[data.search.aggs]: Create agg types function for terms agg.

[lukeelmers]

related #61768, #61769

Summary

First pass at creating an agg type function. I went with terms because it is one of the more complex ones. Once we decide we are happy with this approach, we'll create functions for all other aggs, and then refactor esaggs to use them.

The goal here is to make sure we are comfortable with the general approach, before we repeat it for a bunch of aggs 😄

I tried to implement this in a way that makes it as easy as possible to "scale" and add new functions, since we will need to add a bunch of them (one for each agg type). At this point adding a new function looks like this:

1. Create an interface for the params of the agg type you are writing a function for
2. Add the interface to AggParamsMapping in types.ts
3. Write the function, using the interface you wrote in step 1
4. Write tests for the function
5. Import your function in agg_types.ts and add it to getAggTypesFunctions
6. Rinse and repeat for each of the ~30ish aggs 💦

cc @ppisljar @alexwizp

[elasticmachine]

Pinging @elastic/kibana-app-arch (Team:AppArch)

[lukeelmers]

This is the shape of the agg_type which every agg type function returns. That way we can enforce that esaggs only accept agg_type as an argument, instead of a stringified JSON blob.

[lukeelmers]

This generic helps to streamline the process of typing arguments for an expression function. It uses the provided Name to retrieve the params for the agg from the AggParamsMapping below. Then it combines those with id, enabled, and schema from the agg config options -- these arguments will be required for each of the agg type expression functions, however the params will be different between each.

[lukeelmers]

We can add each new agg type here, along with the params for that agg. This way you can access all the agg params from this mapping.

Eventually we can make this interface public, and folks could even add to it using declare module to append their agg to the interface. (This would be necessary if, for example, there were a special type of aggregation which was only provided by x-pack).

[lukeelmers]

@elasticmachine merge upstream

[ppisljar]

we also need a way to get from aggConfig to expression function. once we have expression builder ready:

const as = agg.toJSON();
createExpression().addFunction(as.type, as.params)

can we assume that serialized params are exact params to the expression function ?
with terms the problem is with orderAgg parameter, which in expression function should be a subexpression. if serialized params and expression function params are different we will need to extend agg type to return expressionparams :|

how do you feel about adding .toAST() function to aggConfig ?

[alexwizp]

I see you tried to do it in a way like we do for aggTypes. But maybe we can create a factory method (or you can use some other pattern) to encapsulate forEach methods and make the setup hook easier for reading. It will be awesome if you will do it for aggTypes and aggFuncitons

[lukeelmers]

@alexwizp Just to make sure I follow; how would you envision this looking? Just removing the forEach from search_service entirely & passing expressions down so it is handled inside of the function instead?

[lukeelmers]

The reason I ask is because readability is what I was aiming for here, but of course I'm always open to ideas to make things more readable. Rather than having the magic of stuff getting registered in another function, I liked the idea of seeing very clearly what this service was registering during setup.

[lukeelmers]

As a start I removed the bucket/metric differentiation per Peter's suggestion; lmk what your thoughts are and I'm happy to adjust further

[alexwizp]

It's not clear enough why you created that interface in buckets/terms.ts not in buckets/terms_fn.ts
From my point of view we can simplify your approach and remove AggExpressionFunctionArgs and AggParamsMapping types at all. I don't see any benefits for today.

what about to do all buckets/terms_fn.ts:

....
import { AggExpressionType, AggConfigJson, IAggExpressionFunction } from '../';

const aggTypeName = 'terms';
const fnName = 'aggTypeTerms';

type Input = any;
type Output = AggExpressionType;

export interface AggParamsTerms extends IAggExpressionFunction {
  field: string;
  order: 'asc' | 'desc';
  orderBy: string;
  orderAgg?: AggConfigJson;
  size?: number;
  missingBucket?: boolean;
  missingBucketLabel?: string;
  otherBucket?: boolean;
  otherBucketLabel?: string;
  // advanced
  exclude?: string;
  include?: string;
}

type FunctionDefinition = ExpressionFunctionDefinition<
  typeof fnName,
  Input,
  AggParamsTerms,
  Output
>;

export const aggTypeTerms = (): FunctionDefinition => ({
....

[lukeelmers]

Part of the reason I put this with terms.ts & created the AggParamsMapping was because I felt that having typings for each of the agg type params would be generally useful outside of the expression functions & didn't have anything to do with expressions in particular.

For example, say I want to create a new agg config. Without the mapping, I could do:

const configState = {
  type: 'terms',
  enabled: true,
  params: {
    field: 'machine.os.keyword',
    // ...etc
  }
};
data.search.aggs.createAggConfigs(indexPattern, [configState]);

But configState is only enforced to match the general shape of params (currently pretty useless: Record<string, any>). What would be ideal would be to have the ability to pull in an interface for the params you are using so that you can have more type safety in your params. Perhaps by making the serialized agg config interface a generic:

const configState: AggConfigSerialized<AggParamsTerms> = {
  type: 'terms',
  enabled: true,
  params: {
    field: 'machine.os.keyword',
    // ...now this would be more strictly typed
  }
};

Or even better, using the type from the global AggParamsMapping based on the type name string.

More concretely, as we have an agg types registry which technically could have anything registered to it, having a mapping of these interfaces that others can add to via declare module will be necessary eventually anyway.

[lukeelmers]

@ppisljar

we also need a way to get from aggConfig to expression function. once we have expression builder ready

Oops, totally forgot about this, thanks for the reminder.

how do you feel about adding .toAST() function to aggConfig ?

IMHO we should start with a separate little utility function that takes in the aggConfig & returns the expression AST. This would be easy to remove later when the expression builder is ready, at which point folks can just use toJSON() (or serialized()). Adding toAST to aggconfigs feels like mixing concerns a little more than we probably need to, but I'm not strongly opposed to it if you think it's the right path.

can we assume that serialized params are exact params to the expression function ?

That is my assumption. The expression args are serialized params (which differ on a per-agg basis), and enabled, id, + schema. As you pointed out, I think it'd be problematic/confusing if there was a divergence between expression args & params. This is why I'd like to eventually change how we are typing the agg params so that they are using the same interface as the expression fn is using for the args. (Edit: Also a param of type agg gets serialized by calling toJSON in the param type)

with terms the problem is with orderAgg parameter, which in expression function should be a subexpression.

I'm looking into this, but even if it were necessary i think it would be okay, since the subexpression could just be another agg type expression function with the serialized info for that agg.

[ppisljar]

code LGTM, left two small comments. once we have AST builder ready this can be cleaned up (should probably the first case for using AST builder).

[kibanamachine]

💚 Build Succeeded

• continuous-integration/kibana-ci/pull-request
• Commit: e17a097

History

• 💚 Build #42624 succeeded 0588f1e
• 💚 Build #42331 succeeded 6296718
• 💔 Build #42299 failed a0ac52f
• 💔 Build #42066 failed 09d97e5
• 💚 Build #41382 succeeded 37b7af7

To update your PR or re-run it, just comment with:
@elasticmachine merge upstream