Data Migration Checklist: Mapping, Validation and Cutover

ソースインベントリを構築します。

移動されるデータセットとそれを理解している人々をリストします。レコード数、日付範囲、関係、および異常な値を特定します。サンプル エクスポートでは形式の問題が明らかになる可能性がありますが、完全なデータセットの一貫性は確立されません。移行スケジュールを確定する前に、より広範なレビューを計画してください。

値がどのように変換されるかについて同意します。

空の値や履歴フォーマットを含め、ソースフィールドから宛先フィールドへのマッピングを文書化します。重複する顧客の統合などのビジネス上の決定を技術的な変換から分離します。比較できるように元のデータを保存し、各変換ルールを誰が承認したかを記録します。

引き継ぎ前に調整します。

ビジネス記録を反映する比較を定義します: カウント、選択された合計、および代表的なリンクされたトランザクション。例外がどのように記録され、修正されるかを計画します。結果が許容できるかどうかを誰が決定するかに同意し、人々が依存するシステムを変更する前に回復アプローチを文書化します。

Rehearse extraction, transformation and reconciliation

Create an inventory of source tables or files and assign a business owner to their meaning. Inspect representative records, including missing values, duplicates, historic formats and relationships. Write transformation rules explicitly so a reviewer can explain why the destination value differs from the source.

Run a rehearsal in a suitable non-production environment. Record rejected records and validate relationships, important totals and selected record details. Agree how changes made during the transition are captured. Time the rehearsal to inform the cutover discussion, then review restoration and recovery options. Obtain approval for unresolved exceptions before scheduling the production move, rather than treating them as cleanup that somebody will handle later.

A migration mapping and cutover worksheet

Maintain the worksheet alongside the migration scripts. Use a separate decision log where different departments must agree a mapping.

• Source and destination: dataset, field, type, identifier and responsible business owner.
• Transformation: format conversion, default, lookup or rule, with a concrete example.
• Exception policy: duplicates, missing mandatory values, invalid references and rejection ownership.
• Validation: record comparisons, relationship checks, totals and agreed acceptance limits.
• Cutover: write restrictions, final extract, change capture and communication responsibilities.
• Recovery: decision authority, restoration procedure, verification and treatment of new writes.

An import count is not a complete validation

Equal row counts do not establish that values, relationships or business totals are correct. Include checks reflecting how the destination will be used. A migration can load every row and still associate a transaction with the wrong account.

Do not assume rollback is simply restoring yesterday’s backup. If the destination has accepted new work, a recovery decision must address those writes. Keep the source available as agreed, restrict sensitive data appropriately and rehearse the recovery approach before the production cutover.

Can we clean all data during the move?

Only include cleanup rules that have a clear owner and agreed interpretation. Unresolved business decisions can make a migration unpredictable. Separate essential transformations from later quality improvements, and preserve a record of changes so reviewers can reconcile the migrated result.

ILLUSTRATIVE PLANNING EXAMPLE

Example: migrating product records

Export a representative sample of product records from the old system. Check duplicate identifiers, missing units and categories that do not exist in the destination. Write mapping rules and test them in a non-production environment. Include discontinued products and unusual prices in the review rather than testing only clean records.

Agree reconciliation totals, rejected-record handling and who approves the cutover. Document the recovery decision if validation fails; a successful import alone is not acceptance.

FURTHER READING

Official resources to explore

Explore the related official guidance below. The templates and examples are practical starting points to adapt to your own project.

• AWS DMS: data validation