How to Migrate Legacy Software Without Downtime

A step-by-step checklist for cutting over a legacy ERP or billing system without a maintenance window, with reconciliation gates and rollback drills.

Maysan Engineering Team · · 5 min read

Xin

Most legacy migrations do not fail because the new system is badly built. They fail in the six hours when both systems are half-live, invoice numbers have gaps, and nobody is certain whether last night's ledger batch ran once or twice. If you run engineering or operations at an Indian SMB sitting on a decade-old ERP, a custom billing module, or an Access-era internal tool, that fear is rational. The fix is sequencing and reversibility, not a bigger team.

What "migrate legacy software without downtime" actually requires

It requires that the old system keep serving real traffic until the new one has already handled that traffic in production, and that every switch you flip is reversible within minutes. That rules out the big-bang cutover most teams default to — a Friday-night export, a DNS change, and a weekend of hope. Downtime is not avoided by moving faster; it is avoided by never having a moment when no system is authoritative. Two invariants carry the whole project: exactly one source of truth per record, and routing you can point back to the old system without a code deploy. Everything below is machinery for those two invariants.

Step 1: Map every consumer and freeze the migration scope

Write down every consumer of the legacy system — human, script and device — because the ones you miss are the ones that cause the outage. In practice that list is longer than the architecture diagram suggests:

  • Cron jobs and nightly exports on an app server nobody has logged into since 2021
  • Excel macros and shared-folder CSV drops that feed accounts
  • Tally or SAP import scripts that run on the 3rd of every month
  • A mobile app or WhatsApp bot hitting an undocumented endpoint
  • Biometric attendance or warehouse scanners writing straight to the database
  • Three people's ad-hoc report queries that will never be documented

Then freeze scope. Migrate one bounded slice — order intake, invoicing, or the customer master — not "the ERP". A slice that keeps two to three engineers busy for more than eight weeks is usually two slices wearing a trench coat.

Step 2: Put a seam in front of the legacy application

Insert a routing layer — a reverse proxy or a thin API gateway — between the legacy system and everything that calls it, before you write new feature code. It is the cheapest insurance you will buy. Once traffic passes through the seam, you can send 5% of read requests to the new service, diff the responses, and send them back, without touching the legacy codebase. On the data side, the equivalent is a set of database views or a change-data-capture stream, so the new system reads a stable shape while the old one keeps writing. Sketch this layer first; it constrains every later decision, and there is a walkthrough of the pattern on our architecture page.

Step 3: Dual-write, backfill in batches, and reconcile daily

Run both systems in parallel with writes going to each, and treat reconciliation as a daily gate rather than a check at the end. Backfill history in checkpointed batches during off-peak hours, throttled so the legacy disk does not spike — 5,000 rows per batch with a 200ms pause is a reasonable starting point. Reconcile on totals, not samples: invoice value sums per GSTIN per day, row counts per table, hashes of last-modified timestamps. Do the maths on tolerance beforehand. A 0.01% mismatch on a 50,000-row table is five rows, and those five rows are the entire project. Keep the parallel-run environment provisioned for the whole project rather than rebuilt per test — it is one of the few places where cloud infrastructure choices directly change migration risk.

Step 4: Cut over in slices with a rehearsed rollback

Cut over one slice at a time, in ascending order of business damage, and only after you have executed the rollback in production at least once. The rehearsal is the step teams skip and the one that decides whether the migration is calm or catastrophic. A checklist that holds up:

  • Freeze deploys on both systems for the window, with no exception for "quick fixes"
  • Take a backup and restore it on a separate box, timing the restore
  • Confirm exactly one scheduler owns each cron job; duplication here means double invoicing
  • Route internal staff first, then 5% of customers, then the rest
  • Name one person on the call who can call rollback without a debate
  • Set a hard time box: if error rates breach your threshold inside 30 minutes, roll back and diagnose

What a realistic timeline looks like

Take a 60-person distributor with a 2014 custom billing app, about 180 invoices a day and 9,000 in peak GST months. One module — not the whole system — typically takes eight to ten weeks: two weeks mapping and scope freeze, one week on the seam, three weeks of dual-write and reconciliation, two weeks of sliced cutover. A project of that shape usually runs with three engineers partly allocated, and the only visible interruption is a single-digit-minutes read-only window at night, announced two days ahead. Treat these as planning ranges, not promises; the mapping phase is the one that overruns.

Where to start

Pick the smallest slice that still touches money — invoicing or order intake — and spend the first week only on the consumer map, because that document decides everything after it. If you want a second opinion on the seam and the reconciliation gates before production traffic points at anything new, that is the kind of review we do at Maysan Labs. Start with the map, not the rewrite.

Buying software or a website?

Twelve questions to ask any vendor before you sign: price, your data, fit for an Indian business and support.

Get the free checklist

Written by Maysan Engineering Team at Maysan Labs, a software studio in Gurgaon, India. Follow us on LinkedIn