3

I am trying to convince my fellow workers and management as part of DevOps transformation to adopt "fail fast" in almost everything. Amazon has two principles "bias for action" & "are right a lot" which I belive are quite similar. my dilemma is espically when adopting a new solution. as an Enterprise, they tend to evaluate several products first, then run many proofs of concepts, then an open bid. which is a waste of time and resources. my take is that decide on one solution quickly based on technical and financial merits, start using it if you fail, you can select another technology, reversing a decision on technology is not a big deal, isn't DevOps about being Agile and having quick feedback loops? I would like to hear your input.

Walid
  • 131
  • 3
  • 1
    "run many proofs of concepts" this seems to go into the right direction but not the way you would prefer? I am not sure what to answer yet, but in general I think this is question of culture; I am thinking of the Three Ways and maybe you could shape your answer towards a specific situation? – Ta Mu Jun 01 '18 at 04:02
  • 2
    Yeah, sign a 3 year contract for dozen of bucks each year and decide after 3 month it is not a good solution... Does it ring a bell of why it could be a bad approach ? – Tensibai Jun 01 '18 at 08:10
  • @Tensibai you do not have to start with the enterprise solution, most solutions now days have an upstream open source solution – Walid Jun 01 '18 at 18:47
  • @PeterMuryshkin take an example of PaaS, one can evaluate Openshift, Rancher, Docker EE, Xebialabs, or raw kubernetes with several add-ons. or just set mind to start using one and in his on-going learning journy can find the limitations and check if other solutions suffer from the same limitation, or they can be resolved in chosen platform, why spend time in several POCs while time can be invested in getting things done. – Walid Jun 01 '18 at 18:52
  • @walid you're asking a different question then. When describing an open bid as end of process, free OSS was out of context for the question for me. – Tensibai Jun 02 '18 at 06:28
  • @Tensibai I value your input, I know I might be wrong in generalizing "Fail fast" on all processes, however, the evaluation of new technology with the intent of adopting it via POC, open bid is one use case. another use case is adopting an OSS e.g. configuration management/deployment framework. there is a tendency in the enterprise to not settle fast on one technology/tools unless several have been evaluated. – Walid Jun 02 '18 at 12:16
  • 1
    And there's reason for that, just the salary of people taking the learning curve of a technology is a large cost, you'd better know if the tech will last more than a few month, fail fast is for trying ideas/implementation , not technologies as far as I know – Tensibai Jun 02 '18 at 16:54
  • @Tensibai agree, one use case is Pivotal vs. Openshift. as far I can see both are mature technologies and can provide the same requirements. it's actually a waste of resources and high salaries to not make a fast decision, I am saying to think long term, I am also saying, we have so many technologies that are mature with minor differences or strategies that are well known. I can see that I should not generalize. thank you for your input. – Walid Jun 02 '18 at 19:48

2 Answers2

4

start using it if you fail, you can select another technology

This approach has two major drawbacks:

  1. Technical debt

    While it may be tempting to apply "fail fast" mentality to architecture and design, skipping or abbreviating the design phase would be a mistake. fast-fail should be used in conjunction with good design practices, not in lieu of them. Making hasty decisions can often result in poorly scaled solutions. While using MySQL seems like a good idea when you are working with a 100mb database to generate reports, it scales poorly into the terabyt range for doing your archival reports. Using DNSMasq to run an overlapping DNS Zone might seem like a good political decision at the time, migrating off of that and resolving the disparity to between your DNS Zone and the overlapping upstream zone a few years later turns into a nightmare.

  2. Expense

    Technology can be expensive. Buying a solution only to discard it later cares with it hugely wastful expenses. But even more expensive is the time expended by personnel. Salaries are every companies' largest expense. Wasting time developing a solution only to discard it and develop a replacement is wildly inefficient.

This however should not be construed to mean that decision paralysis or analysis paralysis isn't a problem. It is! Fail-fast just isn't the best way to combat it. There are other, much more tried and true ways to combat this without comprimising good design principles.

James Shewey
  • 3,752
  • 1
  • 17
  • 38
2

I believe in 'fail fast' but it's also important to tackle understanding an opportunity in the right way.

If you come up with an idea, code in a vacuum for a bit, then release it to people, you will undoubtedly fail fast, but not in a way you would like. A better approach is if you come up with an idea, clearly articulate your hypotheses on why you think it's a great idea, then validate that idea with potential clients to see what they think. This will give you a great opportunity to clarify the idea, refine it, and possibly reject it outright.

I never cease to be amazed at how much my initial idea improves from just a little collaboration with people that I would want to 'sell' it to. And - to be honest, I've discovered why many of my ideas really weren't all that great - and shelved them fast (before any code was written often).

mike
  • 139
  • 3