In defense of immutability

I thought I’d whip together a short blog post as a reaction to Mark Burgess' (@markburgess_osl) keynote at the Cloudstack Collaboration Conference.

I thoroughly enjoyed the keynote but was a bit taken aback by one of the points that was made. It can be summed up in a quote that has since been floating around:

I think a few things are misguided in Mark’s analysis of what is meant by immutable infrastructure

Immutable infrastructure semantics

What people refer to when promoting immutable infrastructure is predominantly the promotion of immutable data structure to represent and converge infrastructure. In now way does it mean that systems are viewed as immobile.

It is also a realization that many parts of infrastructure can now be treated as fully stateless automatons and for which lifecycle management can partly happen at a different level than previously namely by replacing instance instead of reconverging the state of an existing system.

Immutability promotes persistent data structures and a better audit trail

Since immutable data structure enforce building new copies of data on change, instead of silently mutating state, it provides a better avenue to create a complete history of state for data.

One of the main thing immutable data provides is a consistent view of data at a certain point in time - which again, does not limit its ability to evolve over time. This property is key in building simple audit trails.

For one thing, if immutability was such a limiting factor, I don’t think so many programming languages would be built around it.

Immutable infrastructure does not conflict with configuration management

Many people now turning to this new way to think about lifecycle management of infrastructure and systems as a whole come from years of experience with configuration management and in no way are trying to get rid of it. It is more a reflexion on what conf management is today, where it happens and how it could evolve.


I’ll now go read Mark’s book which he pitched rather well, except for this minor nitpick (ok that and the cheapshot at lambda calculus, but I won’t cover that!) :-)