[anchochilis]

I think the author has a point here. In an ideal world, we would of course spin up a new microservice every time a PR in the foobar-widget-generator service begins to deviate from generating foobar widgets. In practice, we make delivery tradeoffs all the time. It's not at all uncommon for service scope to creep while a new, urgent feature is being experimented with. And launching a new service is never, ever going to be as cheap as updating an existing, well-maintained one.

My own hard-line requirement when it comes to naming services is that they should be a single word with some relationship to the service's purpose. Ideally a common English word, but proper nouns are permitted if they improve clarity. Brevity must ALWAYS take precedence over clarity. There's only so much you can express in a name anyway; a detailed explanation of exactly what a service does should exist in documentation.

Otherwise you end up with long names like "horizon-blob-profile-server" or "batch-process-execution-engine". Multiple words lead to ambiguity. Inevitably you end up using dashes, camel case, underscores, or no differentiation at all in order to represent word boundaries in different systems, because VCS, filesytems, domain names, and cloud systems all have different sets of permitted characters. This makes automating your infrastructure painful.

And of course people resort to acronyms when discussing the services, which means everyone is forever getting the dfkg service confused with dkfg. "Whoops, I deleted the wrong database!"

[allknowingfrog]

I have this exact regret. Code may be meaningless, but acronyms are both meaningless and confusable.