I remember seeing this used inside config files for the Caddy webserver.
Google seem to like these executable config languages because they've got another open source one ("Starlark") a few notches up in expressivity.
Google seem to like these executable config languages because they've got another open source one ("Starlark") a few notches up in expressivity.
1. Build an AST on top of YAML a la CloudFormation. Now you're programming in YAML, hurray!
2. Extend your static language with executable features a la Terraform/HCL, basically reinventing (and very badly, at that) more traditional language features
3. Use text templates, a la Helm--now you can generate syntactically invalid configuration! (and absolutely trivially, at that)
4. Use an expression language (familiar, ergonomics) a la Pulumi, Starlark, Nix, Nickel, Dhall, etc
Note that in these conversations, someone inevitably shouts "use the simplest tool for the job!" ignoring that static configuration languages (and options 1-3 above) are strictly more complex for the reusability use cases outlined above.
EDIT: Pulumi isn't an expression language; rather, it lets you use real languages to generate configuration, and these languages often include powerful expression features. AWS's CDK is also in this category.
The scope and capabilities of the config language needs to be limited, otherwise we lose the ergonomic benefits of configuration in the first place.
I think one problem is that configuration needs start out simple and evolve to complex - as opposed to being obvious from the start you'll need dozens of discrete components to deploy. At that early stage, anything beyond a few lines of YAML seems like definite overkill.
Eventually, it becomes clear that it's very hard to maintain, but by then there's thousands of lines of hard-knocks battle-tested, working production config to try and basically recreate from scratch, without breaking anything.
Until you've gone through this process once (or maybe a couple times..) it's hard to see why you should make that initial leap to a much more complex, tools-required workflow. "But eventually, we will probably need..." is a tough sell against someone arguing to do the simplest thing.
I think it's wise to try and skip as many of those intermediate stages as possible. Of course; that's not a clear-cut solution strategy either; because what's "as possible"? Exactly how high up the language chain do you need to go; conversely which language (and environment) features are too powerful, rendering the language difficult to contain?
Me too. I feel like there should be some eponymous law about this. Every declarative language that starts out trumpeting "simplicity and not being Turing complete is a feature!" ultimately grows features until it is an imperative language, or gets replaced by one that is.
If you're gonna get there anyway, you may as well design for that instead of bolting on features poorly after the fact.
I agree, but it becomes very important to limit scope. For example, Azure templates allow looping and conditionals and all sorts of fancy stuff. Approaching a typical library of ARM templates is a massive undertaking: reading a JSON (or YAML) if statement is not ergonomic in any way - causal relationships can be multiple screens (or files) apart because of the sheer amount of JSON required to represent executable code.
It should be kept relatively lightweight, with stuff like CloudFormation GetAtt to glue deployed things together. Anything more complex should be solved with tooling designed for computation, i.e. programming languages (that emit config, e.g. Pulumi).
- file:
path: /etc/foo.conf
owner: foo
group: foo
mode: 0644Nearly every programming language has a YAML serialization library†. And before that serialization happens, your config can be expressed using the regular-ass coding features of your program, however you like. (For optimal clarity, I personally would suggest creating a builder DSL and using it.)
† Technically a language doesn't even need a YAML serialization library to emit valid YAML; because valid JSON is also valid YAML. You can just serialize to JSON on your end, and feed the result into anything that's expecting YAML.
I have run into the same type of issues with salt.
This probably sounds like a strawman, but it's not. It's how a lot of e.g. Python projects are configured - the "config" file is just a normal bit of code that gets run to produce a value. Unless you're using a programming language that absolutely sucks at expressing plain values (e.g. C or Java), it's much better than separate config files, IMO.
Ideological answer: For the same reason HTTP/2.0’s binary protocol didn’t instantly obviate/deprecate HTTP/1.0’s text protocol. Text has advantages: text is debuggable, and prototypable. If the interface between two programs is a text based declarative language, you can audit that text, diff that text, edit that text to see how changes affect the result, mock one side or the other by producing or consuming that text, etc. “GitOps” style config management would never work if config was all opaque binary blobs. These are all reasons that major software projects standardize on YAML or other widely-supported textual data serialization formats for their config.
Pragmatic answer: because we’re talking about production configuration management, here, which is, 99% of the time, about writing configuring and managing the third-party black-box components in your stack, not your own components. Your own business layer usually can be configured conventionally, with minimal explicit config, for your use case, since you built it to work idiomatically for that use-case. It’s all the third-party stuff that has an impedance mismatch to your use-case assumptions, translating to needing tons of config to do what you need.
And, obviously, if you don’t control the other end, you don’t decide how the other end does its config. Usually, these days, it’s YAML (or TOML) — for the ideological reasons mentioned above.
Example: Kubernetes. Big consumer of complex YAML. Many people try to template that YAML. Much simpler and less error-prone to just write a program to generate said YAML. No reason to assume you’re writing in whatever language the k8s orchestrator is written in. (In fact, there are multiple orchestrators, written in different languages, and the shared YAML resource spec is the only formal interface they share.)
I can see the argument for using a textual format (although I think it's weaker than you say; if we're generating this config with code then we don't want to diff or edit the generated config), but YAML seems like a singularly poor choice if you want reliable diffs and editing; it's like picking tag-soup HTML. Straight JSON (ideally with a schema), TOML or even XML seems like a better bet if you're generating it programmatically.
> And, obviously, if you don’t control the other end, you don’t decide how the other end does its config.
Right, in that case it's all moot. I took GP to be talking about what formats these tools should use. IMO if the tool is intended to consume a machine-generated config then it would be better to use a machine-oriented config format. I think the option of something like protobuf (which is language-independent) is underappreciated, but even restricting ourselves to textual options, something stricter than YAML seems like a better bet.
Other examples of formats like this, that are hand-authored in the small but generated in the large: RSS, SQL, CSV.
Again, Kubernetes is a prime example of this. K8s config YAML is designed with the intention of being hand-authored and hand-edited. It’s only when devs or their tools need to auto-generate entire k8s cluster definitions, that you begin needing to machine-generate this YAML. This generated YAML is expected to still be audited by eye and patched by hand after insertion, though, so it still needs to be in a format amenable to those cases, rather than in a format optimal for machine consumption.
> if we're generating this config with code then we don't want to diff or edit the generated config
Look more into GitOps. The idea behind it is that whatever tooling you’re using to generate config is run and the resulting config is committed to a “deployment” repo as a PR; ops staff (who don’t necessarily trust the tooling that generated the config) can then audit the PR, and the low-level changes it describes, before accepting it as the new converged system state. It puts a human veto in the pipeline between machine-generated config and continuous deployment; and allows for debugging when upstream tweaks aren’t having the low-level side-effects on system state one would expect.
I think the real issue is reproducibility; and that boils down to purity. Fully fledged languages all come with lots of apis and features to interact with the rest of the world, and it's quite unclear which apis have such dependencies and which do not - and it's seductively easy to do something actually useful in a "real" programming language that will make the whole configuration process unwieldy later - like, say, reading parts of the config from disk, getting some services public key off the internet, embedding a timestap, or even writing some computed config like a random key to a bit of storage for a later config process to consume. And once you do that, then the whole thing gets flaky, fast.
If you can rigorously avoid that, there's not too much advantage to a static config language.
But keep in mind that we’re not inherently talking about programming languages here — nor are we necessarily talking about people capable of programming as our configurators. We’re talking about third-party components that need to be configured by ops people, who may or may not be DevOps people. Usually they’re not — most ops people are just pure ops, and don’t know any programming languages. As well, most amateur integrators (e.g. a person setting up their own blog) aren’t programmers either.
The goal of these systems, when choosing a configuration solution, is twofold: to give pure-ops and amateur integrators a config language they can author directly, in a text editor, without learning programming; while also making that language formal/structured enough that it’s easy to machine-generate from your programming runtime of choice, if you do have those skills, and a rigorous mindset.
Sure, programming languages don’t necessarily require you to use the full-fledged expression syntax they enable, and so can “reduce” to a configuration-language-like subset of themselves.
But remember, again — ops people and amateur integrators. What do such people tend to do, to create their config? Read the reference config schema? No. They tend to look up tutorials with samples, or StackOverflow “solutions”, from arbitrary places on the Internet.
And what do the creators of those samples have in abundance? Cleverness and a desire for clarity of meaning. Traits that cause them to use the expressive features of whatever the configuration language is, in order to make their answers more “pithy”.
Which means that, to wield these “pithy” samples/solutions, the ops people and amateur integrators now have to understand how to “patch” one arbitrary piece of complex code into another increasingly-arbitrary piece of complex code.
The thing a static data-serialization format gets you, is that the rules for merging any two expression-nodes in it are very simple to learn, because there just aren’t that many types of expressions. There’s no way to be “pithy” with the configuration that requires people to learn entirely-new-to-them syntax.
By choosing to configure your system in YAML, you’re guaranteeing that the samples these ops people and amateur integrators find and attempt to glue together, will also just be pure YAML. And since their existing config file, and each new sample, are pure YAML, they’ll likely succeed at doing this gluing-together.
Meanwhile, DevOps people and enterprise integrators can create their own programs to generate the YAML — but since there’s no first-party framework for doing this, there won’t be much value in sharing these programs around, and so the samples the pure-ops people and amateur integrators find will never be given “in terms of” writing code for such a framework, but rather only in terms of the config YAML itself.
> I think the real issue is reproducibility; and that boils down to purity. [...] If you can rigorously avoid that, there's not too much advantage to a static config language.
Individual users might be able to rigorously avoid that (though expecting a rigorous approach to formal expression from non-programmers is a bit much.) But often it's the system itself that needs purity and reproducibility.
Remember, config formats are usually something executed at every startup — in other words, they're durable state that happens to be human-modifiable. (Think: the Windows Registry.) As the designer of a system, you don't want the same state you serialized today to deserialize to something else tomorrow; and you especially don't want the meaning of your state to depend contextually on the environment. You want to "pin down" your state.
A good example: programming-language package-ecosystem "lock files." In most languages, dependency-constraint specification is done in a programming language, such that the generation of those constraint expressions has access Turing-complete features. But once you lock those constraints down to a baked set of choices, the lockfile itself — the predetermined set of choices, that should be environment-independent — is not expressed in a Turing complete language (in any runtime I know of, at least) but rather is always expressed in its own little static declarative language; or at most in a limited "data-expressions only" subset of the parent language (e.g. Erlang's `file:consult/1` format.)
In this case, dep-constraints are the inputs to a config-generator program; while the lockfile is the config format itself. The config format is a necessary intermediate here; it'd be impossible for the runtime to make the same static guarantees about package management if it wasn't! (In fact, see e.g. Python's setup.py, where exactly that problem stymies any package-manager the Python ecosystem introduces from pre-determining dependency graphs before actually downloading and attempting installation of the dependencies.)
By the way, you conflate purity with turing completeness; but the two are not really all that strongly related. It's possible to have a turing incomplete language that is nevertheless impure (public I/O without unconstrained repetition), and conversely a turing complete language that is pure (i.e. keep your tape private).
I'd argue that turing completeness isn't as relevant as people make it out to be here. It's not a good thing, mind you, but it's just not that problematic either; externally imposed termination and storage limitation can render any turing complete system into a turing incomplete system - that's easy - but a system with uncontrolled sideeffects is almost intrinsically hard to manage. In fact, even technically turing-incomplete systems may well need to impose similar limitations anyhow, because a technically turing incomplete language that allows (say) nested loops or iteration - albeit bounded - may well not practically terminate, or nevertheless cause too much I/O. Some languages are really limited, and perhaps then you can get away without externally imposed resource constraints, but it's not clear to me how realistic that scenario is.
The real problem (to my mind) in general-purpose languages when it comes to using them for config-specification is not turing completeness, it's purity (i.e. reproducibility). And that's not even really a language issue alone, it's because those languages tend to come with large, pervasively used libraries, to the point that it's not trivial to just take some code off stackoverflow (say) and reliably tell whether it's pure or not - because that depends on the internals of all of those library methods too.
A Turing complete language can't be pure because there is no value that is equivalent to a nonterminating computation.
It's not reproducible if it's not a value. The point of a pure function is that you can replace it with the value that it evaluates to.
If you include nontermination as a value in your language then your language becomes almost impossible to reason about as you break almost every equivalence property you could think of. E.g. you can no longer say x * 0 = 0.
> Lack of side effects matters in the context of configuration, non-termination does not (and see the thread you're replying to for an argument as to why that is).
I don't find "but terminating code may still take a long time" to be a convincing argument that nontermination isn't important; rather it's an argument that code taking a long time might also be important (at least to the extent that it actually comes up in practice, which I'm not convinced of).
(A) - you're exceedingly unlikely to run into this issue in the first place. Nontermination limmits aren't set to things you're likely to hit without a runaway loop or recursion.
(B) - when you do hit the a forced termination issue, such replacements are usually irrelevant, i.e if your algebraic rewrite doesn't affect the recursion or loop it won't affect bail out either. Depending on how you implement forced termination, you can likely guarantee this, but it's not very valuable to.
(C) - The alternative isn't real if you allow theoretically bounded loops but with high limits. You can decide not to specify forced termination, but that doesn't mean you don't have it; it simply means the OS or user will terminate the process instead, and reasoning about that is much, much harder. A system with small limits is possible, but those are much less practical to start with. And there have been quite a few systems over the years that tried to impose such limits by design and then it turned out that there were escape-hatched that could be abused to nevertheless impose huge load (if you can do any kind of doubling in an iteration, you don't need many to cause denial-of-service).
(D) - Although it's possible an algebraic rewrite could affect termination, the scope for this is pretty constrained. Either it works, or the whole system fails to terminate; there's no middle ground. That means if you simply assume termination will occur and deal with the code as if it were pure, you'll either end up with a functioning system, or a clearly non-functioning system, but without corruption or any unacceptable uncertainty. (It's possible to shoot yourself in the foot here, but I don't it's possible to do so accidentally).
I mean, if you want to make the argument that all of this is tricky - yes, it sure is; and there are a few risks and some complexity to all this! But simultaneously, I don't think you're going to do a lot better if you want the kind of flexibility that recursion and looping allow. These complexities are pretty manageable in practice; the risks limited. And if you need even tighter guarantees you're going to need to lose most recursion and loops, likely even bounded loops. I've never used it, but earlier in this thread somebody mentioned starlark - and while they clearly tried to avoid turing completeness (loops are bounded, and no recursion by the looks of it), they do allow nested loops with large bounds; i.e., given whatever time you think your OS or user will be willing to wait before pulling the plug you wont hit those bounds: in terms of reasoning, you cannot rely on termination.
But I think that's a decent trade-off. The restrictions needed to reliably terminate in some bounded amount of resources are just too onerous to leave room for a language that can come close to one that does not have those restrictions. As such, it's fine to have either a deterministic, side-effect free language with the risk of (practical) non-termination, or a language with very, very limited looping (e.g. no nesting, and perhaps only constructs like map as opposed to iterating over ranges) - but not much room for anything in between.
Again, the context is the kind of languages you might consider for configuration. And in that context I don't think that turing completeness is all that relevant, compared to determinism and no side-effects, assuming termination. It's those latter aspects that really have a huge impact, and termination mostly in theory, not in practice.
But writing the YAML is fiddly and annoying, so that's a good example of something where it is better to generate it via troposphere (a python module) or some similar system.
To be less specific I guess the answer is that sometimes you don't control both ends - the part that emits and the part that consumes, and having faught ansible, and similar tools, if I can avoid it I'd never want to write YAML by hand for non-trivial purposes if I could script it instead.
Which is the root of a ton of different problems and issues and generally regarded as a bad idea. See pep518 and pyproject.toml vs setup.py
The moment you encode alternation in the configuration file, its time to think about biting the whole Turing complete bullet.
"CEL evaluates in linear time, is mutation free, and not Turing-complete. This limitation is a feature of the language design, which allows the implementation to evaluate orders of magnitude faster than equivalently sandboxed JavaScript."
As mentioned, the goals are security policies (it was first used internally as the Security Rules for Cloud Storage for Firebase and the Cloud Firestore) and proto contracts (e.g. you could define addons to your proto to specify the data matched certain behavior):
I forget the exact syntax for the contract, but it looked something like this...
``` message person { @contract(matches(/* RE2 phone number regex */)) string phone_number = 1; ... } ```
That data could enforce client side checks as well as be used server side (in different implementation languages).
I always wanted to see it combined with the proto to Firebase Security Rules generator (https://firebaseopensource.com/projects/firebaseextended/pro...) to do client and server validation.
Sort of. Starlark doesn't (or at least didn't originally) support recursion or while loops or a number of other structures. There's also a few other differences that make starlark "better" for configs (some immutability is different, there's no such thing as a `class`, etc.)
I still support loops in a configuration language
for x in sequence:
generate_complex_thing(x)
[generate_complex_thing(x) for x in seq]
Cartesian product, map, and reduce operations over finite sets and lists are really handy in configuration. ("For each server in SetA look at each path in SetB and ...") But, if you find yourself starting to write general loops (as opposed to loops implementing map, reduce, and Cartesian product in languages that don't have them built-in), it's a sign you're starting to blur the line between configuration and business logic.
Unit-testing configurations is difficult, especially if they can be non-deterministic (depend on data/time/random()) and aren't modular.
In some sense, all programs with configuration files are really interpreters for the language of their configuration files. (As mentioned before, many of these abstract machines are just push down automata.) Taken too far, the configuration becomes the real program.
I've seen a (now retired) automated trading system with a powerful XML-based configuration language where a few times people got themselves into trouble (and caused trading losses) when their complex tower of configuration fell over. Part of the problem was there existed a few people who weren't trusted to write application logic, but who were trusted to "just update configurations". When the only tool some of your people are allowed to use is a hammer, hammer marks start mysteriously showing up everywhere. Additionally, this was over 10 years ago, and prior to these trading losses, configuration underwent less stringent review. I don't think my experience was atypical.
I've also seen configuration loading get stuck because someone added some code to the config to hit a REST endpoint in the middle of the configuration file. Ideally, you'd leave any I/O to the main program logic, where it's easier to perform the I/O asynchronously, or otherwise non-blocking.
Deterministic non-Turing-complete immutable "executable" configuration languages (or at least ones where it's difficult to get unbounded recursion) tend to be a happy medium. Also, declarative rather than imperative configuration languages tend to be easier to read.
Back when I was a developer in web search infra at Google, I vaguely remember once or twice using a language (maybe Borg's config language, borgconfig) that completely lacked mutability and essentially used object prototyping (A is created as a copy of B, with differences specified at object creation time.)
This ... yes. Being the ops guy backing up second line support at an ISP for a while brought me many examples of this and inspired many in-house tools for them to use instead.