• DataMigration.AI
  • Posts
  • GitHub Is Migrating Mid-Surge. Is Your Migration Plan Already Stale?

GitHub Is Migrating Mid-Surge. Is Your Migration Plan Already Stale?

The system you assessed is not the system you will cut over

What's in it?

  • GitHub is migrating to a multi-region Azure setup while its traffic explodes

  • Why a migration plan is a snapshot of a system that keeps moving

  • The four numbers that quietly go stale between assessment and cutover

  • How continuous profiling keeps the plan matched to the real system

  • Three questions to answer before you lock your cutover date

GitHub logged 7.4 billion commits in September 2026, five times the volume of a year earlier, according to The Register.

And it is migrating its platform at the same time.

At its Universe conference this week, Microsoft's Jay Parikh wrote that GitHub is migrating to a multi-region deployment in Azure while it keeps serving developer traffic. He expects the migration to finish early next year.

The pressure is easy to see. Based on reconstructed status-page data, GitHub's 90-day availability reportedly dipped to about 84 percent in May.

Here is the part most migration leaders will miss.

The system GitHub planned to move is not the system it is moving now.

A platform being migrated while traffic surges behind it
OpenMap: We are building The Agent Network. We're raising a £5M seed. Invest in OpenMap.

What Changes Between Your Assessment and Your Cutover?

Every migration starts with a picture of the source system.

Someone profiles the data. Someone counts the tables, the records and the jobs. Someone estimates how long the bulk load will take and how big the delta will be on cutover night.

Then the program runs for nine, twelve, eighteen months.

And the source system keeps living its life. In that time, a typical estate can pick up:

  • new tables and columns added by teams who never heard about the migration

  • record volumes that grew far faster than the forecast

  • new integrations, feeds and scheduled jobs reading from the old system

  • new kinds of users, like automated agents, hitting it in patterns nobody modeled

The plan still looks complete.

It describes a system that no longer exists.

Your source system has changed since you assessed it. Find out how much before cutover night tells you the hard way.

Why Do Migration Baselines Go Stale So Quietly?

GitHub's case is extreme, but the pattern is ordinary.

Parikh put it plainly: agents are changing the volume and shape of software development. Volume is the obvious part. Shape is the dangerous part.

A migration plan is built on assumptions about both.

How many records move. How fast they change. Which processes write to them. Which reports read from them. How much can be reconciled in the cutover window.

When those assumptions are measured once, at the start, they start aging immediately. Nobody is assigned to notice.

A migration blueprint that no longer matches the system it describes

The numbers that drift most between assessment and cutover:

  • Volume: the bulk load that fit in a weekend can quietly grow into a four-day job

  • Change rate: the delta you planned to sync in an hour now arrives faster than you can apply it

  • Schema: fields added after the mapping was signed off have no target

  • Consumers: new jobs and feeds that nobody listed as an interface

None of these show up as a red flag on a status report.

They show up on cutover night.

A Snapshot Plan vs a Living Plan

Most teams do the right things. They just do them once.

What Teams Do

What They Miss

Business Impact

Profile the source at kickoff

Volume and change rate growth since then

Cutover window overruns

Sign off the field mapping

Columns and tables added later

Data silently dropped at load

List interfaces in the design phase

New feeds built during the program

Downstream systems break after go-live

Rehearse with a copy of production

How live traffic behaves under load

Performance surprises in week one

A rehearsal on old data proves the old plan works.

It does not prove the new reality fits.

This Is Not a Capacity Problem. It Is a Baseline Problem.

When a migration slips, the first instinct is to add capacity. More compute. More storage. More people on the cutover bridge.

GitHub is doing that too, with 4 million CPU cores and 160 PB of high-speed storage added this year. It is also redesigning its architecture to decouple storage from compute.

Capacity buys you room.

It does not tell you what changed.

Picture a payments team that sized its cutover in January. By September, transaction volume has grown, a fraud model reads from three new tables, and a partner feed now runs every five minutes instead of nightly. The runbook still says six hours.

Nobody made a mistake.

Nobody re-measured.

How This Gets Solved For You

Instead of profiling the source once and hoping it holds, DataMigration.AI keeps the picture current for the whole life of the program.

Profile AI runs automated profiling so volume, growth and data patterns are measured again as the source changes. Discovery AI surfaces new tables, jobs and interfaces as they appear. Map AI applies intelligent schema mapping to fields that arrive after sign-off. Quality AI and Reconcile AI provide continuous validation and real-time monitoring, so every rehearsal and every wave is checked against the system as it is today, not as it was at kickoff.

AI agents continuously profiling and validating data during a migration

Program leads see drift weeks before it reaches the cutover window.

Business owners can confirm totals match on both sides.

Engineers size the final load on real numbers, not old ones.

And migrations run 60% faster and 60% cheaper, with zero-downtime capability built in.

What Happens If You Skip This Step

Every long migration carries changes nobody measured.

  • Some will never matter.

  • Some will stretch your cutover window past the weekend.

  • Some will drop data that a regulator or a customer asks about later.

The target platform is rarely the reason.

The reason is a baseline nobody refreshed.

Before You Lock Your Cutover Date

When was your source system last profiled, and how much has it grown since?

Which tables, jobs and feeds were added after your mapping was signed off?

Was your last rehearsal run on data that looks like today's production?

Skip any one of these, and your cutover date is a guess.

Answer all three, and it becomes a commitment.

Don't Migrate the System You Assessed. Migrate the System You Actually Have.

A leader steering a migration through rising data volumes with live readings

Thank you for reading

DataMigration.AI & Team