Can services adopt the future of devops?
Startups and products can move faster than agencies that serve clients as there is no feedback loops and manual QA steps by an external authority that can halt a build going live.
One of the roundtable discussions that popped up this week while we’re all in Minsk is that agencies which practice Agile transparently as SystemSeed do see a common trade-off. CI/CD (Continuous Integration / Continuous Deployment) isn’t quite possible as long as you have manual QA and that lead time baked-in.
Non-Agile (or “Waterfall”) agencies can potentially supply work faster but without any insight by the client, inevitably then needing change requests which I’ve always visualised as the false economy of Waterfall as demonstrated here:
Would the client prefer Waterfall+change requests and being kept in the dark throughout the development but all work is potentially delivered faster (and never in the final state), or would they prefer full transparency, having to check all story details, QA and sign off as well as multi-stakeholder oversight… in short - it can get complicated.
CI and CD isn’t truly possible when a manual review step is mandatory. Today we maintain a thorough manual QA by ourselves and our clients before deploy using a “standard” (feature branch -> dev -> stage -> production) devops process, where manual QA and automated test suites occur both at the feature branch level and just before deployment (Stage). Pantheon provides this hosting infrastructure and makes this simple as visualised below:
This week we brainstormed Blue & Green live environments which may allow for full Continuous Integration whereby deploys are automated whenever scripted tests pass, specifically without manual client sign off. What this does is add a fully live clone of the Production environment to the chain whereby new changes are always deployed out to the clone of live and at any time the system can be switched from pointing at the “Green” production environment, to the “Blue” clone or back again.
Assuming typical rollbacks are simple and databases are either in sync or both Green and Blue codebases link to a single DB, then this theory is well supported and could well be the future of devops. Especially when deploys are best made “immediately” and not the next morning or in times of low traffic.
In this case clients would be approving work already deployed to a production-ready environment which will be switched to as soon as their manual QA step is completed.
One argument made was that our Pantheon standard model allows for this in Stage already, we just need an automated process to push from Stage to Live once QA is passed. We’ll write more on this if our own processes move in this direction.