eQuantic.Core.Data.MongoDb 5.6.0

eQuantic.Core.Data

A provider-agnostic Repository + Unit of Work for .NET, where filtering, sorting and paging are authored typed and fluent — not as stringly-typed configuration.

var repo = unitOfWork.GetAsyncRepository<OrderData, Guid>();

var page = await repo.GetPagedAsync(
    PageRequest.Of(pageIndex: 1, pageSize: 20),
    new QueryOptions<OrderData>()
        .Where(o => o.Total, FilterOperator.GreaterThan, 100m)
        .And(o => o.Customer.Name, FilterOperator.Contains, term)
        .OrderByDescending(o => o.CreatedAt)
        .Include(nameof(OrderData.Customer))
        .NoTracking());

// page is a PagedResult<OrderData>: Items + TotalCount + PageIndex/PageSize/PageCount + Has*Page

Why

The Repository pattern keeps your domain ignorant of the persistence engine — you code against IRepository<TEntity, TKey>, and the Entity Framework (or other) provider supplies the implementation. What usually rots is the query surface: dozens of GetPaged/GetFiltered overloads and Action<Configuration> callbacks, with filters and sorts passed as magic strings.

eQuantic.Core.Data v5 collapses that: one method per operation, each taking a single QueryOptions<TEntity> that you compose fluently and typed, backed by the eQuantic.Linq query engine.

Getting started

This package is the contracts — pair it with a provider that supplies the engine:

dotnet add package eQuantic.Core.Data
dotnet add package eQuantic.Core.Data.EntityFramework.SqlServer   # or PostgreSql / MySql / MongoDb / CosmosDb

Give your entities a key via IEntity<TKey> (the key is exposed through GetKey/SetKey — no mandated Id property):

using eQuantic.Core.Data.Repository;

public class OrderData : IEntity<Guid>
{
    public Guid Id { get; set; }
    public decimal Total { get; set; }
    public Guid GetKey() => Id;
    public void SetKey(Guid key) => Id = key;
}

Register the provider's unit of work and repositories — the wiring is provider-specific (see the Entity Framework Core provider) — then inject IUnitOfWork, ask it for a repository, and query with a QueryOptions (the snippet at the top).

How you query

QueryOptions<TEntity> mirrors the eQuantic.Linq query builders, so filters read like code and fail at compile time — not at runtime:

new QueryOptions<OrderData>()
    .Where(o => o.Total, FilterOperator.GreaterThanOrEqual, 100m)   // typed member selector
    .And(o => o.Status, FilterOperator.Equal, OrderStatus.Paid)     // clauses fold left to right:
    .Or(o => o.Customer.IsVip, FilterOperator.Equal, true)          //   (total>=100 AND paid) OR vip
    .OrderByDescending(o => o.CreatedAt)
    .ThenBy("customer.name");                                        // string path for dynamic columns

You reach for whichever filter form fits — all end up as one predicate the provider translates:

The query-string grammar behind the string forms (total:gt(100),status:eq(Paid)) is documented in the eQuantic.Linq query-string reference.

Paging that tells you what you got

PagedResult<OrderData> page = await repo.GetPagedAsync(PageRequest.Of(2, 20), options);
// page.Items, page.TotalCount, page.PageIndex, page.PageSize, page.PageCount,
// page.HasPreviousPage, page.HasNextPage

PageRequest is one-based (Skip/Take derived); PagedResult<T> carries the items and the totals — no second count call, no bare IEnumerable<T>.

Provider-agnostic by design

This package is the contracts (IRepository, IUnitOfWork, QueryOptions, PageRequest, PagedResult, specifications). The persistence engine is kept out of the type signatures — IRepository<TEntity, TKey>, not IRepository<TUnitOfWork, TEntity, TKey>. The Entity Framework implementation lives in the provider packages (eQuantic.Core.Data.EntityFramework and friends).

Native providers — MongoDB, Azure Cosmos DB, Apache Cassandra

The same contracts implemented directly on each official driver — no Entity Framework — with a lean write model: Add/Modify/Remove buffer typed writes and one native batch runs on Commit (no change tracking or snapshotting). Explicit transactions are opt-in: multi-document sessions on MongoDB (reads inside the transaction see its writes), a single-partition TransactionalBatch on Cosmos, an atomic LOGGED BATCH on Cassandra. All three ship the same fluent, typed migrations, authored with member selectors instead of field strings:

services.AddMongoRepositories("mongodb://localhost:27017", "shop");
services.AddMongoMigrations(typeof(AddProductIndexes).Assembly);

[Migration("Product indexes", 2026, 7, 20, 14, 0, 0)]
public sealed class AddProductIndexes : Migration
{
    public override void Up(IMigrationBuilder migration) =>
        migration.For<Product>(product => product
            .EnsureCollection()
            .Index(x => x.Category)
            .CompositeIndex(keys => keys.Descending(x => x.Price).Ascending(x => x.Name)));
}

// on startup: apply pending migrations, once each
await serviceProvider.GetRequiredService<IMigrationRunner>().RunAsync();

The engine — what the providers add on top of the drivers

• Pushdown + residual filtering (Cassandra) — every clause CQL can express runs on the cluster; the ones it cannot (OR across columns, !=, NULL, arbitrary predicates) run client-side over the fetched rows behind the explicit .AllowClientEvaluation() opt-in, with .AllowFiltering() gating scans. An OR whose branches each pin a partition needs no opt-in at all: it runs as parallel native queries, merged and de-duplicated by primary key.
• Explain() on every provider (IExplainableRepository<T>) — the native statement (CQL, Cosmos SQL, aggregation pipeline), what is pushed down versus evaluated client-side, partition scoping, and the opt-ins a query still needs. Builds the plan without executing anything.
• Computed set-based updates — UpdateMany(filter, x => new E { N = x.N + 1, Tags = x.Tags.Append("vip").ToList() }) renders to the store's atomic form: $inc/$mul/$push/$addToSet/$pullAll on MongoDB (honouring [BsonElement] and custom serializers), PatchOperation.Increment/Add on Cosmos, collection col = col + ? and counter columns on Cassandra. Shapes a store cannot apply atomically are rejected with the reason — an update never silently degrades to read-modify-write.
• Continuation paging and streaming (IContinuationReadRepository<T>, IStreamingReadRepository<T>) — deep pages walk the native path (Cassandra PagingState, Cosmos continuation tokens, MongoDB keyset by id), every page costing the same as the first, and IAsyncEnumerable<T> streams hold one page in memory at a time.
• Partition-key inference and ETag concurrency (Cosmos) — a filter that pins the partition key scopes the query to a single partition automatically (the biggest RU saving there is), and ConcurrencyToken(x => x.ETag) turns Modify into a conditional If-Match replace.
• Prepared statements and LWT (Cassandra) — every repeated CQL text is prepared once per session and bound thereafter; AddIfNotExistsAsync runs INSERT … IF NOT EXISTS and reports whether it applied; WithConsistency(...) sets a per-query consistency level and CommitAsync(o => o.WithTtl(...)) writes expiring rows.
• Global query filters (multi-tenancy) — register new QueryFilters().For<Order>(...) as a singleton and every read and set-based write is scoped by it on all three providers; a per-request factory receives the scope's IServiceProvider (resolve the current tenant there), and IgnoringQueryFilters() opts a read out. A tenant filter on the Cosmos partition key auto-scopes the query to that partition.
• OpenTelemetry built in — subscribe to the eQuantic.Core.Data ActivitySource and every provider emits spans with the statement (placeholders, never values) plus the engine's own facts: client evaluation, split-query count, partition scoping, staged write counts.

The rule behind all of it: a query never lies about its cost. Anything beyond the store's native access path is explicit and opt-in, and Explain() shows exactly what will run. Providers target net10.0.

Install

dotnet add package eQuantic.Core.Data

Targets net8.0 and net10.0. Depends only on the framework-free eQuantic.Linq.Web and eQuantic.Linq.Specification.

Learn more

• Repository Pattern walkthrough — a full example: data entities, unit of work, repository, specifications and domain services.
• v5 contracts design — the rationale, the consolidated interface and the breaking-change/migration summary.
• Releasing — the automated release flow (maintainers).

Upgrading to v5

v5 is a deliberate breaking redesign: one QueryOptions argument per read (not Action<Configuration> + overloads), PagedResult<T> paging, the unit-of-work type parameter removed from IRepository/GetRepository, an IEntity<TKey> constraint (no new()), the SQL/relational surface moved to the provider layer, and the monolithic eQuantic.Linq dependency replaced by eQuantic.Linq.Web + eQuantic.Linq.Specification. The full before/after is in the design doc.

MIT © eQuantic Tech