How to Migrate a Database Without Downtime
An hour of downtime costs real money: lost orders, missed payments, a queue of support tickets. That’s why “how to migrate database without downtime” is one of the first questions teams ask before moving from Oracle to PostgreSQL, or off an ageing Sybase or Informix system.
The honest answer is that when you move between different database platforms, you can’t get to a literal zero. What you can do is shrink the outage from a lost weekend to a few minutes. For most businesses, that’s the win.
Zero downtime vs. near-zero
True zero downtime means the application never stops. In practice, cross-platform migrations aim for near-zero: a short switchover window where writes to the old database pause while the team confirms everything has caught up.
Getting to an actual zero requires so much extra machinery that the effort usually outweighs the benefit. Before you chase it, put a number on what an outage costs you. Three minutes at 4 a.m. is often cheaper than six months of engineering.
Same platform, or a different one
Moving from Oracle 12 to Oracle 19, or from a self-hosted PostgreSQL to a managed one, is a different task from moving Oracle to PostgreSQL. In the first case the schema and the stored procedures already speak the right language. You can lean on the database’s own replication, promote a standby, and finish inside a short maintenance window.
Everything in this article that sounds like hard work applies to the second case: different platform, different SQL dialect, different data types. Conversion has to happen before a single row moves.
Either way, take an inventory before choosing an approach. It’s worth knowing:
- how much data there is, and how much of it is still actively used;
- how many stored procedures and triggers, and roughly how many lines of code;
- which tables have no primary key, since sync mechanisms rely on them;
- what else connects to the database besides the main application: reports, integrations, scheduled jobs, ETL. All of it has to be repointed too, and it’s the usual source of nasty surprises;
- how much writing happens at peak, because that sets how fast the sync has to run.
Three ways to run the move
| Approach | Outage | Complexity | Best for |
|---|---|---|---|
| Big bang | Hours to a weekend | Low | Smaller databases, services with a scheduled maintenance window |
| Phased | Minutes per stage | Medium | Systems that can be split up and moved piece by piece |
| Bulk load + sync | Seconds to minutes | High | Busy production systems where downtime is expensive |
The third option is the real answer to the near-zero question, so it’s worth walking through.
How bulk load plus sync works
The idea is straightforward; the work is in the details.
- Convert the schema. Tables, indexes, keys and views are rebuilt on the new platform, allowing for differences in how each database handles data types.
- Convert the business logic. Stored procedures, functions and triggers. On an Oracle-to-PostgreSQL project this is usually the longest stage.
- Load the data. History moves across in batches while the old system keeps running.
- Keep the two in sync. Everything users do during the load, from new orders to edits and deletions, is copied over to the new database.
- Switch over. Once the gap between old and new is effectively zero, writes pause briefly, the last changes catch up, and the application points at the new database.
The mechanism behind step four is called change data capture, or CDC. Most databases have some version of it built in. On a healthy setup the lag stays under a second; if it sits at minutes and won’t come down, something upstream is the bottleneck.
One tempting shortcut is worth naming: having the application write to both databases at once. It sounds elegant, but if one write succeeds and the other fails, the two copies quietly drift apart. It’s a poor primary strategy.
Changing the structure while people are using it
Schema changes follow a pattern often called expand and contract:
- Expand. Add the new column or table, leave the old one alone.
- Migrate. Fill the new one, keep both in step, then move reads and writes across.
- Contract. Remove the old one once nothing depends on it.
Every stage is reversible, which matters at 2 a.m. For very large tables there are dedicated tools that restructure things in the background instead of locking users out.
A switchover checklist
- Lower DNS caching times in advance so traffic redirects quickly.
- Decide how the application will be pointed at the new database: config change, feature flag, or proxy.
- Set a hard number for acceptable lag before you commit. Anything above it means you postpone.
- Take a final backup of the source right before the cut.
- Keep the old database alive and untouched until the new one has been signed off.
- Rehearse the whole thing at least twice on an environment that mirrors production.
Checking what moved across
Comparing row counts alone isn’t verification. A reasonable minimum:
- row counts across every table, as a first pass;
- checksums on the tables that matter most;
- totals and other aggregates compared on key numeric fields;
- sampling for very large tables, full comparison for anything financial.
Where Ispirer fits
Ispirer Systems has been doing cross-platform migrations since 1999, and its tools cover the two heaviest stages: converting the code and moving the data.
SQLWays handles schema and business logic. It supports more than twenty databases, including Oracle, SQL Server, DB2, Sybase, MySQL, PostgreSQL, Informix and Teradata, and converts tables, indexes, views, procedures, functions and triggers. Ispirer puts automation at up to 95 percent, with the remainder covered by customising the conversion rules for a specific project. Whatever the tool doesn’t convert automatically still has to be picked up by a developer.
The data side is handled by the Ispirer migration tool, Ispirer Data Migrator, released in late 2025. A few things set it apart:
- it runs inside the target PostgreSQL database, with no separate agents or middleware to deploy;
- it connects to the source read-only and doesn’t require extra logging to be switched on there, which matters when the DBA team is protective of a production Oracle instance;
- Ispirer quotes speeds of up to 250 GB per hour, against roughly 60 GB per hour for SQLWays;
- if a load fails partway, it resumes from where it stopped rather than starting again.
In practice the sequence is: run the first load, run repeat catch-up passes while watching the gap shrink, pause the application briefly once it reaches zero, then bring everything up on the new database.
Two things to plan around. The catching up happens in repeated passes rather than as a continuous stream, and the approach is currently built for Oracle moving to PostgreSQL; other combinations move their data through SQLWays at the lower throughput above. The speed and automation figures are Ispirer’s own, and a free assessment tool and trial licence exist so you can check them against your database before committing budget.
Tools for each stage
Planning how to migrate database without downtime largely comes down to which tool covers which stage:
- schema and code conversion: SQLWays, ora2pg, AWS SCT, EDB;
- moving the data: Ispirer Data Migrator, native utilities, AWS DMS;
- continuous replication: GoldenGate, Debezium, Qlik Replicate, Striim, when repeated passes aren’t enough;
- verification: comparison tools, checksums, manual spot checks.
No single product covers all four, and a vendor claiming otherwise is selling you something. A migration that works is a combination of two or three tools, an honest estimate of the outage you can live with, and at least two full rehearsals before the real thing.
Have you read?
Global Human Prosperity Index: Best Countries in the World.
Global Mobility 2026: The World’s Most Powerful Passports.
Top Residency and Citizenship by Investment Programs Ranked.
Top CBI Programs: Best Countries for Second Citizenship.
Top RBI Programs: Best Countries for Residency.
Follow CEOWORLD Magazine on: Google News, LinkedIn, X (Twitter), and Facebook.
Note: The views expressed are those of the authors and do not necessarily reflect those of CEOWORLD Magazine, its Editorial Board, or management. Content is provided "as is," may contain monetized links, and is not professional advice. See our Ethics & Guidelines for details. Reproduction requires prior written permission.
Contact info@ceoworld.biz for inquiries or media requests.