persistence（persistence 搜索）

Persistence Skill

Installation: If not already installed, add the required packages:

• pnpm add @efesto-cloud/entity (for IEntityMapper interface)
• pnpm add @efesto-cloud/maybe (for nullable results)
• pnpm add @efesto-cloud/mongodb-database-context (for transaction support)

Helps you build the persistence layer — document type, repository interface, implementation, and mapper — for a hexagonal architecture TypeScript/MongoDB project following the ports-and-adapters pattern.

Scope: document type, repository interface, repository implementation, mapper. Population (the system for eager-loading related entities via aggregation pipelines) is a separate concern with its own skill — this skill only declares the hook for it where needed, without implementing the internal machinery.

Assumes the entity class and DTO interface already exist and are correct. Read them before writing any persistence code.

Before You Write Anything

1. Read the entity and DTO — identify which fields need type conversion (DateTime → Date, string id → ObjectId, value objects → their serialized form) vs. scalars that pass through unchanged.
2. Scan an existing repo in the project — browse src/repo/, src/repo/impl/, src/mapper/, src/db/Documents/. Match the existing code style: import order, mapper shape, whether they use Foo.create() or new Foo() in mappers.
3. If the project has few existing repo examples, read the reference files — they are fully annotated:

- references/simple-repo-example.ts — no population, uses .find()/.findOne() directly - references/aggregate-repo-example.ts — optional population, aggregation pipeline, streaming, bulk save - references/di-wiring-example.md — the exact 4 files to update for DI registration

What Gets Created

For a new entity Foo stored in collection foo:

Plus 4 existing files to update for DI registration (see DI Wiring).

Document Type

The document type is the MongoDB-level view of the entity. It's a pure TypeScript type alias — not a class, not a schema.

// src/db/Documents/FooDocument.ts
import { ObjectId } from "mongodb";
import IFoo from "~/dto/IFoo.js";
import BarDocument from "./BarDocument.js";

type FooDocument = Overwrite<IFoo, {
    _id: ObjectId;              // DTO has string → document has ObjectId
    deleted_at: Date | null;    // DTO has string|null → document has JS Date|null
    owner_id: ObjectId | null;  // FK: DTO string → document ObjectId
    bar?: BarDocument | null;   // optional — only present after aggregation $lookup
    items?: ItemDocument[];     // optional array — only after aggregation
}>;

export default FooDocument;

Rules:

• Start from Overwrite<IDTO, {...}> and override only what differs from the DTO.
• If some DTO fields are computed (never stored in MongoDB), wrap with Omit<IFoo, "computed_field"> before overriding: Overwrite<Omit<IFoo, "labels">, {...}>.
• _id is always ObjectId.
• DateTime in entity/DTO becomes Date in document; FK fields (*_id) become ObjectId.
• Populated sub-documents are always optional (?) — they're absent on raw stored documents, present only when an aggregation pipeline joined them.
• Polymorphic collections use discriminated unions: type XDocument = ADocument | BDocument.

Repository Interface

// src/repo/IFooRepo.ts
import { Maybe } from "@efesto-cloud/maybe";
import { ObjectId } from "mongodb";
import Foo from "~/entity/Foo.js";

// Export the search query type alongside the interface
export type SearchFoo = {
    name?: string;
    include_deleted?: boolean;
};

interface IFooRepo {
    search(query: SearchFoo): Promise<Foo[]>;   // empty array when nothing matches
    get(id: ObjectId): Promise<Maybe<Foo>>;     // Maybe.none() when not found
    save(entity: Foo): Promise<void>;
}

export default IFooRepo;

Return type guide:

When population is needed — declare an options parameter and a namespace with an Options type. The population skill handles everything else; the interface just exposes the hook:

import type { Populate } from '@efesto-cloud/population';
import type { FooShape } from './shape/FooShape.js';

interface IFooRepo {
    search(query: SearchFoo, options?: IFooRepo.Options): Promise<Foo[]>;
    get(id: ObjectId, options?: IFooRepo.Options): Promise<Maybe<Foo>>;
    save(entity: Foo): Promise<void>;
}

namespace IFooRepo {
    export type Options = {
        populate?: Populate<FooShape>; // Populate + FooShape come from the population system
    };
}

Repository Implementation

// src/repo/impl/FooRepoImpl.ts
import { Maybe } from "@efesto-cloud/maybe";
import { inject, injectable } from "inversify";
import { Collection, Filter, ObjectId } from "mongodb";
import type IDatabaseContext from "~/db/Context/IDatabaseContext.js";
import FooDocument from "~/db/Documents/FooDocument.js";
import Symbols from "~/di/Symbols.js";
import Foo from "~/entity/Foo.js";
import FooMapper from "~/mapper/FooMapper.js";
import IFooRepo, { SearchFoo } from "../IFooRepo.js";

@injectable()
export default class FooRepoImpl implements IFooRepo {
    constructor(
        @inject(Symbols.Collections.foo) private readonly coll: Collection<FooDocument>,
        @inject(Symbols.DatabaseContext) private readonly db: IDatabaseContext,
    ) {}

    async search(query: SearchFoo): Promise<Foo[]> {
        const filter: Filter<FooDocument> = {};
        if (query.name) filter.name = new RegExp(`^${query.name}`, "i");
        if (!query.include_deleted) filter.deleted_at = null;

        const docs = await this.coll
            .find(filter, { session: this.db.session }) // session is required — wires into transactions
            .sort({ name: 1 })
            .toArray();
        return docs.map(FooMapper.from);
    }

    async get(id: ObjectId): Promise<Maybe<Foo>> {
        const doc = await this.coll.findOne({ _id: id }, { session: this.db.session });
        return Maybe.maybe(doc).map(FooMapper.from);
    }

    async save(entity: Foo): Promise<void> {
        const raw = FooMapper.to(entity);
        await this.coll.updateOne(
            { _id: raw._id },
            { $set: raw },
            { upsert: true, session: this.db.session },
        );
    }
}

The session rule — every MongoDB driver call must include {session: this.db.session}. Without it, the call silently runs outside any active transaction. IDatabaseContext exposes session as undefined when there's no transaction; the MongoDB driver ignores undefined gracefully, so it's always safe to pass.

Simple vs. aggregate reads:

• Use .find()/.findOne() for straightforward queries with no population.
• When the repo supports population, use an aggregation pipeline built via XQueryBuilder. The query builder is part of the population system — see references/aggregate-repo-example.ts for the pattern.

Mapper

The mapper is a plain object (not a class) that transforms between entity and document.

// src/mapper/FooMapper.ts
import { IEntityMapper } from "@efesto-cloud/entity";
import { DateTime } from "luxon";
import FooDocument from "~/db/Documents/FooDocument.js";
import Foo from "~/entity/Foo.js";

const FooMapper: IEntityMapper<Foo, FooDocument> = {
    /**
     * from: document → entity (read path)
     * Convert MongoDB types to domain types. Patch in populated sub-documents after construction.
     */
    from: (doc: FooDocument): Foo => {
        const entity = new Foo({
            name: doc.name,
            owner_id: doc.owner_id,
            deleted_at: doc.deleted_at
                ? DateTime.fromJSDate(doc.deleted_at) as DateTime<true>
                : null,
            bar: null,   // default; overwritten below if the pipeline populated it
        }, doc._id);

        if (doc.bar) entity.props.bar = BarMapper.from(doc.bar);
        if (doc.items) entity.props.items = doc.items.map(ItemMapper.from);

        return entity;
    },

    /**
     * to: entity → document (write path)
     * Only include fields the collection actually stores. Populated join results are never written back.
     */
    to: (domain: Foo): FooDocument => ({
        _id: domain._id,
        name: domain.props.name,
        owner_id: domain.props.owner_id,
        deleted_at: domain.deleted_at?.toJSDate() ?? null,
    }),
};

export default FooMapper;

from vs. to asymmetry — from is read-time and may encounter in-place populated sub-documents from an aggregation; construct the entity, then patch them in. to is write-time; serialize only own stored scalar fields and FK ObjectIds — never populated relations.

new Foo() vs. Foo.create() — in mappers, the direct constructor is appropriate because you have complete, already-validated stored state and don't need create()'s defaults. Use Foo.create() in use cases where you're working with partial user input.

Two mapper styles in the wild — some projects use namespace FooMapper {export function from(...)}. Match what already exists in the project.

DI Wiring

Touch these 4 files when adding a new collection. See references/di-wiring-example.md for exact code snippets.

1. src/enum/CollectionNameEnum.ts — add foo = "foo"
2. src/db/ICollectionsDocument.ts — add foo: FooDocument
3. src/di/Symbols.ts — add FooRepo: Symbol.for("FooRepo") in the Repo section
4. src/di/container.ts — add container.bind(Symbols.Repo.FooRepo).to(FooRepoImpl).inRequestScope()

Symbols.Collections is auto-generated from CollectionNameEnum — you do not need to touch it manually.

Special Cases

Soft-delete

Entity has deleted_at: DateTime<true> | null. Records stay in the collection; filtered omitting deleted by default.

// In save():
if (entity.isDeleted()) {
    await this.coll.updateOne(
        { _id: raw._id },
        { $set: { deleted_at: entity.deleted_at!.toJSDate() } },
        { session: this.db.session },
    );
} else {
    await this.coll.updateOne({ _id: raw._id }, { $set: raw }, { upsert: true, session: this.db.session });
}

// In search() filter:
if (!query.include_deleted) filter.deleted_at = null;

Hard-delete

if (entity.isDeleted()) {
    await this.coll.deleteOne({ _id: raw._id }, { session: this.db.session });
} else {
    await this.coll.updateOne({ _id: raw._id }, { $set: raw }, { upsert: true, session: this.db.session });
}

Bulk save via saveMany

import prepareBulkOps from "~/db/prepareBulkOps.js";

async saveMany(entities: Foo[]): Promise<void> {
    const ops = prepareBulkOps(entities, FooMapper);
    if (!ops.length) return;
    await this.coll.bulkWrite(ops, { session: this.db.session });
}

Saving child entities from a parent repo

When a parent's save() must also persist child entities in their own collection:

constructor(
    @inject(Symbols.Collections.foo) private readonly coll: Collection<FooDocument>,
    @inject(Symbols.Repo.BarRepo) private readonly barRepo: IBarRepo,  // inject child repo
    @inject(Symbols.DatabaseContext) private readonly db: IDatabaseContext,
) {}

async save(entity: Foo): Promise<void> {
    const raw = FooMapper.to(entity);
    await this.coll.updateOne({ _id: raw._id }, { $set: raw }, { upsert: true, session: this.db.session });
    await this.barRepo.saveMany(entity.bars); // delegate to child repo — same transaction session
}

Polymorphic collections

When one MongoDB collection stores multiple entity variants:

// Document: discriminated union
type FooDocument = FooADocument | FooBDocument; // each has `type: "A" | "B"` discriminator

// Mapper: dispatch on doc.type
from: (doc: FooDocument): Foo => {
    if (doc.type === "A") return FooAMapper.from(doc as FooADocument);
    if (doc.type === "B") return FooBMapper.from(doc as FooBDocument);
    throw new Error(`Unknown Foo type: ${(doc as any).type}`);
}

Streaming large result sets

Use a Readable with cursor iteration rather than .toArray() to avoid loading the full collection into memory. See references/aggregate-repo-example.ts for the complete streaming implementation pattern.

Checklist — New Repository

• Read the entity + DTO before writing anything
• FooDocument.ts created with Overwrite<IFoo, {_id: ObjectId;...}>
• IFooRepo.ts created with interface + exported search type
• FooRepoImpl.ts created — @injectable(), every call passes {session: this.db.session}
• FooMapper.ts created — from() handles optional populated fields; to() only own stored scalars
• CollectionNameEnum updated
• ICollectionsDocument updated
• Symbols.Repo.FooRepo added
• container.bind(...).inRequestScope() added
• run typecheck

Checklist — Modifying Existing

• Read all four files before changing anything
• New field → update FooDocument + FooMapper.from() + FooMapper.to() + interface if the signature changes
• run typecheck