Back to Daft Choices

Draft paper 001

Distributed Legacy

The strangler pattern is a sensible way to replace a legacy system. The daft choice is assuming the transition state will stay temporary just because the diagram says so.

Prompted by a LinkedIn post from Hatem Moushir on how strangler-pattern migrations can turn into distributed legacy.

The reasonable idea

Replacing a system piece by piece is usually more realistic than a big-bang rewrite. You keep the business running, reduce the blast radius, and learn from the parts you have already moved.

That is the appeal of the strangler pattern. It gives teams a way to change a live system without pretending the organisation can stop breathing while the new one is built.

The awkward middle

The middle state is where the real cost lives. The old system still exists. The new services need care. Data has to move between them. Monitoring doubles. Release paths multiply. Nobody is quite sure which side owns a rule that was never written down.

At first this is acceptable. It is the price of moving carefully. The problem starts when the organisation funds the start of the migration but not the end of it.

The hidden incentive

Starting modernisation is visible. Finishing it is often thankless. Leaders can point to new services, dashboards and delivery milestones while the remaining coupling sits behind the curtain.

Teams then optimise for progress that can be shown, not complexity that has actually been removed. The transition architecture becomes a place where everybody can claim movement while nobody has to pay down the full operational bill.

The daft choice

The daft choice is not using the strangler pattern. It is treating the temporary hybrid as a neutral phase instead of a system with its own incentives, owners and failure modes.

Once the hybrid is relied on for long enough, it stops being a migration plan. It becomes the architecture.

The pattern to watch for

Watch for migrations where every new slice adds another bridge, adapter or synchronisation job, but no old responsibility is retired. The system is not being strangled. It is being wrapped.

Questions worth asking

  • What part of the old system will be turned off next?
  • Who owns the cost of running both worlds at once?
  • Which duplicated rule, report or data path is allowed to disappear?
  • What would prove that complexity has gone down, not merely moved?

Comment carefully

Send examples or corrections by email. Responses are private by default and manually reviewed before anything is quoted. Respond to this draft.