---
title: "Oracle to PostgreSQL Migration Tools: Ora2Pg vs AWS DMS"
description: "What Ora2Pg, AWS DMS Schema Conversion, pgrecon and AI code agents each convert, what they leave to people, and how to plan a cutover with rollback."
url: https://axiointelligence.com/blog/oracle_to_postgresql_migration_tools_rollback
---

[← Blog](https://axiointelligence.com/blog)

oracle to postgresql migration, ora2pg, aws dms schema conversion, oracle migration tools, database migration, cutover rollback, change data capture

# Ora2Pg, AWS DMS Schema Conversion or an AI Agent? Choosing Oracle to PostgreSQL Migration Tools and a Cutover You Can Roll Back

Brice Ayres / November 2, 2026

Short answer

No single tool migrates Oracle to PostgreSQL: Ora2Pg or AWS DMS Schema Conversion convert most schema objects, assessment tools such as pgrecon size what's left, and SQL embedded in application code needs its own pass, often with an AI code agent. Move the data with a full load plus change data capture, and keep reverse replication back to Oracle running after cutover so a rehearsed rollback loses no transactions.

Choose tools by layer, not one tool for everything. For the schema, Ora2Pg (open source) or AWS DMS Schema Conversion will convert most tables, indexes, views and a good share of PL/SQL, and flag what they can’t. Assessment tools such as Ora2Pg’s cost report or the new open-source pgrecon size the leftover work before you commit a budget. The SQL embedded in Java, Python or shell scripts needs its own pass, which is where AI code agents have recently made a real difference. For the data, run a full load plus change data capture (CDC), then keep **reverse replication from PostgreSQL back to Oracle** after cutover so you can roll back without losing a transaction.

If you’re still deciding whether to leave Oracle at all, start with our [Oracle to Postgres cost post](https://axiointelligence.com/blog/oracle_to_postgres_migration_cost). This one is for teams that have decided and now need tools and a cutover plan.

## What does each Oracle to PostgreSQL tool actually convert?

| Tool | What it does well | What it leaves to you |
| --- | --- | --- |
| [Ora2Pg](https://ora2pg.darold.net/) | Free and open source. Connects to Oracle, extracts structure and data, generates PostgreSQL SQL scripts, and produces a migration cost assessment | Package state, autonomous transactions, complex PL/SQL, app-embedded SQL |
| [AWS DMS Schema Conversion](https://docs.aws.amazon.com/dms/latest/userguide/CHAP_SchemaConversion.html) | Converts Oracle schema objects to Aurora or RDS PostgreSQL DDL with a rules engine and flags objects needing manual work | Same categories; tied to AWS targets |
| [pgrecon](https://www.postgresql.org/about/news/pgrecon-open-source-oracle-to-postgresql-migration-assessment-and-schema-conversion-3410/) | Apache-2.0, public since Oct 22, 2026. Works offline from a read-only SQL\*Plus dump, runs 80 rules with line-level evidence, estimates effort in person-day ranges, and emits DDL only for what it can prove (for example, CONNECT BY becomes WITH RECURSIVE) | Everything in its “residue report”; it’s young (release 0.7.3) |
| AI code agents | Finding and rewriting Oracle SQL embedded across many languages, with test-and-retry loops | Business sign-off, testing, anything the tests don’t cover |

A few notes on the table:

- **Ora2Pg is the default starting point.** It’s free, mature and widely used, and its assessment report is a quick way to size a schema. The [PostgreSQL wiki’s Oracle conversion page](https://wiki.postgresql.org/wiki/Oracle_to_Postgres_Conversion) collects many of the manual fixes.
- **DMS Schema Conversion fits when the target is Aurora or RDS** and you plan to use DMS for the data too. In The New York Times migration below, AWS reports it automated most of the schema objects and flagged the rest.
- **pgrecon is worth a look for assessment** when you can’t connect a tool to production, because it never connects to either database. Its benchmark claim (423 statements from Oracle sample schemas, zero rejected) is the project’s own, so test it on your schema.

## What do the tools leave for people to fix?

Every tool above stops at roughly the same places:

- **PL/SQL packages** with package-level variables and initialization code. PostgreSQL has no packages.
- **Autonomous transactions** (`PRAGMA AUTONOMOUS_TRANSACTION`), often used for audit logging.
- **CONNECT BY edge cases.** Simple hierarchies convert to `WITH RECURSIVE`; `NOCYCLE`, `SYS_CONNECT_BY_PATH` and ordering rules need review.
- **NUMBER precision.** Unqualified `NUMBER` becomes `numeric`, which is correct but slow where `integer` or `bigint` would do.
- **Empty string vs NULL.** Oracle treats `''` as NULL; PostgreSQL doesn’t, so results can change silently.
- **SQL embedded in application code.** This is usually the biggest gap. Schema tools convert the database, not the Java, Python, Go or shell scripts that send it queries.

We covered these in more depth, with examples, in our [cost post](https://axiointelligence.com/blog/oracle_to_postgres_migration_cost).

## Where do AI agents help, and what did The New York Times do?

AI agents help most with the embedded SQL. In a case study AWS published on Oct 26, 2026, [The New York Times moved PLATO, its subscription revenue system, from Oracle to Aurora PostgreSQL](https://aws.amazon.com/blogs/database/the-new-york-times-cut-oracle-migration-time-from-18-to-5-months/). According to the post:

- The initial scan found about **110,000 conversion items** across three schemas and more than 8,000 application files with nearly **6 million lines of code**, with Oracle SQL embedded in Java, Gradle/Groovy, Python, Go and shell scripts.
- Rule-based estimates put the effort at up to **18 months**. The team finished in about **5 months**.
- DMS Schema Conversion handled the schema. An AI agent on Amazon Bedrock converted embedded SQL, generated parity tests and fed errors back for correction. DMS handled continuous replication.
- Before cutover, they mirrored live production traffic and ran shadow queries against PostgreSQL behind feature flags while Oracle stayed the system of record.
- At go-live they sustained 18,500 requests per second with zero revenue impact and 100% row-level data accuracy, as reported.

The lesson we’d take isn’t “let AI do it.” It’s that the team paired fast conversion with heavy verification: parity tests, shadow traffic and a rehearsed rollback.

**Planning an Oracle exit and want a second opinion on tools and cutover?** [Book a call](https://axiointelligence.com/book) and we’ll review your schema, code and rollback plan with you.

## How do you plan a cutover you can roll back?

Keep Oracle current after you switch, so going back is a decision, not a disaster. The pattern from the NYT case, which is the one we’d use:

1.  **Full load, then CDC.** Load PostgreSQL, then stream ongoing Oracle changes with a CDC tool such as [AWS DMS, which supports Oracle as a CDC source](https://docs.aws.amazon.com/dms/latest/userguide/CHAP_Source.Oracle.html), until lag is zero.
2.  **Validate continuously.** Use row-level validation (DMS has [built-in data validation](https://docs.aws.amazon.com/dms/latest/userguide/CHAP_Validating.html)), compare query results, and run shadow traffic if you can.
3.  **Set up reverse replication before cutover.** From PostgreSQL back to an Oracle copy, tested in rehearsal. NYT kept an Oracle clone current with every transaction from Aurora after go-live.
4.  **Pre-provision the rollback target** so Oracle performs like production if you need it, and document and rehearse the rollback runbook.
5.  **Cut over with a short freeze:** stop writes, confirm zero lag, fix sequences, switch connection strings, start reverse replication, and watch.
6.  **Agree on rollback triggers in advance,** such as error rates, failed reconciliations or performance thresholds, and on how long you keep the reverse path running.

For the general dual-run pattern across systems, see [migrating without a migration freeze](https://axiointelligence.com/blog/migration_without_downtime).

## Which tool combination should you pick?

- **Small to mid-size schema, any target:** Ora2Pg for assessment and conversion, an open-source or managed CDC tool for data.
- **Moving to Aurora or RDS PostgreSQL:** DMS Schema Conversion plus DMS for full load, CDC and reverse replication. Ora2Pg is still a useful second opinion.
- **No tool access to production:** pgrecon’s offline dump for the first assessment.
- **Lots of SQL in application code:** add a code-conversion pass, AI-assisted or not, with parity tests that compare results on both databases.

Whatever you choose, budget most of the effort for testing, reconciliation and the cutover rehearsal, not the conversion itself. That’s the work our [database migration services](https://axiointelligence.com/services/database-migration) focus on.

## FAQ

### Is Ora2Pg still the best free Oracle to PostgreSQL tool?

It’s the most established. Ora2Pg is free, open source and covers schema, data and a cost assessment. Newer tools such as pgrecon add offline assessment, but are much younger.

### What is the difference between AWS SCT and DMS Schema Conversion?

Both convert Oracle schemas for AWS targets. DMS Schema Conversion is the managed version inside the DMS console; the Schema Conversion Tool is a desktop application. Check AWS’s current docs for which features each supports.

### Can AI convert Oracle PL/SQL and embedded SQL to PostgreSQL?

It can speed up the conversion a lot, as the NYT case shows, but only alongside parity tests, data validation and human review. Treat AI output like any other unverified code.

### How do you roll back an Oracle to PostgreSQL migration?

Keep reverse replication from PostgreSQL to an up-to-date Oracle copy running after cutover, pre-provision that Oracle environment, and rehearse the rollback runbook before go-live.

### How long does an Oracle to PostgreSQL migration take?

It depends on schema size, PL/SQL volume and how much SQL sits in application code. NYT’s estimate was up to 18 months and it took about 5. Our [cost post](https://axiointelligence.com/blog/oracle_to_postgres_migration_cost) covers what drives the timeline.

## Where to go from here

Assess first with Ora2Pg or pgrecon, convert the schema with the tool that fits your target, give embedded SQL its own pass, and design the cutover around CDC and a rehearsed reverse path. If you’d like help with any of it, [book a call](https://axiointelligence.com/book).

## We do this work for a living.

A 30-minute scoping call with the engineers who would do the work.

[Book a scoping call](https://axiointelligence.com/book)

Or email [contact@axiointelligence.com](mailto:contact@axiointelligence.com?subject=Project%20inquiry) and we’ll reply.

## Related posts

Oct 19, 2026

### [Oracle to Postgres Migration: Real Cost and When to Hire](https://axiointelligence.com/blog/oracle_to_postgres_migration_cost)

Oracle support renews every year whether you migrate or not. Here's what moving to PostgreSQL really takes, and where projects stall.

oracle to postgres, database migration, postgresql

Nov 4, 2026

### [Keeping Db2 for z/OS and PostgreSQL in Sync During a Mainframe Exit: Your Change Data Capture Options](https://axiointelligence.com/blog/db2_zos_postgresql_change_data_capture)

AWS DMS can't do CDC from Db2 for z/OS. Compare log-based CDC tools, scheduled full-load refresh and app-level sync, and what each costs on the mainframe.

db2 z/os, change data capture, db2 cdc

Oct 26, 2026

### [Db2 11.5 Support Ends April 30, 2027: Upgrade to 12.1, Pay for Extended Support, or Move to PostgreSQL](https://axiointelligence.com/blog/db2_11_5_end_of_support_postgresql)

Db2 11.5 base support ends April 30, 2027. Compare upgrading to 12.1, paying for Extended Support to 2031, or a Db2 to PostgreSQL migration.

db2 to postgresql, database migration, postgresql