Overview

Master Ecto's powerful Query DSL to build efficient, composable database queries. This skill covers the query syntax, filtering, joining, aggregation, preloading associations, and advanced query composition patterns.

Basic Query with from Macro

import Ecto.Query, only: [from: 2]

# Basic query using keyword syntax
query = from u in "users",
          where: u.age > 18,
          select: u.name

# Execute the query
MyApp.Repo.all(query)

Queries are built using the from/2 macro and only sent to the database when passed to a Repo function like all/1, one/1, or get/2. The keyword syntax provides a readable way to construct queries.

Query with Schema Module

query = from u in MyApp.User,
          where: u.age > 18,
          select: u.name

MyApp.Repo.all(query)

Using a schema module instead of a table name string provides better type safety and allows Ecto to use the schema's field definitions for validation and casting.

Bindingless Query Construction

from MyApp.Post,
  where: [category: "fresh and new"],
  order_by: [desc: :published_at],
  select: [:id, :title, :body]

Bindingless syntax allows building queries without explicit variable bindings. This works well for simple queries and when using keyword list syntax for conditions.

Query with Explicit Bindings

query = from p in MyApp.Post,
          where: p.category == "fresh and new",
          order_by: [desc: p.published_at],
          select: struct(p, [:id, :title, :body])

MyApp.Repo.all(query)

Explicit bindings (like p for posts) allow for more complex conditions and selections. The struct/2 function selects only specific fields from the schema.

Dynamic Query Variables

category = "fresh and new"
order_by = [desc: :published_at]
select_fields = [:id, :title, :body]

query = from MyApp.Post,
          where: [category: ^category],
          order_by: ^order_by,
          select: ^select_fields

MyApp.Repo.all(query)

The pin operator ^ allows interpolating Elixir values into queries. This is essential for parameterized queries and prevents SQL injection.

Where Clause with Expressions

query = from u in MyApp.User,
          where: u.age > 0,
          select: u.name

# Multiple where clauses are combined with AND
query = from u in MyApp.User,
          where: u.age > 18,
          where: u.confirmed == true,
          select: u

MyApp.Repo.all(query)

Query expressions support field access, comparison operators, and literals. Multiple where clauses are automatically combined with AND logic.

Composable Queries

# Create a base query
query = from u in MyApp.User, where: u.age > 18

# Extend the query
query = from u in query, select: u.name

MyApp.Repo.all(query)

Queries are composable - you can build on existing queries by using them in the in clause. This enables powerful query abstraction and reusability.

Query Composition Function Pattern

def most_recent_from(query, minimum_date) do
  from p in query,
    where: p.published_at > ^minimum_date,
    order_by: [desc: p.published_at]
end

# Usage
MyApp.Post
|> most_recent_from(~N[2024-01-01 00:00:00])
|> MyApp.Repo.all()

Extracting query logic into functions creates reusable, testable query components. This pattern is fundamental to building maintainable query code.

Or Where Conditions

from p in MyApp.Post,
  where: p.category == "elixir" or p.category == "phoenix",
  select: p

Use the or keyword for alternative conditions. For more complex OR logic, consider using Ecto.Query.dynamic/2.

IN Query with List

categories = ["elixir", "phoenix", "ecto"]

query = from p in MyApp.Post,
          where: p.category in ^categories,
          select: p

MyApp.Repo.all(query)

The in operator checks if a field value exists in a list of values. Use the pin operator to interpolate the list variable.

Like and ILike for Pattern Matching

search_term = "%elixir%"

query = from p in MyApp.Post,
          where: like(p.title, ^search_term),
          select: p

# Case-insensitive version
query = from p in MyApp.Post,
          where: ilike(p.title, ^search_term),
          select: p

Use like/2 for case-sensitive pattern matching and ilike/2 for case-insensitive matching. Wildcards % match any characters.

Selecting Specific Fields

# Select multiple fields
query = from p in MyApp.Post,
          select: {p.id, p.title}

MyApp.Repo.all(query)  # Returns [{1, "Title 1"}, {2, "Title 2"}]

# Select as map
query = from p in MyApp.Post,
          select: %{id: p.id, title: p.title}

MyApp.Repo.all(query)  # Returns [%{id: 1, title: "Title 1"}, ...]

# Select struct with specific fields
query = from p in MyApp.Post,
          select: struct(p, [:id, :title, :body])

MyApp.Repo.all(query)  # Returns Post structs with only selected fields loaded

Selecting specific fields instead of entire records improves query performance by reducing data transfer and memory usage.

Aggregation Functions

# Count records
query = from p in MyApp.Post,
          select: count(p.id)

MyApp.Repo.one(query)  # Returns integer count

# Average
query = from p in MyApp.Post,
          select: avg(p.rating)

# Sum
query = from o in MyApp.Order,
          select: sum(o.total)

# Min and Max
query = from p in MyApp.Product,
          select: {min(p.price), max(p.price)}

Ecto supports standard SQL aggregation functions including count/1, avg/1, sum/1, min/1, and max/1.

Group By and Having

query = from p in MyApp.Post,
          group_by: p.category,
          select: {p.category, count(p.id)}

MyApp.Repo.all(query)  # Returns [{"elixir", 10}, {"phoenix", 5}]

# With having clause
query = from p in MyApp.Post,
          group_by: p.category,
          having: count(p.id) > 5,
          select: {p.category, count(p.id)}

Use group_by to group results by field values and having to filter groups based on aggregate values.

Order By

# Single field ascending
query = from p in MyApp.Post,
          order_by: p.published_at

# Single field descending
query = from p in MyApp.Post,
          order_by: [desc: p.published_at]

# Multiple fields
query = from p in MyApp.Post,
          order_by: [desc: p.published_at, asc: p.title]

# With nulls positioning
query = from p in MyApp.Post,
          order_by: [desc_nulls_last: p.published_at]

The order_by option controls result ordering. You can specify ascending or descending order, multiple fields, and null positioning.

Limit and Offset for Pagination

# Simple limit
query = from p in MyApp.Post,
          limit: 10

# With offset for pagination
page = 2
per_page = 10

query = from p in MyApp.Post,
          order_by: [desc: p.published_at],
          limit: ^per_page,
          offset: ^((page - 1) * per_page)

MyApp.Repo.all(query)

Use limit and offset for pagination. Always include an order_by clause to ensure consistent pagination results.

Inner Join

query = from p in MyApp.Post,
          join: c in MyApp.Comment,
          on: c.post_id == p.id,
          select: {p.title, c.body}

MyApp.Repo.all(query)

Inner joins return only records that have matching records in both tables. The on clause specifies the join condition.

Join with assoc Helper

query = from p in MyApp.Post,
          join: c in assoc(p, :comments),
          select: {p, c}

MyApp.Repo.all(query)

The assoc/2 helper uses the association definition from your schema, making joins more maintainable and less error-prone than manually specifying foreign keys.

Left Join

query = from p in MyApp.Post,
          left_join: c in assoc(p, :comments),
          select: {p, c}

MyApp.Repo.all(query)

Left joins return all records from the left table (posts) even if there are no matching records in the right table (comments). Unmatched fields are nil.

Preload Associations

# Preload in separate query
MyApp.Repo.all(from p in MyApp.Post, preload: [:comments])

# Preload multiple associations
MyApp.Repo.all(from p in MyApp.Post, preload: [:comments, :author])

# Nested preload
MyApp.Repo.all(from p in MyApp.Post, preload: [:author, comments: :likes])

Preloading fetches associated data efficiently, preventing N+1 query problems. Separate query preloading is simpler but may require more database round trips.

Preload with Join

query = from p in MyApp.Post,
          join: c in assoc(p, :comments),
          where: c.published_at > p.updated_at,
          preload: [comments: c]

MyApp.Repo.all(query)

When you join an association and want to filter it, you can preload the joined data using the binding variable. This creates a single, more efficient query.

Complex Nested Preload with Joins

query = from p in MyApp.Post,
          join: c in assoc(p, :comments),
          join: l in assoc(c, :likes),
          where: l.inserted_at > c.updated_at,
          preload: [:author, comments: {c, likes: l}]

MyApp.Repo.all(query)

You can preload multiple levels of nested associations while maintaining join filters. The nested tuple syntax preserves the join bindings.

Preload After Query

posts = MyApp.Repo.all(MyApp.Post)
posts_with_comments = MyApp.Repo.preload(posts, :comments)

# Preload with custom query
comments_query = from c in MyApp.Comment, order_by: [desc: c.inserted_at]
posts_with_recent_comments = MyApp.Repo.preload(posts, comments: comments_query)

The preload/2 function can preload associations after fetching records. You can also customize the preload query for fine-grained control.

Subquery

# Define subquery
subquery = from p in MyApp.Post,
             where: p.published == true,
             select: %{category: p.category, count: count(p.id)},
             group_by: p.category

# Use subquery
query = from s in subquery(subquery),
          where: s.count > 10,
          select: s.category

MyApp.Repo.all(query)

Subqueries allow using the result of one query as input to another, enabling complex analytical queries.

Fragment for Raw SQL

# Use SQL fragment
query = from p in MyApp.Post,
          where: fragment("lower(?)", p.title) == "elixir",
          select: p

# Fragment with parameters
search = "elixir"
query = from p in MyApp.Post,
          where: fragment("lower(?) LIKE ?", p.title, ^"%#{search}%"),
          select: p

The fragment/1 function allows embedding raw SQL in queries when Ecto's DSL doesn't support a specific database feature. Use sparingly as it reduces portability.

Query Hints

query = from p in MyApp.Post,
          hints: ["USE INDEX FOO"],
          where: p.title == "title"

# Multiple hints
query = from p in MyApp.Post,
          hints: "TABLESAMPLE SYSTEM(1)"

# Dynamic hints
sample = "SYSTEM_ROWS(1)"
query = from p in MyApp.Post,
          hints: ["TABLESAMPLE", unsafe_fragment(^sample)]

Query hints provide database-specific optimization instructions like index usage or table sampling. Hints are database-specific and may not be portable.

Dynamic Query Building

defmodule MyApp.PostQueries do
  import Ecto.Query

  def filter(query \\ MyApp.Post, filters) do
    query
    |> filter_by_category(filters[:category])
    |> filter_by_published(filters[:published])
    |> filter_by_search(filters[:search])
  end

  defp filter_by_category(query, nil), do: query
  defp filter_by_category(query, category) do
    from p in query, where: p.category == ^category
  end

  defp filter_by_published(query, nil), do: query
  defp filter_by_published(query, published) do
    from p in query, where: p.published == ^published
  end

  defp filter_by_search(query, nil), do: query
  defp filter_by_search(query, search) do
    from p in query, where: ilike(p.title, ^"%#{search}%")
  end
end

# Usage
filters = %{category: "elixir", published: true, search: "ecto"}
MyApp.PostQueries.filter(filters) |> MyApp.Repo.all()

Building queries dynamically allows handling optional filters and complex search criteria. Pattern matching on nil values keeps the code clean and readable.

Ecto.Query.dynamic for Complex Conditions

defmodule MyApp.PostQueries do
  import Ecto.Query

  def search(filters) do
    MyApp.Post
    |> where(^build_where_clause(filters))
    |> MyApp.Repo.all()
  end

  defp build_where_clause(filters) do
    Enum.reduce(filters, dynamic(true), fn
      {:category, value}, dynamic ->
        dynamic([p], ^dynamic and p.category == ^value)

      {:published, value}, dynamic ->
        dynamic([p], ^dynamic and p.published == ^value)

      {:min_rating, value}, dynamic ->
        dynamic([p], ^dynamic and p.rating >= ^value)

      _, dynamic ->
        dynamic
    end)
  end
end

The dynamic/2 macro builds query fragments that can be composed at runtime. This is more flexible than string-based query building and prevents SQL injection.

Distinct Queries

# Distinct on all selected fields
query = from p in MyApp.Post,
          distinct: true,
          select: p.category

# Distinct on specific fields
query = from p in MyApp.Post,
          distinct: [desc: p.published_at],
          select: p

The distinct option removes duplicate rows from results. You can specify which fields to use for determining uniqueness.

Union Queries

posts_query = from p in MyApp.Post,
                where: p.published == true,
                select: %{type: "post", title: p.title}

pages_query = from p in MyApp.Page,
                where: p.active == true,
                select: %{type: "page", title: p.title}

# Union
query = posts_query |> union(^pages_query)
MyApp.Repo.all(query)

# Union all (includes duplicates)
query = posts_query |> union_all(^pages_query)

Union combines results from multiple queries. Use union/2 to remove duplicates or union_all/2 to keep them.

Locking for Concurrency Control

# Pessimistic locking
query = from p in MyApp.Post,
          where: p.id == ^post_id,
          lock: "FOR UPDATE"

post = MyApp.Repo.one(query)

# Optimistic locking (using version field in schema)
changeset = MyApp.Post.changeset(post, params)
case MyApp.Repo.update(changeset) do
  {:ok, updated_post} -> # Success
  {:error, changeset} -> # Failed, possibly due to concurrent update
end

Locking prevents race conditions in concurrent operations. Pessimistic locking uses database locks, while optimistic locking uses version fields.

Lateral Joins for Correlated Subqueries

defp newest_records(parent_ids, assoc, n) do
  %{related_key: related_key, queryable: queryable} = assoc

  squery = from q in queryable,
             where: field(q, ^related_key) == parent_as(:parent_ids).id,
             order_by: {:desc, :created_at},
             limit: ^n

  query = from f in fragment("SELECT id from UNNEST(?::int[]) AS id", ^parent_ids),
             as: :parent_ids,
             inner_lateral_join: s in subquery(squery),
             on: true,
             select: s

  MyApp.Repo.all(query)
end

Lateral joins allow subqueries that reference columns from the outer query, enabling complex correlated queries like "top N per group."

Named Bindings

query = from p in MyApp.Post, as: :posts
query = from [posts: p] in query,
          join: c in assoc(p, :comments), as: :comments
query = from [posts: p, comments: c] in query,
          where: c.score > 10,
          select: {p.title, c.body}

Named bindings make complex queries more readable by giving explicit names to each table or subquery in the query.

When to Use This Skill

Use ecto-query-patterns when you need to:

Query database records with filtering, sorting, and pagination
Join multiple tables to fetch related data
Preload associations to avoid N+1 query problems
Aggregate data using count, sum, average, or other functions
Build dynamic queries based on user input or application logic
Perform complex analytical queries with subqueries and grouping
Optimize query performance with hints and indexes
Handle concurrent updates with locking mechanisms
Create reusable query components through composition
Implement search functionality with pattern matching

Best Practices

Always use the pin operator ^ for external values to prevent SQL injection
Compose queries into small, reusable functions
Use preload to avoid N+1 query problems with associations
Select only the fields you need to reduce data transfer
Add order_by when using limit and offset for consistent pagination
Use assoc/2 helper instead of manual foreign key joins
Leverage Ecto.Query.dynamic/2 for complex conditional queries
Keep query logic in dedicated query modules, not controllers
Use subqueries for complex aggregations and analytical queries
Profile queries in development to identify performance issues
Use database indexes for frequently queried fields
Prefer preloading with joins when filtering associated data
Use named bindings for complex multi-join queries
Test query functions independently from your business logic
Document complex queries with comments explaining the logic

Common Pitfalls

Forgetting the pin operator ^, causing compilation errors
Not preloading associations, leading to N+1 query problems
Selecting entire structs when only a few fields are needed
Using Repo.all/1 in loops instead of batch operations
Building queries with string concatenation (SQL injection risk)
Not adding order_by when using pagination
Joining tables without filtering, causing cartesian products
Using fragments excessively, reducing query portability
Not handling nil values in dynamic query building
Performing aggregations in application code instead of database
Forgetting to wrap updates in transactions when necessary
Using Repo.preload/2 in loops instead of batch preloading
Not utilizing query composition for reusable logic
Mixing business logic with query construction
Over-optimizing queries prematurely without profiling
Using distinct without understanding its performance impact
Not leveraging database-specific features when appropriate
Creating overly complex queries that are hard to maintain
Ignoring database query logs during development
Not testing edge cases like empty results or nil values

📖 ecto-query-patterns