EA0E034080020AE3E0A

Another way to avoid uniqueness conflicts is to create a sequence at each of the databases that shares data and concatenate the database name (or other globally unique value) with the local sequence. This approach helps to avoid any duplicate sequence values and helps to prevent uniqueness conflicts.

Finally, you can create a customized sequence at each of the databases that shares data so that no two databases can generate the same value. You can accomplish this by using a combination of starting, incrementing, and maximum values in the CREATE SEQUENCE statement. For example, you might configure the following sequences:

Using a similar approach, you can define different ranges for each database by specifying a START WITH and MAXVALUE that would produce a unique range for each database.

Avoid Delete Conflicts in a Streams Environment

Always avoid delete conflicts in shared data environments. In general, applications that operate within a shared ownership data model should not delete rows using DELETE statements. Instead, applications should mark rows for deletion and then configure the system to purge logically deleted rows periodically.

Avoid Update Conflicts in a Streams Environment

After trying to eliminate the possibility of uniqueness and delete conflicts, you should also try to limit the number of possible update conflicts. However, in a shared ownership data model, update conflicts cannot be avoided in all cases. If you cannot avoid all update conflicts, then you must understand the types of conflicts possible and configure the system to resolve them if they occur.

Types of Prebuilt Update Conflict Handlers

Oracle provides the following types of prebuilt update conflict handlers for a Streams environment: OVERWRITE, DISCARD, MAXIMUM, and MINIMUM.

The description for each type of handler later in this section refers to the following conflict scenario:

1. The following update is made at the dbs1.net source database:

   UPDATE hr.employees SET salary = 4900 WHERE employee_id = 200;
   COMMIT;
   

   This update changes the salary for employee 200 from 4400 to 4900.

2. At nearly the same time, the following update is made at the dbs2.net destination database:

   UPDATE hr.employees SET salary = 5000 WHERE employee_id = 200;
   COMMIT;
   

3. A capture process captures the update at the dbs1.net source database and puts the resulting row LCR in a queue.

4. A propagation propagates the row LCR from the queue at dbs1.net to a queue at dbs2.net.

5. An apply process at dbs2.net attempts to apply the row LCR to the hr.employees table but encounters a conflict because the salary value at dbs2.net is 5000, which does not match the old value for the salary in the row LCR (4400).

The following sections describe each prebuilt conflict handler and explain how the handler resolves this conflict.

CONNECT hr/hr

ALTER TABLE hr.employees ADD (time TIMESTAMP WITH TIME ZONE);

CREATE OR REPLACE TRIGGER hr.insert_time_employees
BEFORE 
  INSERT OR UPDATE OF job_id, salary, commission_pct ON hr.employees
FOR EACH ROW
BEGIN
   -- Consider time synchronization problems. The previous update to this 
   -- row might have originated from a site with a clock time ahead of the 
   -- local clock time.
   IF :OLD.TIME IS NULL OR :OLD.TIME < SYSTIMESTAMP THEN
     :NEW.TIME := SYSTIMESTAMP;
   ELSE
     :NEW.TIME := :OLD.TIME + 1 / 86400;
   END IF;
END;
/

Column Lists

Each time you specify a prebuilt update conflict handler for a table, you must specify a column list. A column list is a list of columns for which the update conflict handler is called. If an update conflict occurs for one or more of the columns in the list when an apply process tries to apply a row LCR, then the update conflict handler is called to resolve the conflict. The update conflict handler is not called if a conflict occurs only in columns that are not in the list. The scope of conflict resolution is a single column list on a single row LCR.

You can specify more than one update conflict handler for a particular table, but the same column cannot be in more than one column list. For example, suppose you specify two prebuilt update conflict handlers on hr.employees table:

• The first update conflict handler has the following columns in its column list: salary and commission_pct.

• The second update conflict handler has the following columns in its column list: job_id and department_id.

Also, assume that no other conflict handlers exist for this table. In this case, if a conflict occurs for the salary column when an apply process tries to apply a row LCR, then the first update conflict handler is called to resolve the conflict. If, however, a conflict occurs for the department_id column, then the second update conflict handler is called to resolve the conflict. If a conflict occurs for a column that is not in a column list for any conflict handler, then no conflict handler is called, and an error results. In this example, if a conflict occurs for the manager_id column in the hr.employees table, then an error results. If conflicts occur in more than one column list when a row LCR is being applied, and there are no conflicts in any columns that are not in a column list, then the appropriate update conflict handler is invoked for each column list with a conflict.

Column lists enable you to use different handlers to resolve conflicts for different types of data. For example, numeric data is often suited for a maximum or minimum conflict handler, while an overwrite or discard conflict handler might be preferred for character data.

If a conflict occurs in a column that is not in a column list, then the error handler for the specific operation on the table attempts to resolve the conflict. If the error handler cannot resolve the conflict, or if there is no such error handler, then the transaction that caused the conflict is moved to the error queue.

Also, if a conflict occurs for a column in a column list that uses either the OVERWRITE, MAXIMUM, or MINIMUM prebuilt handler, and the row LCR does not contain all of the columns in this column list, then the conflict cannot be resolved because all of the values are not available. In this case, the transaction that caused the conflict is moved to the error queue. If the column list uses the DISCARD prebuilt method, then the row LCR is discarded and no error results, even if the row LCR does not contain all of the columns in this column list.

A conditional supplemental log group must be specified for the columns specified in a column list if more than one column at the source database affects the column list at the destination database. Supplemental logging is specified at the source database and adds additional information to the LCR, which is needed to resolve conflicts properly. Typically, a conditional supplemental log group must be specified for the columns in a column list if there is more than one column in the column list, but not if there is only one column in the column list.

However, in some cases, a conditional supplemental log group is required even if there is only one column in a column list. That is, an apply handler or custom rule-based transformation can combine multiple columns from the source database into a single column in the column list at the destination database. For example, a custom rule-based transformation can take three columns that store street, state, and postal code data from a source database and combine the data into a single address column at a destination database.

Also, in some cases, no conditional supplemental log group is required even if there is more than one column in a column list. For example, an apply handler or custom rule-based transformation can separate one address column from the source database into multiple columns that are in a column list at the destination database. A custom rule-based transformation can take an address that includes street, state, and postal code data in one address column at a source database and separate the data into three columns at a destination database.

Resolution Columns

The resolution column is the column used to identify a prebuilt update conflict handler. If you use a MAXIMUM or MINIMUM prebuilt update conflict handler, then the resolution column is also the column used to resolve the conflict. The resolution column must be one of the columns in the column list for the handler.

For example, if the salary column in the hr.employees table is specified as the resolution column for a maximum or minimum conflict handler, then the salary column is evaluated to determine whether column list values in the row LCR are applied or the destination database values for the column list are retained.

In either of the following situations involving a resolution column for a conflict, the apply process moves the transaction containing the row LCR that caused the conflict to the error queue, if the error handler cannot resolve the problem. In these cases, the conflict cannot be resolved and the values of the columns at the destination database remain unchanged:

• The new LCR value and the destination row value for the resolution column are the same (for example, if the resolution column was not the column causing the conflict).

• Either the new LCR value of the resolution column or the current value of the resolution column at the destination database is NULL.

Data Convergence

When you share data between multiple databases, and you want the data to be the same at all of these databases, then make sure you use conflict resolution handlers that cause the data to converge at all databases. If you allow changes to shared data at all of your databases, then data convergence for a table is possible only if all databases that are sharing data capture changes to the shared data and propagate these changes to all of the other databases that are sharing the data.

In such an environment, the MAXIMUM conflict resolution method can guarantee convergence only if the values in the resolution column are always increasing. A time-based resolution column meets this requirement, as long as successive timestamps on a row are distinct. The MINIMUM conflict resolution method can guarantee convergence in such an environment only if the values in the resolution column are always decreasing.