The problem:
PAAS and auto-scaling are brilliant, I love them, but I hate the hidden price...Vendor lock-in.IAAS and cloud VMs are great, I love them, but I hate the hidden price...Vendor lock-in.
Neither solution works in the long run.
Either we run into a missing facility in the PAAS, or our costs scale out of control as IAAS gains traction with the developers, testers and everyone starts spinning up VMs all over the place.
We end up creating a Heath Robinson-esque machine, solving each little problem by adding one more part to our monster. The complexity grows and grows and grows. Soon enough no more features can be added as keeping the machine running takes all our time.
Eventually we are going to realise that we need more/better/bigger and will have to change everything or go and find it somewhere else and we still have to change everything.
So vendor lock-in is the biggest elephant in the room, and we face it where ever we look at the moment. Yes, you can have all their shiny features, but you must create everything their way, and woe betide you if you think you can change provider easily.
Everything we do becomes tightly coupled to the providers infrastructure or systems:
development, deployment, testing, roll-back, DevOps, Sysops, metrics, and the list goes on...
There are ways around these issues, but they always feel like edge case coding, and I know another edge case is waiting just around the corner, in fact I have 3 in the backlog.
Are you ready? Here are the words, just in case:
"We welcome our Heath Robinson overlords..."
The dream:
For a long time, I have had a dream ... of a system that allows everyone to get the best of all worlds, without any group suffering to support other, and everyone is capable of working to improve their conditions.- my applications should be able to live anywhere
- without requiring a complete refit, refactor and rebuild just to move home
- I can code it, test it, build it, deploy it and manage it easily, uncoupled from the infrastructure
- No more herculean efforts just to get tests to run locally or in the build
- Use the best of breed as standards, but avoiding the Heath Robinson effect
- A simple configuration system that has defaults for everything, yet all can be overridden
- Simple to change, and old parts replaced or new injected without downtime
This thought has bothered me for a long time.
The solution:
Eventually, I fell back on time honoured solution.To de-couple 2 things, create an abstraction layer between them.
An Infrastructure Abstraction Platform: IAP.
Create an abstraction layer, a platform for my code, that runs on top of any IAAS, that can manage itself, and all the applications inside it, as well as interface with the containing IAAS, but avoid coupling my applications to the IAAS. Everything external to my platform and applications should be attached through discovery and configuration, so an IAAS provider change should only require a redeploy and a config change.
If you have any idea how much time/effort this would take, you can understand my reluctance to even think about starting it. That was until I discovered Vertx.
Next:
[Coming soon: What Vertx lets me achieve][Coming soon: Why I use ENYO to build UIs]
[Coming soon: Why I can now build my own platform on IAAS]
No comments:
Post a Comment