Skip to content

How it works

The database migration process, stage by stage

Assess the source, plan and convert the code, move the data, validate it, then switch over.

15 h 40 min

total run for Meridian Retail (example), from schema deploy to the end of validation

Run data flows forwardFindings and warnings

The run

Follow ERP-Oracle to ERP-PG from start to switch-over

Oracle Database 19c to PostgreSQL 16.4 for Meridian Retail (example), with 5 schemas, 1,036 tables and 540 code objects.

  1. 1

    Stage 1 · Assess

    See the whole job before it starts

    Datalobe reads every schema, table and code object, then scores readiness and lists what needs a person.

    • 1,036 tables and 540 code objects in scope
    • 214 findings, grouped by code
    • Readiness 71 of 100

    Assessment Report

    ERP-Oracle to ERP-PG / Assessment Report

    71of 100

    Ready with remediation
    Tables
    1,036
    Rows
    2,141,876,402
    Columns
    71,940
    Constraints
    7,969
    Indexes
    4,212
    Code objects
    540

    Top findings

    • DFLUX-301CONNECT BYINV.V_STOCK_POSITION and 17 views
    • DFLUX-305Package stateFIN.PKG_GL_POSTING and 9 packages
    • DFLUX-322NUMBER without precision1,284 columns

    Stage 1 of 5

  2. 2

    Stage 2 · Plan and convert

    Agree the plan, review the code

    Fixed rules convert tables, columns, types, keys and the data copy. AI converts stored code (procedures, functions, triggers, complex views). Nothing AI-written is deployed silently.

    • Code first, then 3 waves
    • In the example run, 540 of 540 compiled on the target before deploy
    • Plan approved 1 Oct 2026, 16:20 UTC

    Migration Plan

    ERP-Oracle to ERP-PG / Migration Plan

    • First540 code objects
    • Wave 1COMMONHR
    • Wave 2INVFIN
    • Wave 3SALES

    Decisions

    • NUMBER without precision to NUMERIC(38,10)Sarah Mitchell
    • CHAR(1) 'Y'/'N' to BOOLEANapproved

    Plan approved 1 Oct 2026, 16:20 UTC · Sarah Mitchell

    Stage 2 of 5

  3. 3

    Stage 3 · Migrate and sync

    Copy in parallel, keep the target current

    Data moves by schema in parallel. Change capture keeps PostgreSQL in step with Oracle until switch-over.

    • 2,141,876,402 rows in 13 h 48 min
    • 43,100 rows/s average, 61,200 peak in the example run
    • 1,204,318 changes synced

    ERP-Oracle to ERP-PG / Run progress

    Rows copied

    2,141,876,402

    • SALES312 tables
    • FIN248 tables
    • INV196 tables
    • HR142 tables
    • COMMON138 tables
    Average
    43,100 rows/s
    Peak
    61,200 rows/s
    Copy time
    13 h 48 min

    Changes synced 1,204,318 · lag at cut-over 4 s

    Stage 3 of 5

  4. 4

    Stage 4 · Validate

    Check everything against the source

    17 checks run on every table in six groups, and deterministic read-only functions are executed on both databases. Code computes the verdict.

    • 17,612 check runs, 0 failed in the example run
    • 7 FIN rounding warnings accepted by Sarah Mitchell

    Validation ReportCertificate of Migration

    ERP-Oracle to ERP-PG / Validation Report

    Checks
    17,612
    Failed
    0
    Warnings accepted
    7
    • Row counts match on 1,036 tables
    • Values compared
    • Keys and constraints hold
    • 7 rounding warnings accepted by Sarah Mitchell
    • 61 functions give the same answers

    Stage 4 of 5

  5. 5

    Stage 5 · Cut over

    Switch over with the evidence in hand

    Applications switch from Oracle to PostgreSQL once lag is low and the verdict is in.

    • Lag at switch-over 4 s in the example run
    • Downtime 18 min in the example run

    Migration Report

    ERP-Oracle to ERP-PG / Cut-over summary

    Downtime

    18 min

    • Cut-over 4 Oct 2026, 02:00 UTC
    • Lag at cut-over 4 s
    • Applications switched to ERP-PG

    Stage 5 of 5

Roles

Who does what?

Four roles, and each approval is recorded with a name and time.

  • Org admin

    Runs the account

    • Sarah MitchellHead of Data Platform

    Does everything an Operator does, and manages users and roles and sets the AI model key.

  • Operator

    Runs the migration

    • Emily BrooksLead DBA
    • Marcus ReedDatabase engineer
    • Daniel OrtizDatabase engineer

    Connects databases, starts runs, decides type mappings, edits converted code and cuts over.

  • Auditor

    Checks the record

    • Olivia HayesInternal audit

    Read-only access to every report, the certificate and the run record, for audit and compliance teams.

  • Viewer

    Follows the migration

      Read-only access to reports, for stakeholders who need to follow the migration.

    What can each role do?

    Role permissions: the actions each of the four roles can take
    ActionOrg adminOperatorAuditorViewer
    Add source and target connectionsYesYesNoNo
    Start assessments and migration runsYesYesNoNo
    Approve a run, never one you startedYesYesNoNo
    Decide type mappings and accept, edit or rewrite converted codeYesYesNoNo
    Accept validation warningsYesYesNoNo
    Cut over with change data captureYesYesNoNo
    Deploy code without validationYesNoNoNo
    View reports and download the certificateYesYesYesYes
    Manage users and roles and set the AI model keyYesNoNoNo

    The person who starts a run can never approve it.

    Timeline

    The full run, hour by hour

    15 h 40 min from schema deploy to the end of validation, starting on 2 Oct 2026 UTC.

    Run clock 13:40 UTC

    2 Oct 2026 22:00 to 13:40 UTC the next day

    1. Schema and code22:00 to 23:10 · 1 h 10 min
    2. Data copy, wave 1 · COMMON and HR23:10 to 23:50 · 40 min
    3. Data copy, wave 2 · INV and FIN23:50 to 05:20 · 5 h 30 min
    4. Data copy, wave 3 · SALES05:20 to 12:58 · 7 h 38 min
    5. Validation12:58 to 13:40 · 42 min

    Then change data capture keeps the target in sync until cut-over on 4 Oct 2026 02:00 UTC. In the example run: lag 4 s, application downtime 18 min.

    Related

    Questions

    Frequently asked questions

    How long does a database migration take?

    It depends on size, code and the cut-over window. The example run (1,036 tables, 2.14 billion rows) took 15 h 40 min from schema deploy to validation. Your Assessment Report estimates the effort.

    Do we have to rehearse?

    Rehearsing is a good idea, and the process supports it. In the example, a rehearsal on 26 to 27 Sep 2026 failed validation, which led to a type mapping decision.

    Who approves what?

    An Org admin or an Operator can approve a run, but never one they started. The same two roles decide type mappings and accept validation warnings. Each approval is recorded with a name and time.

    What do we get at the end?

    The Assessment Report, Migration Plan, Validation Report and Migration Report, the Certificate of Migration when the verdict allows it, plus DDL and rollback scripts, the execution log and a manifest.

    Free assessment

    Bring your hardest schema.

    Stage 1 is free. We run the assessment and send you the report.