Data migration
Moving data from one place or shape to another, such as between clouds, out of a legacy system or into a new structure, without losing or corrupting any of it.
Updated October 2026 · with examples from our own builds
What does data migration mean in practice?
Migrations come in two kinds: moving data between systems, such as from an old ERP to a new one or between storage providers, and reshaping data inside one system, such as adding a new structure to every record, often called a backfill.
Projects slip here more than anywhere. Real data is messier than anyone expected, runs take longer than estimated, and a half-finished migration leaves the system in neither state. Good ones can pause, resume, verify what they copied and be rehearsed before the real run.
Before any migration, ask three things: how will we know every record arrived intact, what happens if it stops halfway, and what will it cost to run?
Data migration at a glance
- Two kinds
- Between systems, or reshaping data in place
- Must have
- Verification, resume after failure and a rehearsal
- Often forgotten
- Transfer fees and run time on real volumes
Where has data migration come up in our builds?
Migrations copy objects in key order with a checkpoint after each clean copy, stream every object through a SHA-256 hash and verify it at one of three levels, and delete the source of a move only after the copy verifies. Egress is estimated first, and above $10 the user must acknowledge it.
In April 2026 we restructured the backend from four flat program types into two verticals, Academy and Training, with academy courses holding dated batches whose sessions the server generates.
Moving to MongoDB's geospatial queries would need a backfill across the five models that store coordinates as separate fields, so it is recorded as the next step rather than done under deadline.
Common questions about data migration
How do I know a migration didn't lose anything?
Count and checksum. Every record or file should be counted at both ends and, for files, hashed during the copy and compared at the destination. Storafleet doesn't trust provider ETags for this, because providers compute multipart ETags differently.
Can we keep using the old system while we migrate?
Often, if the migration runs in batches and catches up on changes made in the meantime. Agree a cut-over moment when writes stop on the old system, and rehearse it on a copy first.
Related terms
Planning a product and not sure where data migration fits? Ask us, or browse all 116 terms in the glossary.