-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
This may be quite clear in its simplicity, and particularly to those familiar with etcha/jsonnet. However, what invariably happens is that you start seeing things like vars embedded in templates, embedded in the config mgmt implementation language, embedded in strings. The source of the values becomes incredibly difficult to reason about and to make changes against; does the value come from the target host's env, from the runner's env, from the packaging step's env, from a network request made by one of these stages (eg to a secret server), etc.
Take a look at an example of what, IMO, is an absolutely horrid helm chart that Gitlab ships for installing their CI/CD runner: https://gitlab.com/gitlab-org/charts/gitlab-runner/-/blob/ma...
Bash code, in yaml, in golang template. Besides even the most advanced IDEs failing to grok such a freak of technical nature, there's no way I would believe any dev that told me they understand what the state of their system will be given some input to this morass.
In a recent position I was asked to try and make a nomad installation viable in a pretty standard corporate environment (not some special operational space e.g. cloudflare), and it was even worse; some configuration expansion was 5 layers deep, with 3 different templating engines, once consul templates were involved in generating an app's config, and the nomad config being env-generalized through generation by a higher-level helm-like tooling.
Re state bag:
I'm glad you mentioned nix, as I think it, and to a looser extent containers, really approach the issue in the only humanly-tenable fashion (again IMO): starting mutation from a known state. In a lot of cases that state is "nothing" as it's the simplest known state not only to position the beginning of some configuration flow at, but also the most straight-forward from which to deterministically derive a desired end state from.
I definitely applaud having tests as a core component of your system, the problem is that you can not derive determinism from nondeterminism even with the best tests.
Because you are operating over a nondeterministic bag of state, you can never guarantee that your tests provide a representation of a transfrom from any potential state to the desired end state, only for some particular input state (or set) which may or may not representative of what is found on the actual targets.