ora_migrator: Oracle to PostgreSQL migration tools

Summary: ora_migrator is a plugin for db_migrator that uses oracle_fdw to migrate an Oracle database to PostgreSQL.

Table of Contents

Introduction

ora_migrator is a plugin for db_migrator that uses oracle_fdw to migrate an Oracle database to PostgreSQL.

Please read the db_migrator documentation for usage instructions; this article only covers the setup of the plugin as well as additional features that are not covered in the general documentation.

In addition to that, ora_migrator offers a replication functionality from Oracle to PostgreSQL which can be used for almost zero down time migration from Oracle. See Replication for details.

Note that since schema names are usually in upper case in Oracle, you will need to use upper case schema names for the only_schemas parameter of the db_migrator functions.

Options

The following option can be used for db_migrate_prepare, db_migrate_mkforeign and db_migrate:

• max_long (integer, default value 32767): will be used to set the max_long option on the foreign tables. This determines the maximal length of LONG, LONG RAW and XMLTYPE columns.

Prerequisites

• You need PostgreSQL 9.5 or later.

• The oracle_fdw and db_migrator extensions must be installed.

• A foreign server must be defined for the Oracle database you want to access.

• The user who calls the create_oraviews function to create the foreign tables must have the USAGE privilege on the foreign server.

• A user mapping must exist for the user who calls the create_oraviews function.

• The Oracle user used in the user mapping must have privileges to read the following Oracle dictionary views:

  • DBA_COL_PRIVS
  • DBA_CONS_COLUMNS
  • DBA_CONSTRAINTS
  • DBA_IND_COLUMNS
  • DBA_IND_EXPRESSIONS
  • DBA_INDEXES
  • DBA_MVIEWS
  • DBA_MVIEW_LOGS
  • DBA_PROCEDURES
  • DBA_SEGMENTS
  • DBA_SEQUENCES
  • DBA_SOURCE
  • DBA_TAB_COLUMNS
  • DBA_TAB_PRIVS
  • DBA_TABLES
  • DBA_TRIGGERS
  • DBA_USERS
  • DBA_VIEWS

  You can choose to grant the user the SELECT ANY DICTIONARY system privilege instead, which includes all of the above.

• To use replication, the user must have the CREATE TABLE and CREATE TRIGGER privileges.

  To use replication for tables not owned by the Oracle user, the user must have the CREATE ANY TABLE, CREATE ANY INDEX, CREATE ANY TRIGGER, DROP ANY TABLE, DROP ANY TRIGGER and SELECT ANY TABLE privilege (this is required to create and drop logging tables and triggers).

Objects created by the extension

Migration functions

The db_migrator callback function db_migrator_callback() returns the migration functions provided by the extension. See the db_migrator documentation for details.

The “metadata view creation function” create_oraviews creates some additional objects in the FDW stage that provide information that will be helpful for Oracle migrations:

package definitions

packages (
   schema       text    NOT NULL,
   package_name text    NOT NULL,
   is_body      boolean NOT NULL,
   source       text    NOT NULL
)

• is_body is FALSE for the package definition and TRUE for the package body definition

This view can be used to make the transition of package code easier.

segments

segments (
   schema       text   NOT NULL,
   segment_name text   NOT NULL,
   segment_type text   NOT NULL,
   bytes        bigint NOT NULL
)

This foreign table is most useful for assessing the size of tables and indexes in Oracle.

migration cost estimate

migration_cost_estimate (
   schema          text    NOT NULL,
   task_type       text    NOT NULL,
   task_content    bigint  NOT NULL,
   task_unit       text    NOT NULL,
   migration_hours integer NOT NULL
)

• task_type is one of tables, data_migration, functions, triggers, packages and views.

• task_content is the quantity for this task type

• task_unit is the unit of task_content

• migration_hours is a rough estimate of the hours it may take to complete this task

This view can help to assess the migration costs for an Oracle database.

Additional objects

table function oracle_test_table

This function tests an Oracle table for potential migration problems. You have to run it after db_migrate_prepare.

The parameters are:

• server: the name of the Oracle foreign server

• schema: the schema name

• table_name: the table name

• pgstage_schema (default pgsql_stage): The name of the PostgreSQL stage created by db_migrate_prepare.

schema and table_name must be values from the columns of the same name of the tables table in the PostgreSQL stage.

This is a table function and returns the Oracle ROWID of the problematic rows as well as a message describing the problem.

Currently there are tests for two problems:

• zero bytes chr(0) in string columns

• values in string columns that are not in the database encoding

function oracle_migrate_test_data

This function calls oracle_test_table for all tables in the PostgreSQL staging schema and records the results in the table test_error in the FDW stage (after emptying the table).

In addition, an error summary is added to the table test_error_stats in the FDW stage. This is useful for measuring the progress of cleaning up bad data in Oracle over time.

The function returns the total number of errors encountered.

The function parameters are:

• server: the name of the Oracle foreign server

• staging_schema (default fdw_stage): name of the remote staging schema

• pgstage_schema (default pgsql_stage): The name of the PostgreSQL stage created by oracle_migrate_prepare.

• only_schemas (default NULL): An array of Oracle schema names that should be migrated to PostgreSQL. If NULL, all schemas except Oracle system schemas are processed. The names must be as they appear in Oracle, which is usually in upper case.

tables oracle_test_table and test_error_stats

These tables contain individual and summary results for runs of oracle_migrate_test_data.

Replication functions

function oracle_replication_start

This function creates all the objects necessary for replication in the Oracle and PostgreSQL databases. PostgreSQL objects will be created in the Postgres staging schema, Oracle objects in the same schema as the replicated table.

This function should be called right before db_migrate_tables, and no data modification activity should occur on Oracle between the time when you start oracle_replication_start and the time you call db_migrate_tables.

The function parameters are:

• server: the name of the Oracle foreign server

• pgstage_schema (default pgsql_stage): The name of the PostgreSQL stage created by oracle_migrate_prepare

The objects created by the function are:

• a PostgreSQL foreign table __ReplicationEnd that shows a timestamp guaranteed to be earlier than the oldest active transaction on Oracle

• a PostgreSQL table __ReplicationStart used to store the starting point for the next replication catch-up

For each table in the tables table of the Postgres stage that has migrate set to TRUE, the following objects are created:

• an Oracle table __Log_<tablename> to collect changes to <tablename>

• an Oracle trigger __Log_<tablename>_TRIG on <tablename>

• a PostgreSQL foreign table __Log_<schema>/<tablename> for the Oracle change log table

function oracle_catchup_table

Copies data that have changed during a certain time interval from an Oracle table to PostgreSQL.

This requires that oracle_replication_start has created the required objects and that the data migration has finished.

Parameters:

• schema: the schema of the migrated table

• table_name: the name of the migrated table

• from_ts: replicate changes later than that timestamp

• to_ts: replicate changes up to and including that timestamp

This is a “low level” function called by oracle_replication_catchup; it can be used if you want to parallelize catch-up by running it concurrently for different tables.

function oracle_catchup_sequence

Parameters:

• schema: the schema of the migrated sequence

• sequence_name: the name of the migrated sequence

• staging_schema (default fdw_stage): name of the remote staging schema

Queries the current value of the Oracle sequence on the remote side and sets the migrated sequence to that value.

function oracle_replication_catchup

Copies all changes in all Oracle tables and sequences since the last catch-up to PostgreSQL.

The start timestamp is taken from __ReplicationStart, the end from __ReplicationEnd (which contains the latest safe timestamp). After successful completion, the replicaton end time is saved in __ReplicationStart for the next time.

Parameters:

• staging_schema (default fdw_stage): name of the remote staging schema

• pgstage_schema (default pgsql_stage): The name of the PostgreSQL stage created by oracle_migrate_prepare

You can call this function anytime after oracle_replication_start has completed.

Unless you have no triggers or foreign key constraints in your database, you should set the configuration parameter session_replication_role to replica when calling this function. Then triggers don’t fire, and foreign key constraints are not checked.

oracle_replication_catchup uses the SERIALIZABLE isolation level on Oracle, so it sees a fixed snapshot of the Oracle database, and the data will be consistent on the PostgreSQL side, even if the Oracle database is modified concurrently.

If you want to use replication for near-zero down time migration, call it twice in short succession and make sure that there is no data modification activity on Oracle during the second call. Once the second catch-up has completed, you can switch the application over to PostgreSQL immediately.

function oracle_replication_finish

Removes all objects created by oracle_replication_start in PostgreSQL and Oracle.

This is used to clean up after you have finished migrating from Oracle.

• server: the name of the Oracle foreign server

• pgstage_schema (default pgsql_stage): The name of the PostgreSQL stage created by oracle_migrate_prepare

Limitations

Not all Oracle partitioning options are supported by ora_migrator. Only list, hash and range partitioning are supported in PostgreSQL, so other partitioning strategies are not supported. Also, list and range partitioning with a partitioning key with two or more columns is not supported. When ora_migrator encounters a partitioning schema that it cannot migrate, it will migrate the table to an unpartitioned table.

Replication

ora_migrator offers a simple trigger-based replication functionality from Oracle to PostgreSQL.

This can be used to migrate databases from Oracle to PostgreSQL with almost no down time.

The procedure is as follows:

• Prepare migration as described in the db_migrator documentation by calling db_migrate_prepare and db_migrate_mkforeign.

• Suspend all data modification activity on the Oracle database. This is necessary because Oracle does not support transactional DDL.

• Then call oracle_replication_start to set up all the required objects. This will create log tables and triggers in the Oracle database.

• Then start the data migration as usual with db_migrate_tables.

  As soon as db_migrate_tables has started, data modification activity on the Oracle database can resume. The migration will run using the SERIALIZABLE transaction isolation level, so the migrated data will be consistent.

  Make sure that you have enough UNDO space on Oracle, else the data migration may fail.

• Migrate constraints and indexes with db_migrate_constraints and other objects as described in the db_migrator documentation.

• At any time, you can call oracle_replication_catchup to transfer changed data from Oracle to PostgreSQL.

  This calls oracle_catchup_table for all affected tables, so to parallelize operation, you can call that latter function directly for all affected tables.

  Note that catching up will not purge the log tables on Oracle.

  To avoid problems with foreign key constraints in PostgreSQL, make sure that the configuration parameter session_replication_role is set to replica while you are running oracle_catchup_table.

  For near-zero down time migration, the last call to oracle_replication_catchup must also be performed while there is no data modification activity on the Oracle database. After that call, switch the application over to PostgreSQL.

• To end replication, call oracle_replication_finish. That will delete all the objects created for replication.

• Finally, call db_migrate_finish to drop all auxiliary objects.