What happens when my migration fails at 90% completion?

This happens more often than Google admits. [Migration failures can be restarted from checkpoints](https://cloud.google.com/database-migration/docs/mysql/diagnose-issues), so you don't lose all progress. Common causes: - **Permission errors** - the service lost access to your source database - **Network timeouts** - VPC peering broke or network congestion - **Source database overload** - your DB couldn't keep up with the extraction Check the [error logs in the Console](https://cloud.google.com/database-migration/docs/postgres/diagnose-issues) first (they're usually useless generic messages like "Connection failed", "Migration job encountered an error", or my personal favorite: "An unexpected error occurred" - but try anyway). Most failures are fixable without starting over.

My CDC replication is falling behind - what do I do?

[Change Data Capture can lag](https://docs.aws.amazon.com/dms/latest/userguide/CHAP_Troubleshooting.html) when your source database is under heavy write load. CDC processes every insert/update/delete, and if your app is doing bulk operations, replication falls behind. You'll see this in the monitoring dashboard: CDC lag jumps from milliseconds to minutes or hours, the "Events behind source" counter keeps climbing, and your migration ETA changes from "2 hours" to "sometime next week". **Fix:** Pause heavy batch jobs during migration, increase your source database resources temporarily, or wait it out. The migration will catch up eventually (physics applies - you can't replicate faster than you write).

Can I migrate from AWS RDS without paying massive egress fees?

Yes, but [AWS charges for data transfer out](https://docs.aws.amazon.com/dms/latest/userguide/CHAP_ReplicationInstance.html). For a 1TB database, expect $100+ in AWS egress charges. Budget for this or consider migrating during AWS free tier periods if you qualify. (Spoiler: you probably don't.) Google's [Database Migration Service is free for homogeneous migrations](https://cloud.google.com/database-migration/pricing), but AWS still charges you to leave.

What do I do when Oracle's stored procedures turn into garbage PostgreSQL code?

Google's [AI conversion](https://cloud.google.com/database-migration/docs/convert-sql-with-dms#gemini-conversion) is surprisingly good but not magic. Complex PL/SQL with cursors, nested procedures, or Oracle-specific bullshit will need manual fixes. Plan for this or you'll be rewriting code at 2am. **Reality check:** Plan for manual conversion work. The AI handles maybe 80% correctly - you'll spend time fixing the other 20%. Budget for this in your migration timeline.

How do I know if my app will break after migration?

Test migrations are your friend. Create a [test migration job](https://cloud.google.com/database-migration/docs/overview), copy your app to point at the test database, and run your test suite. Data types, constraints, and subtle behavior differences between engines can break things. **Pro tip:** Oracle's NUMBER type doesn't map perfectly to PostgreSQL. SQL Server's DATETIME behaves differently than PostgreSQL's TIMESTAMP. Test everything.

What happens if Google's migration service just stops working?

[Google's service status page](https://status.cloud.google.com/products/vY4CRgRFNbqUXWWyYGFS/history) shows historical outages. When DMS breaks, your options are: - Wait for Google to fix it - Use [manual migration tools](https://cloud.google.com/sql/docs/mysql/import-export/) like pg_dump/restore - Try [third-party tools](https://www.striim.com/blog/an-introduction-to-database-migration-strategy-and-best-practices/) as backup Always have a Plan B for critical migrations.

My migration is stuck - how do I tell if it's actually working?

[Monitor CDC lag and transfer rates](https://cloud.google.com/database-migration/docs/overview) in the Console. If numbers aren't changing for hours, something's wrong. Check: - Source database connectivity - Network bandwidth utilization - Error logs for cryptic failures - Source database load (high CPU can slow extraction) "Stuck" migrations usually mean networking problems or resource constraints.

Can I rollback if everything goes to hell?

Before final cutover, yes - your source database is unchanged. After cutover, rolling back means losing any new data written to the destination. **There's no magic "undo" button.** **Best practice:** Test cutover procedures thoroughly, have rollback plans documented, and keep your source database running until you're 100% confident.

Will my ancient MySQL 5.6 instance actually migrate?

**Current support (what we've actually tested):** - MySQL 5.6+ → Cloud SQL for MySQL (8.0 works great, 5.6 is painful) - PostgreSQL 9.4+ → Cloud SQL/AlloyDB for PostgreSQL (probably works with 16/17, who knows) - SQL Server 2008 R2+ → Cloud SQL for SQL Server (yes, that ancient version still haunts us) - Oracle 11g+ → PostgreSQL (good luck with those stored procedures) **Not supported:** MongoDB, Cassandra, Redis, or anything NoSQL. This is for relational databases only.

How much does this actually cost?

[Homogeneous migrations are free](https://cloud.google.com/database-migration/pricing) (same database engine). Heterogeneous migrations cost [$40 setup fee + $0.40-$2.00/GiB](https://cloud.google.com/database-migration/pricing). **Hidden costs:** AWS egress fees (they'll get you on the way out), destination database sizing (always bigger than you think), testing environments, and the three weekends you'll spend debugging schema conversion failures.

Will my 10TB Oracle database work?

Technically yes, but [Google recommends calling sales](https://cloud.google.com/database-migration/pricing) for anything over 10TB. Large migrations need custom configurations, dedicated support, and lots of planning. **Reality:** Expect weeks of testing, multiple trial runs, and careful coordination. Enterprise migrations are projects, not weekend tasks.

What if the DBA who knew everything just quit?

This is why people migrate to managed services. [Database Migration Service documentation](https://cloud.google.com/database-migration/docs/) covers most scenarios, but tribal knowledge is hard to replace. **Survival tips:** Document everything about your current setup, export schemas/configs, and test migrations thoroughly. Managed services mean less tribal knowledge needed going forward.

Currently viewing the AI version

Switch to human version

Google Cloud Database Migration Service - AI Technical Reference

Configuration Requirements

Supported Database Versions

PostgreSQL: 12+ → AlloyDB or Cloud SQL (tested, works reliably)
MySQL: 8.0 → Cloud SQL (tested twice, smooth operation)
Oracle: 11g+ → PostgreSQL (high complexity, manual intervention required)
SQL Server: 2008 R2+ → Cloud SQL (legacy support available)

Critical Network Configuration

VPC peering setup: Automated by DMS (saves 3 hours of manual configuration)
Firewall rules: Auto-configured (prevents 80% of common failures)
Cross-cloud networking: Supports AWS → GCP migrations with additional egress costs

Performance Specifications

Migration Downtime

Successful DMS migration: <5 minutes downtime
Manual migration: 8+ hours downtime
AWS DMS equivalent: <10 minutes downtime

Data Transfer Rates

800GB PostgreSQL database: 6 hours transfer time from AWS
CDC replication lag: Milliseconds to hours depending on write load
Large database threshold: 10TB+ requires custom configuration and dedicated support

Cost Analysis

Service Pricing

Homogeneous migrations: Free (same database engine)
Heterogeneous migrations: $40 setup fee + $0.40-$2.00/GiB
AWS egress charges: $100+ for 1TB database migration from AWS

License Cost Comparisons

Oracle Enterprise Edition: $47,500 per processor + support fees
Oracle Standard Edition: $17,500 per processor + support fees
PostgreSQL: Free (eliminates licensing complexity)

Critical Failure Scenarios

High-Probability Issues (80% of problems)

VPC peering failures: Network connectivity between source and destination
Permission errors: Missing user grants or username formatting issues
CDC lag: Replication falls behind during heavy write operations
Migration failures at 90% completion: Connection timeouts or resource exhaustion

Network-Related Failures

Corporate VPN interference: Blocks migration traffic (6+ hour debugging sessions)
Username formatting: Usernames with dashes cause authentication failures
Cross-cloud routing: VPN compatibility issues with Google's network routes

Data Conversion Issues

Oracle stored procedure conversion: 70% accuracy rate, 30% requires manual rewriting
AI conversion limitations: Complex PL/SQL, cursors, nested procedures fail conversion
Data type mapping: Oracle NUMBER → PostgreSQL, SQL Server DATETIME → PostgreSQL TIMESTAMP

Resource Requirements

Time Investments

Simple PostgreSQL migration: 5 minutes to 6 hours depending on size
Oracle → PostgreSQL migration: 3 weeks longer than planned due to stored procedure conversion
Schema conversion debugging: Manual rewriting at 2am for failed AI conversions

Expertise Requirements

Homogeneous migrations: Minimal DBA expertise required
Heterogeneous migrations: Requires DBA familiar with both source and destination engines
Oracle conversions: Deep knowledge of both Oracle PL/SQL and PostgreSQL required

Infrastructure Prerequisites

Source database connectivity: Direct network access or VPN
Destination database sizing: Always larger than source estimates
Testing environments: Required for validation before production cutover

Implementation Reality vs Documentation

What Actually Works

PostgreSQL → AlloyDB: Reliable, tested multiple times
MySQL 8.0 migrations: Smooth operation, good compatibility
Automated networking setup: Saves significant manual configuration time
Migration restart capability: Failures resume from checkpoints, don't restart from beginning

Hidden Operational Costs

AWS egress fees: Not included in Google pricing, budget separately
Testing environment costs: Required for pre-production validation
Weekend debugging sessions: Plan for 3am troubleshooting when things break
Manual procedure conversion: 20% of Oracle stored procedures need manual rewriting

Breaking Points and Limitations

CDC overload: Heavy write loads cause replication lag and migration delays
10TB+ databases: Require sales consultation and custom configuration
Legacy database versions: MySQL 5.6 is "painful", newer versions recommended
NoSQL databases: Not supported (MongoDB, Cassandra, Redis excluded)

Migration Decision Matrix

Use Google Cloud DMS When

Migrating PostgreSQL → AlloyDB: Proven reliable performance
Escaping Oracle licensing costs: $500K+ annual savings possible
Reducing operational overhead: Managed services eliminate 3am database pages
Hardware is failing: Emergency migration from dying on-premises systems

Avoid Google Cloud DMS When

Complex Oracle stored procedures: >30% manual conversion work required
NoSQL databases: Service doesn't support non-relational databases
Very old database versions: Upgrade source database first
Critical uptime requirements: Test extensively before production migration

Troubleshooting Decision Tree

Migration Stuck/Failed

Check CDC lag: If increasing, reduce write load on source database
Verify network connectivity: VPC peering and firewall rule issues
Review permission errors: User grants and authentication problems
Monitor resource utilization: Source database CPU/memory constraints

Post-Migration Validation

Data integrity checks: Compare row counts and key data samples
Application compatibility testing: Data type and behavior differences
Performance validation: Query performance on destination database
Rollback procedures: Document and test before final cutover

Support and Recovery Options

When Things Break

Google Cloud Support: Required for production migrations, engineers are knowledgeable
Migration restart: Automatic checkpoint recovery prevents full restart
Manual fallback tools: pg_dump/restore, mysqldump as backup options
Third-party alternatives: Striim and other tools for service outages

Rollback Limitations

Pre-cutover: Source database unchanged, full rollback possible
Post-cutover: No magic undo, new data on destination lost in rollback
Source database retention: Keep running until 100% confidence in migration

Comparative Analysis

Feature	Google DMS	AWS DMS	Manual Migration
Cost	Free (homogeneous)	$0.20/hour + storage	Weekend overtime
Downtime	<5 minutes	<10 minutes	8+ hours
Network Setup	Automated VPC	Manual VPC peering	SSH tunnels
Monitoring	Built-in dashboards	CloudWatch complexity	tail -f logs
Failure Recovery	Checkpoint restart	Replication instances	Start over

Success Factors

Technical Requirements

Network connectivity: Stable, high-bandwidth connection between source and destination
Resource planning: Adequate CPU/memory on source database during migration
Schema compatibility: Understanding of data type mappings and behavioral differences

Operational Requirements

Testing strategy: Multiple test migrations before production cutover
Rollback planning: Documented procedures and decision criteria
Team expertise: DBA knowledge appropriate for migration complexity
Timeline planning: Factor in debugging time and manual conversion work

Risk Mitigation

Backup validation: Verify source database backups before migration
Application testing: Full test suite execution against migrated database
Monitoring setup: CDC lag and migration progress tracking
Support engagement: Google Cloud Support for production migrations

Useful Links for Further Investigation

Resources Worth Your Time (And Some That Aren't)

Link	Description
Database Migration Service Product Page	The marketing page actually has useful info (shocking, I know). Read this first to understand what you're getting into.
Database Migration Service Documentation	Google's docs are surprisingly readable (unlike AWS's documentation nightmare). Start here if you want to avoid the usual tutorial hell where step 3 assumes you've already completed step 17.
Database Migration Service Pricing	The only pricing page that doesn't make you call sales for real numbers. Homogeneous migrations are free - that's not a typo.
PostgreSQL Migration Quickstart	Actually works as advertised. Follow this for your first PostgreSQL migration - it'll save you hours.
MySQL Migration Guide	Covers the networking gotchas that usually break MySQL migrations. Read the VPC section twice.
Oracle to PostgreSQL Migration	Prepare for pain. This guide helps, but converting Oracle stored procedures is still going to hurt.
Cloud Skills Boost Training Path	Hands-on labs that don't completely suck. Use these to practice before touching production.
Database Migration Concepts and Principles	Dense but comprehensive. Read this if you want to understand why things break (and they will break).
PostgreSQL Migration with DMS Architecture	Enterprise-grade patterns. Useful if you're migrating more than a toy database.
Database Migration Service REST API	API docs that don't make you want to cry. Use this to script your migrations.
gCloud CLI Commands	Command-line tools for when clicking through the console gets old (after about 5 minutes).
Terraform Provider Documentation	Infrastructure as Code. Because clicking buttons in production is for amateurs.
Google Cloud Community Forums	Mostly people asking the same five questions, but occasionally someone posts the exact error message you're getting at 3am.
Stack Overflow - Google Cloud DMS	Where you'll find the real answers buried under 20 "did you try turning it off and on again" responses.
Google Cloud Support	Expensive but worth it for production migrations. Their engineers actually know what they're talking about.
Cloud SQL Documentation	Managed MySQL, PostgreSQL, and SQL Server. Works as advertised, monitoring is solid.
AlloyDB for PostgreSQL	PostgreSQL but faster and more expensive. Worth it if you need the performance.

Google Cloud Database Migration Service - AI Technical Reference

Configuration Requirements

Supported Database Versions

Critical Network Configuration

Performance Specifications

Migration Downtime

Data Transfer Rates

Cost Analysis

Service Pricing

License Cost Comparisons

Critical Failure Scenarios

High-Probability Issues (80% of problems)

Network-Related Failures

Data Conversion Issues

Resource Requirements

Time Investments

Expertise Requirements

Infrastructure Prerequisites

Implementation Reality vs Documentation

What Actually Works

Hidden Operational Costs

Breaking Points and Limitations

Migration Decision Matrix

Use Google Cloud DMS When

Avoid Google Cloud DMS When

Troubleshooting Decision Tree

Migration Stuck/Failed

Post-Migration Validation

Support and Recovery Options

When Things Break

Rollback Limitations

Comparative Analysis

Success Factors

Technical Requirements

Operational Requirements

Risk Mitigation

Useful Links for Further Investigation

Resources Worth Your Time (And Some That Aren't)

Related Tools & Recommendations

Google Cloud SQL - Database Hosting That Doesn't Require a DBA

Google Cloud Platform - After 3 Years, I Still Don't Hate It

AWS Database Migration Service - When You Need to Move Your Database Without Getting Fired

Azure Database Migration Service - Migrate SQL Server Databases to Azure

Migrate Your Infrastructure to Google Cloud Without Losing Your Mind

Striim - Enterprise CDC That Actually Doesn't Suck

jQuery - The Library That Won't Die

I've Migrated 15 Production Systems from AWS to GCP - Here's What Actually Works

Google Cloud CDN - Decent Performance if You're Already Paying Google

Hoppscotch - Open Source API Development Ecosystem

Stop Jira from Sucking: Performance Troubleshooting That Works

Debezium - Database Change Capture Without the Pain

Google Cloud Storage Transfer Service - Move Your Shit Without Losing Your Mind

Northflank - Deploy Stuff Without Kubernetes Nightmares

Google Cloud Developer Tools - Deploy Your Shit Without Losing Your Mind

AWS Migration Hub - Track Your Migration So You Don't Lose Your Mind

Oracle Zero Downtime Migration - Free Database Migration Tool That Actually Works

AWS RDS - Amazon's Managed Database Service

LM Studio MCP Integration - Connect Your Local AI to Real Tools

Google Kubernetes Engine (GKE) - Google's Managed Kubernetes (That Actually Works Most of the Time)