Oracle to Azure PostgreSQL schema conversion limitations

This article summarizes the known limitations, unsupported objects, and considerations for converting Oracle schemas to Azure Database for PostgreSQL flexible server or Azure HorizonDB in Visual Studio Code.

Unsupported and partially supported objects

The following sections describe unsupported objects, partial conversion, and runtime considerations that need manual review. An individual object's outcome depends on its source definition, dependencies, and target prerequisites.

Discovery, extraction, conversion, and deployment describe different stages. An object can be discovered but excluded from extraction, or extracted but not converted. Review Discovery results for source scope, the HTML conversion report and Conversion Object Explorer for conversion results and notes, and Setup Experience reports for deployment results.

Oracle system and built-in schemas aren't extracted or converted. This includes schemas such as SYS, SYSTEM, XDB, MDSYS, CTXSYS, and WMSYS.

Important

A Converted result doesn't by itself establish compilation success, a complete implementation, deployment inclusion, or equivalent runtime behavior. Compilation checks syntax and structure, not data or application behavior. Review both Converted and Not converted objects and validate runtime behavior during data testing and user acceptance testing (UAT), before production use.

Address conversion issues with GitHub Copilot

Select Fix for Copilot-assisted corrections and choose from the latest models available in the GitHub Copilot Chat model picker. Model availability depends on the Copilot plan and organizational policies. For more information, see Change the AI model in GitHub Copilot Chat.

Use the recorded reason to decide what to do next. Not supported in Discovery, an extraction failure, and Not converted in conversion results describe different conditions.

For an object available in Conversion Object Explorer:

  1. Open its available definitions and Conversion notes. Identify missing source, a missing dependency, a compilation error, or functionality that needs a different implementation.
  2. Select Ask to open Copilot assistance and explain the limitation and possible PostgreSQL approaches.
  3. Select Fix for Copilot-assisted edits when a PostgreSQL correction is appropriate. Review the changes to the object and its dependencies, and inspect the compilation results. Corrected dependent objects are resolved automatically after successful compilation.
  4. If a reviewed and successfully compiled correction still needs a status update, select Resolve. It changes the recorded status to Converted; it doesn't open Copilot Chat, compile SQL, or deploy it.
  5. Select Refresh to update the conversion reports. Keep the tested corrections in the deployment artifacts and check Setup deployment scope.

For an object excluded from extraction or absent from Conversion Object Explorer, review Discovery and extraction results and the available Oracle source. Plan manual implementation, an alternative service, or exclusion from the conversion scope. Don't assume that an object-level Fix action is available or that Copilot can recover missing source logic.

Copilot assistance doesn't expand automatic conversion coverage or establish equivalent behavior. Independently validate the chosen implementation. For a worked example, see Correct an object and update its report.

Placeholder output

When the tool can't fully convert an object, it might emit best-effort or placeholder DDL. Some objects have no usable PostgreSQL definition.

A placeholder doesn't implement the original business logic. For example, a generated routine whose body raises an error can still appear as Converted. Inspect the generated definition and any conversion notes rather than relying on the status alone.

Objects that depend on a placeholder can require manual changes or be omitted from deployment. Review these dependencies in Conversion Object Explorer before deployment.

Cases that can require manual completion include:

  • Autonomous transactions. Conversion of routines that use PRAGMA AUTONOMOUS_TRANSACTION can depend on dblink. Missing required extensions or unsupported transaction logic can prevent a complete conversion.
  • Unresolved references. Missing or unresolved dependencies can prevent complete conversion or compilation.
  • Transaction control statements. Constructs such as ROLLBACK TO SAVEPOINT don't always convert cleanly.
  • Wrapped PL/SQL. The source body isn't available for conversion. For more information, see Wrapped PL/SQL.

Programmable objects

Object type Limitation or consideration What to do
Package Package-state behavior requires independent validation. Validate initialization and package-state persistence across calls, transactions, and sessions.
Routine using autonomous transactions Autonomous-transaction behavior requires independent validation. Validate required extensions, separate-connection identity and session context, data visibility, commits, rollbacks, and error handling.
Database and session-event trigger Oracle event handling can require a different target implementation. Validate the target event, scope, and permissions for LOGON, LOGOFF, STARTUP, SHUTDOWN, and SERVERERROR logic, and reimplement logic for unsupported events.
Type body Type bodies can be extracted, but extraction doesn't guarantee a complete PostgreSQL implementation of their methods. Review the generated definitions for type methods and their dependencies. Implement any missing logic manually.
Java source, Java stored procedure, and Java class Not extracted. Rewrite the logic in PL/pgSQL or move it to the application layer.
External C library (CREATE LIBRARY) Not extracted. Rewrite the logic in PL/pgSQL or move it to the application layer.
Wrapped PL/SQL The body can't be read. Generated output can contain placeholders rather than the original implementation. For more information, see Wrapped PL/SQL. Obtain the original unwrapped source before starting conversion, or implement the missing PostgreSQL logic manually.

For broader runtime testing guidance, see Validate complex code objects.

Wrapped PL/SQL

The tool can't read the implementation of PL/SQL units stored in wrapped (obfuscated) form by using the Oracle wrap utility or DBMS_DDL.WRAP. It might use available metadata to generate a placeholder, such as a routine that raises an error when called. This output doesn't reconstruct the original business logic.

Review wrapped units in Conversion Object Explorer even if their status is Converted. Inspect the generated definitions, conversion notes, and dependencies to identify missing implementations. A placeholder doesn't guarantee that dependent objects convert or deploy.

Obtain the original unwrapped source before starting conversion. If conversion has already completed, implement and validate the missing PostgreSQL logic manually. Rerunning a completed conversion isn't supported.

Data and storage objects

Object type Limitation or consideration What to do
User-defined operator (CREATE OPERATOR) Conversion isn't guaranteed and can produce placeholder output. Review the generated operator and test it.
Index (bitmap, bitmap join, global partitioned, or domain) These Oracle index definitions aren't extracted for automatic conversion. Review application dependencies on index-specific features, choose appropriate target indexes, and test query behavior and performance.
Materialized view log (MLOG$) Not extracted. Define and test the target materialized-view refresh strategy, including refresh schedules and data-freshness requirements.
External table Not extracted. Load the file into a table with a client-side \copy or COPY ... FROM STDIN, with a storage extension supported by the selected target service, or through the application.
Advanced Queuing (AQ) table Not extracted. Oracle Advanced Queuing carries application logic. Redesign it with an application queue, a PostgreSQL extension, or an Azure messaging service.
Blockchain table Not extracted. Reimplement the integrity requirement in the application layer.
Cluster and hash cluster Not extracted. Recreate as a plain or partitioned table.
Global temporary table Temporary-data behavior requires independent validation. Validate temporary-data lifetime, session isolation, and transaction behavior.
Directory object (CREATE DIRECTORY) Not extracted. Handle file access in the application layer.
User-defined index type (CREATE INDEXTYPE) Not extracted. Recreate the access method with a PostgreSQL extension index where one exists.
Database link Not converted automatically. Manually configure postgres_fdw or oracle_fdw, depending on the remote database. See Database links.
Synonym Some synonyms, including those that reference packages, procedures, or sequences, might not convert. Review target mappings and update dependent SQL to reference the appropriate PostgreSQL objects.
Application context (CREATE CONTEXT, DBMS_SESSION.SET_CONTEXT) Not extracted. Use PostgreSQL session settings with current_setting() for non-security application state. Contexts used for authorization require a separately designed and validated access-control implementation; ordinary custom settings don't preserve Oracle's trusted-setter restriction. See Security and identity objects.
Analytic view, attribute dimension, hierarchy, OLAP dimension, data mining model, and materialized zone map Not extracted for automatic conversion. Rebuild the required output as views or functions, or exclude it from the conversion scope.
LOB storage clause, registered XML schema (DBMS_XMLSCHEMA.REGISTERSCHEMA), and XML schema-bound index Not extracted. Remodel the storage or validation requirement on the target.
Resource Manager plan and consumer group, SQL profile, SQL plan baseline, edition, SQL Translation profile, Information Lifecycle Management policy, lockdown profile, and Workspace Manager object Not extracted. These are Oracle instance, optimizer, or lifecycle artifacts that are out of scope. Recreate them at the application or infrastructure layer, or exclude them from the conversion scope.

The tool doesn't automatically convert Oracle database links. Select postgres_fdw for a remote PostgreSQL database or oracle_fdw for a remote Oracle database and configure remote access manually. Both extensions are supported by Azure Database for PostgreSQL flexible server and Azure HorizonDB.

Manually configure the selected extension, connection, credentials, and permissions. Update dependent SQL, including @link references in synonyms and views, to use the new remote-access objects. Validate remote access and transaction behavior. Keep real credentials out of shared or committed SQL.

Security and identity objects

The tool doesn't support automated conversion of security and identity objects. Plan and validate users, roles, privileges, and data-protection policies separately from converted schema DDL. Use the Oracle source inventory to identify required access controls. Discovery listings or extracted metadata don't establish conversion support or confirm that access controls are configured on the target.

Object type Limitation What to do
User Automated conversion isn't supported. Provision logins separately on Azure. Recreate each user as a PostgreSQL role, and validate the resulting access before production use.
Role Automated conversion isn't supported. Recreate each role on the target.
System privilege Automated conversion isn't supported. Map the required access to PostgreSQL privileges, then apply the appropriate GRANT statements.
Role membership Automated conversion isn't supported. Recreate as GRANT <role> TO <role> statements.
Object privilege and grant Source grant metadata doesn't configure equivalent target permissions. Recreate grants from the source inventory, then validate them.
Row-level security and Virtual Private Database (DBMS_RLS) Not converted. Recreate the policies as PostgreSQL row-level security policies.
Oracle Label Security Not converted. Redesign the data-access boundaries on the target.
Data redaction (DBMS_REDACT) Not converted. Reapply masking on the target, in the application layer or with a PostgreSQL equivalent.
Profile, password policy, and unified or fine-grained audit policy (DBMS_FGA, AUDIT POLICY) Automated conversion isn't supported. Configure the equivalent Azure server parameters and auditing, and manage them as infrastructure as code.

Important

Missing row-level security, Virtual Private Database, Oracle Label Security, or data redaction protections can expose rows or values that Oracle filtered or masked. Implement equivalent protections and verify effective access before routing production traffic to the target.

Scheduler objects

Review job and schedule outcomes individually rather than treating Discovery inventory as proof of conversion. Generated job SQL doesn't establish that the target scheduler is configured, the job is enabled, or the Oracle orchestration behavior is preserved. Plan scheduling and orchestration separately from the routines that jobs call.

Object type Conversion consideration What to do
DBMS_SCHEDULER job and schedule Review the generated output and notes for the job and its schedule; routine conversion alone doesn't configure execution. Configure a scheduler supported by the destination, such as pg_cron where available, or an external scheduler. Verify timing, identity, commands, and activation.
Legacy DBMS_JOB job The job's execution environment and scheduling behavior need separate validation. Review the output and recreate missing scheduling logic with a supported database or application scheduler.
Program, chain, credential, file watcher, job class, group, window, and event-based job Automated conversion of these orchestration objects isn't supported. Rebuild the required orchestration separately, for example in an external scheduler.

Feature-level gaps

The following items are Oracle features rather than object types. They can require changes beyond creating valid PostgreSQL DDL. Review the source inventory as well as the converted definitions, because feature-level differences might not appear as separate objects in the conversion results.

Feature Limitation or consideration What to do
Hybrid partitioned table (Oracle 19.3 and later) Converts as an ordinary table, and the external partitions are lost. Identify hybrid partitioned tables in the source, and load the external data separately.
Editioning view and cross-edition trigger A PostgreSQL view or trigger definition alone doesn't establish equivalent edition-based behavior. Validate object visibility and trigger behavior for the converted edition during application rollout.
Identity, generated, and invisible columns Column behavior requires independent validation. Validate value generation, expression results, and column visibility to application SQL.
Immutable table (Oracle 19.11 and later) The immutable-table requirement isn't detected or recreated by the converter. Design and validate the required write and deletion restrictions separately.
Flashback Data Archive Temporal history isn't modeled. Use a history table pattern or a temporal extension on the target.
Flashback query (AS OF TIMESTAMP, AS OF SCN) Not converted. Rewrite queries that read historical versions of a row.
Oracle Sharding The converter doesn't model the source distribution strategy. Plan distribution and routing separately for the destination architecture.
BFILE and SecureFile LOB specifics External file contents and LOB storage behavior require separate migration planning. Review external file references and dependencies on Oracle-specific LOB features. Plan and test the target storage and access pattern.

Private temporary tables exist only at runtime, so they don't appear in the source inventory.

Operational boundaries

Keep these boundaries in mind when planning schema conversion:

  • Schema conversion doesn't migrate row data. Plan data migration and data validation separately from schema conversion.
  • Completed conversions can't be rerun. Review and correct the generated PostgreSQL definitions through Conversion Object Explorer. Use Fix for object and dependency corrections, or compile manual edits separately. Keep the tested corrections in postgres_ddl. Resolve updates object status, and Refresh updates reports; neither action reruns conversion. For an incomplete run, follow conversion failure guidance.

Get help

Start with the assisted correction workflow and the best practices. If a problem remains, use PostgreSQL extension support and include the extension version, failed stage, and redacted diagnostics. Don't include credentials or sensitive schema details.