A distinct query expression.

When true, only keeps distinct values from the resulting select expression.

If supported by your database, you can also pass query expressions to distinct and it will generate a query with DISTINCT ON. In such cases, distinct accepts exactly the same expressions as order_by and any distinct expression will be automatically prepended to the order_by expressions in case there is any order_by expression.

Keywords examples

# Returns the list of different categories in the Post schema
from(p in Post, distinct: true, select: p.category)

# If your database supports DISTINCT ON(),
# you can pass expressions to distinct too
from(p in Post,
   distinct: p.category,
   order_by: [p.date])

# The DISTINCT ON() also supports ordering similar to ORDER BY.
from(p in Post,
   distinct: [desc: p.category],
   order_by: [p.date])

# Using atoms
from(p in Post, distinct: :category, order_by: :date)

Expressions example

Post
|> distinct(true)
|> order_by([p], [p.category, p.author])

Builds a dynamic query expression.

Dynamic query expressions allow developers to compose query expressions bit by bit, so that they can be interpolated into parts of a query or another dynamic expression later on.

Examples

Imagine you have a set of conditions you want to build your query on:

conditions = false

conditions =
  if params["is_public"] do
    dynamic([p], p.is_public or ^conditions)
  else
    conditions
  end

conditions =
  if params["allow_reviewers"] do
    dynamic([p, a], a.reviewer == true or ^conditions)
  else
    conditions
  end

from query, where: ^conditions

In the example above, we were able to build the query expressions bit by bit, using different bindings, and later interpolate it all at once into the actual query.

A dynamic expression can always be interpolated inside another dynamic expression and into the constructs described below.

where, having and a join's `on'

The dynamic macro can be interpolated at the root of a where, having or a join's on.

For example, assuming the conditions variable defined in the previous section, the following is forbidden because it is not at the root of a where:

from q in query, where: q.some_condition and ^conditions

Fortunately that's easily solved by simply rewriting it to:

conditions = dynamic([q], q.some_condition and ^conditions)
from query, where: ^conditions

order_by

Dynamics can be interpolated inside keyword lists at the root of order_by. For example, you can write:

order_by = [
  asc: :some_field,
  desc: dynamic([p], fragment("?>>?", p.another_field, "json_key"))
]

from query, order_by: ^order_by

Dynamics are also supported in order_by/2 clauses inside windows/2.

As with where and friends, it is not possible to pass dynamics outside of a root. For example, this won't work:

from query, order_by: [asc: ^dynamic(...)]

But this will:

from query, order_by: ^[asc: dynamic(...)]

group_by

Dynamics can be interpolated inside keyword lists at the root of group_by. For example, you can write:

group_by = [
  :some_field,
  dynamic([p], fragment("?>>?", p.another_field, "json_key"))
]

from query, group_by: ^group_by

Dynamics are also supported in partition_by/2 clauses inside windows/2.

As with where and friends, it is not possible to pass dynamics outside of a root. For example, this won't work:

from query, group_by: [:some_field, ^dynamic(...)]

But this will:

from query, group_by: ^[:some_field, dynamic(...)]

Updates

A dynamic is also supported inside updates, for example:

updates = [
  set: [average: dynamic([p], p.sum / p.count)]
]

from query, update: ^updates

An except (set difference) query expression.

Takes the difference of the result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Except expression returns only unique rows as if each query returned distinct results. This may cause a performance penalty. If you need to take the difference of multiple result sets without removing duplicate rows consider using except_all/2.

Note that the operations order_by, limit and offset of the current query apply to the result of the set difference.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, except: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> except(^supplier_query)

An except (set difference) query expression.

Takes the difference of the result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Note that the operations order_by, limit and offset of the current query apply to the result of the set difference.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, except_all: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> except_all(^supplier_query)

Resets a previously set field on a query.

It can reset many fields except the query source (from). When excluding a :join, it will remove all types of joins. If you prefer to remove a single type of join, please see paragraph below.

Examples

Ecto.Query.exclude(query, :join)
Ecto.Query.exclude(query, :where)
Ecto.Query.exclude(query, :order_by)
Ecto.Query.exclude(query, :group_by)
Ecto.Query.exclude(query, :having)
Ecto.Query.exclude(query, :distinct)
Ecto.Query.exclude(query, :select)
Ecto.Query.exclude(query, :combinations)
Ecto.Query.exclude(query, :with_ctes)
Ecto.Query.exclude(query, :limit)
Ecto.Query.exclude(query, :offset)
Ecto.Query.exclude(query, :lock)
Ecto.Query.exclude(query, :preload)

You can also remove specific joins as well such as left_join and inner_join:

Ecto.Query.exclude(query, :inner_join)
Ecto.Query.exclude(query, :cross_join)
Ecto.Query.exclude(query, :left_join)
Ecto.Query.exclude(query, :right_join)
Ecto.Query.exclude(query, :full_join)
Ecto.Query.exclude(query, :inner_lateral_join)
Ecto.Query.exclude(query, :left_lateral_join)

However, keep in mind that if a join is removed and its bindings were referenced elsewhere, the bindings won't be removed, leading to a query that won't compile.

Restricts the query to return the first result ordered by primary key.

The query will be automatically ordered by the primary key unless order_by is given or order_by is set in the query. Limit is always set to 1.

Examples

Post |> first |> Repo.one
query |> first(:inserted_at) |> Repo.one

Creates a query.

It can either be a keyword query or a query expression.

If it is a keyword query the first argument must be either an in expression, or a value that implements the Ecto.Queryable protocol. If the query needs a reference to the data source in any other part of the expression, then an in must be used to create a reference variable. The second argument should be a keyword query where the keys are expression types and the values are expressions.

If it is a query expression the first argument must be a value that implements the Ecto.Queryable protocol and the second argument the expression.

Keywords example

from(c in City, select: c)

Expressions example

City |> select([c], c)

Examples

def paginate(query, page, size) do
  from query,
    limit: ^size,
    offset: ^((page-1) * size)
end

The example above does not use in because limit and offset do not require a reference to the data source. However, extending the query with a where expression would require the use of in:

def published(query) do
  from p in query, where: not(is_nil(p.published_at))
end

Notice we have created a p variable to reference the query's original data source. This assumes that the original query only had one source. When the given query has more than one source, positional or named bindings may be used to access the additional sources.

def published_multi(query) do
  from [p,o] in query,
  where: not(is_nil(p.published_at)) and not(is_nil(o.published_at))
end

Note that the variables p and o can be named whatever you like as they have no importance in the query sent to the database.

A group by query expression.

Groups together rows from the schema that have the same values in the given fields. Using group_by "groups" the query giving it different semantics in the select expression. If a query is grouped, only fields that were referenced in the group_by can be used in the select or if the field is given as an argument to an aggregate function.

group_by also accepts a list of atoms where each atom refers to a field in source. For more complicated queries you can access fields directly instead of atoms.

Keywords examples

# Returns the number of posts in each category
from(p in Post,
  group_by: p.category,
  select: {p.category, count(p.id)})

# Using atoms
from(p in Post, group_by: :category, select: {p.category, count(p.id)})

# Using direct fields access
from(p in Post,
  join: c in assoc(p, :category),
  group_by: [p.id, c.name])

Expressions example

Post |> group_by([p], p.category) |> select([p], count(p.id))

Returns true if the query has a binding with the given name, otherwise false.

For more information on named bindings see "Named bindings" in this module doc.

An AND having query expression.

Like where, having filters rows from the schema, but after the grouping is performed giving it the same semantics as select for a grouped query (see group_by/3). having groups the query even if the query has no group_by expression.

Keywords example

# Returns the number of posts in each category where the
# average number of comments is above ten
from(p in Post,
  group_by: p.category,
  having: avg(p.num_comments) > 10,
  select: {p.category, count(p.id)})

Expressions example

Post
|> group_by([p], p.category)
|> having([p], avg(p.num_comments) > 10)
|> select([p], count(p.id))

An intersect query expression.

Takes the overlap of the result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Intersect expression returns only unique rows as if each query returned distinct results. This may cause a performance penalty. If you need to take the intersection of multiple result sets without removing duplicate rows consider using intersect_all/2.

Note that the operations order_by, limit and offset of the current query apply to the result of the set difference.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, intersect: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> intersect(^supplier_query)

An intersect query expression.

Takes the overlap of the result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Note that the operations order_by, limit and offset of the current query apply to the result of the set difference.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, intersect_all: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> intersect_all(^supplier_query)

A join query expression.

Receives a source that is to be joined to the query and a condition for the join. The join condition can be any expression that evaluates to a boolean value. The qualifier must be one of :inner, :left, :right, :cross, :full, :inner_lateral or :left_lateral.

For a keyword query the :join keyword can be changed to :inner_join, :left_join, :right_join, :cross_join, :full_join, :inner_lateral_join or :left_lateral_join. :join is equivalent to :inner_join.

Currently it is possible to join on:

• an Ecto.Schema, such as p in Post
• an interpolated Ecto query with zero or more where clauses, such as c in ^(from "posts", where: [public: true])
• an association, such as c in assoc(post, :comments)
• a subquery, such as c in subquery(another_query)
• a query fragment, such as c in fragment("SOME COMPLEX QUERY"), see "Joining with fragments" below.

Options

Each join accepts the following options:

• :on - a query expression or keyword list to filter the join
• :as - a named binding for the join
• :prefix - the prefix to be used for the join when issuing a database query
• :hints - a string or a list of strings to be used as database hints

In the keyword query syntax, those options must be given immediately after the join. In the expression syntax, the options are given as the fifth argument.

Keywords examples

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

from p in Post,
  left_join: c in assoc(p, :comments),
  select: {p, c}

Keywords can also be given or interpolated as part of on:

from c in Comment,
  join: p in Post,
  on: [id: c.post_id],
  select: {p.title, c.text}

Any key in on will apply to the currently joined expression.

It is also possible to interpolate an Ecto query on the right-hand side of in. For example, the query above can also be written as:

posts = Post
from c in Comment,
  join: p in ^posts,
  on: [id: c.post_id],
  select: {p.title, c.text}

The above is specially useful to dynamically join on existing queries, for example, to dynamically choose a source, or by choosing between public posts or posts that have been recently published:

posts =
  if params["drafts"] do
    from p in Post, where: [drafts: true]
  else
    from p in Post, where: [public: true]
  end

from c in Comment,
  join: p in ^posts, on: [id: c.post_id],
  select: {p.title, c.text}

Only simple queries with where expressions can be interpolated in a join.

Expressions examples

Comment
|> join(:inner, [c], p in Post, on: c.post_id == p.id)
|> select([c, p], {p.title, c.text})

Post
|> join(:left, [p], c in assoc(p, :comments))
|> select([p, c], {p, c})

Post
|> join(:left, [p], c in Comment, on: c.post_id == p.id and c.is_visible == true)
|> select([p, c], {p, c})

Joining with fragments

When you need to join on a complex query, Ecto supports fragments in joins:

Comment
|> join(:inner, [c], p in fragment("SOME COMPLEX QUERY", c.id, ^some_param))

Although using fragments in joins is discouraged in favor of Ecto Query syntax, they are necessary when writing lateral joins as lateral joins require a subquery that refer to previous bindings:

Game
|> join(:inner_lateral, [g], gs in fragment("SELECT * FROM games_sold AS gs WHERE gs.game_id = ? ORDER BY gs.sold_on LIMIT 2", g.id))
|> select([g, gs], {g.name, gs.sold_on})

Note that the join does not automatically wrap the fragment in parentheses, since some expressions require parens and others require no parens. Therefore, in cases such as common table expressions, you will have to explicitly wrap the fragment content in parens.

Hints

from and join also support index hints, as found in databases such as MySQL and MSSQL. Clickhouse.

For example, a developer using MySQL may write:

from p in Post,
  join: c in Comment,
  hints: ["USE INDEX FOO", "USE INDEX BAR"],
  where: p.id == c.post_id,
  select: c

Keep in mind you want to use hints rarely, so don't forget to read the database disclaimers about such functionality.

Hints must be static compile-time strings when they are specified as (list of) strings. Certain Ecto adapters may also accept dynamic hints using the tuple form:

from e in Event,
  hints: [sample: sample_threshold()],
  select: e

Restricts the query to return the last result ordered by primary key.

The query ordering will be automatically reversed, with ASC columns becoming DESC columns (and vice-versa) and limit is set to 1. If there is no ordering, the query will be automatically ordered decreasingly by primary key.

Examples

Post |> last |> Repo.one
query |> last(:inserted_at) |> Repo.one

A limit query expression.

Limits the number of rows returned from the result. Can be any expression but has to evaluate to an integer value and it can't include any field.

If limit is given twice, it overrides the previous value.

Keywords example

from(u in User, where: u.id == ^current_user, limit: 1)

Expressions example

User |> where([u], u.id == ^current_user) |> limit(1)

A lock query expression.

Provides support for row-level pessimistic locking using SELECT ... FOR UPDATE or other, database-specific, locking clauses. expr can be any expression but has to evaluate to a boolean value or to a string and it can't include any fields.

If lock is used more than once, the last one used takes precedence.

Ecto also supports optimistic locking but not through queries. For more information on optimistic locking, have a look at the Ecto.Changeset.optimistic_lock/3 function.

Keywords example

from(u in User, where: u.id == ^current_user, lock: "FOR SHARE NOWAIT")

Expressions example

User |> where(u.id == ^current_user) |> lock("FOR SHARE NOWAIT")

An offset query expression.

Offsets the number of rows selected from the result. Can be any expression but it must evaluate to an integer value and it can't include any field.

If offset is given twice, it overrides the previous value.

Keywords example

# Get all posts on page 4
from(p in Post, limit: 10, offset: 30)

Expressions example

Post |> limit(10) |> offset(30)

An OR having query expression.

Like having but combines with the previous expression by using OR. or_having behaves for having the same way or_where behaves for where.

Keywords example

# Augment a previous group_by with a having condition.
from(p in query, or_having: avg(p.num_comments) > 10)

Expressions example

# Augment a previous group_by with a having condition.
Post |> or_having([p], avg(p.num_comments) > 10)

An OR where query expression.

Behaves exactly the same as where except it combines with any previous expression by using an OR. All expressions have to evaluate to a boolean value.

or_where also accepts a keyword list where each key is a field to be compared with the given value. Each key-value pair will be combined using AND, exactly as in where.

Keywords example

from(c in City, where: [country: "Sweden"], or_where: [country: "Brazil"])

If interpolating keyword lists, the keyword list entries are combined using ANDs and joined to any existing expression with an OR:

filters = [country: "USA", name: "New York"]
from(c in City, where: [country: "Sweden"], or_where: ^filters)

is equivalent to:

from c in City, where: (c.country == "Sweden") or
                       (c.country == "USA" and c.name == "New York")

The behaviour above is by design to keep the changes between where and or_where minimal. Plus, if you have a keyword list and you would like each pair to be combined using or, it can be easily done with Enum.reduce/3:

filters = [country: "USA", is_tax_exempt: true]
Enum.reduce(filters, City, fn {key, value}, query ->
  from q in query, or_where: field(q, ^key) == ^value
end)

which will be equivalent to:

from c in City, or_where: (c.country == "USA"), or_where: c.is_tax_exempt == true

Expressions example

City |> where([c], c.country == "Sweden") |> or_where([c], c.country == "Brazil")

An order by query expression.

Orders the fields based on one or more fields. It accepts a single field or a list of fields. The default direction is ascending (:asc) and can be customized in a keyword list as one of the following:

• :asc
• :asc_nulls_last
• :asc_nulls_first
• :desc
• :desc_nulls_last
• :desc_nulls_first

The *_nulls_first and *_nulls_last variants are not supported by all databases. While all databases default to ascending order, the choice of "nulls first" or "nulls last" is specific to each database implementation.

order_by may be invoked or listed in a query many times. New expressions are always appended to the previous ones.

order_by also accepts a list of atoms where each atom refers to a field in source or a keyword list where the direction is given as key and the field to order as value.

Keywords examples

from(c in City, order_by: c.name, order_by: c.population)
from(c in City, order_by: [c.name, c.population])
from(c in City, order_by: [asc: c.name, desc: c.population])

from(c in City, order_by: [:name, :population])
from(c in City, order_by: [asc: :name, desc_nulls_first: :population])

A keyword list can also be interpolated:

values = [asc: :name, desc_nulls_first: :population]
from(c in City, order_by: ^values)

A fragment can also be used:

from c in City, order_by: [
  # A deterministic shuffled order
  fragment("? % ? DESC", c.id, ^modulus),
  desc: c.id,
]

Expressions examples

City |> order_by([c], asc: c.name, desc: c.population)
City |> order_by(asc: :name) # Sorts by the cities name

Preloads the associations into the result set.

Imagine you have a schema Post with a has_many :comments association and you execute the following query:

Repo.all from p in Post, preload: [:comments]

The example above will fetch all posts from the database and then do a separate query returning all comments associated with the given posts. The comments are then processed and associated to each returned post under the comments field.

Often times, you may want posts and comments to be selected and filtered in the same query. For such cases, you can explicitly tell an existing join to be preloaded into the result set:

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

In the example above, instead of issuing a separate query to fetch comments, Ecto will fetch posts and comments in a single query and then do a separate pass associating each comment to its parent post. Therefore, instead of returning number_of_posts * number_of_comments results, like a join would, it returns only posts with the comments fields properly filled in.

Nested associations can also be preloaded in both formats:

Repo.all from p in Post,
           preload: [comments: :likes]

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

Applying a limit to the association can be achieved with inner_lateral_join:

Repo.all from p in Post, as: :post,
           join: c in assoc(p, :comments),
           inner_lateral_join: top_five in subquery(
             from Comment,
             where: [post_id: parent_as(:post).id],
             order_by: :popularity,
             limit: 5,
             select: [:id]
           ), on: top_five.id == c.id,
           preload: [comments: c]

Preload queries

Preload also allows queries to be given, allowing you to filter or customize how the preloads are fetched:

comments_query = from c in Comment, order_by: c.published_at
Repo.all from p in Post, preload: [comments: ^comments_query]

The example above will issue two queries, one for loading posts and then another for loading the comments associated with the posts. Comments will be ordered by published_at.

When specifying a preload query, you can still preload the associations of those records. For instance, you could preload an author's published posts and the comments on those posts:

posts_query = from p in Post, where: p.state == :published
Repo.all from a in Author, preload: [posts: ^{posts_query, [:comments]}]

Note: keep in mind operations like limit and offset in the preload query will affect the whole result set and not each association. For example, the query below:

comments_query = from c in Comment, order_by: c.popularity, limit: 5
Repo.all from p in Post, preload: [comments: ^comments_query]

won't bring the top of comments per post. Rather, it will only bring the 5 top comments across all posts. Instead, use a window:

ranking_query =
  from c in Comment,
  select: %{id: c.id, row_number: over(row_number(), :posts_partition)},
  windows: [posts_partition: [partition_by: :post_id, order_by: :popularity]]

comments_query =
  from c in Comment,
  join: r in subquery(ranking_query),
  on: c.id == r.id and r.row_number <= 5

Repo.all from p in Post, preload: [comments: ^comments_query]

Preload functions

Preload also allows functions to be given. In such cases, the function receives the IDs of the parent association and it must return the associated data. Ecto then will map this data and sort it by the relationship key:

comment_preloader = fn post_ids -> fetch_comments_by_post_ids(post_ids) end
Repo.all from p in Post, preload: [comments: ^comment_preloader]

This is useful when the whole dataset was already loaded or must be explicitly fetched from elsewhere. The IDs received by the preloading function and the result returned depends on the association type:

• For has_many and belongs_to - the function receives the IDs of the parent association and it must return a list of maps or structs with the associated entries. The associated map/struct must contain the "foreign_key" field. For example, if a post has many comments, when preloading the comments with a custom function, the function will receive a list of "post_ids" as the argument and it must return maps or structs representing the comments. The maps/structs must include the :post_id field

• For has_many :through - it behaves similarly to a regular has_many but note that the IDs received are of the last association. Imagine, for example, a post has many comments and each comment has an author. Therefore, a post may have many comments_authors, written as has_many :comments_authors, through: [:comments, :author]. When preloading authors with a custom function via :comments_authors, the function will receive the IDs of the authors as the last step

• For many_to_many - the function receives the IDs of the parent association and it must return a tuple with the parent id as the first element and the association map or struct as the second. For example, if a post has many tags, when preloading the tags with a custom function, the function will receive a list of "post_ids" as the argument and it must return a tuple in the format of {post_id, tag}

Keywords example

# Returns all posts, their associated comments, and the associated
# likes for those comments.
from(p in Post,
  preload: [comments: :likes],
  select: p
)

Expressions examples

Post |> preload(:comments) |> select([p], p)

Post
|> join(:left, [p], c in assoc(p, :comments))
|> preload([p, c], [:user, comments: c])
|> select([p], p)

Puts the given prefix in a query.

Enables or disables recursive mode for CTEs.

According to the SQL standard it affects all CTEs in the query, not individual ones.

See with_cte/3 on example of how to build a query with a recursive CTE.

Reverses the ordering of the query.

ASC columns become DESC columns (and vice-versa). If the query has no order_bys, it orders by the inverse of the primary key.

Examples

query |> reverse_order() |> Repo.one()
Post |> order(asc: :id) |> reverse_order() == Post |> order(desc: :id)

A select query expression.

Selects which fields will be selected from the schema and any transformations that should be performed on the fields. Any expression that is accepted in a query can be a select field.

Select also allows each expression to be wrapped in lists, tuples or maps as shown in the examples below. A full schema can also be selected.

There can only be one select expression in a query, if the select expression is omitted, the query will by default select the full schema. If select is given more than once, an error is raised. Use exclude/2 if you would like to remove a previous select for overriding or see select_merge/3 for a limited version of select that is composable and can be called multiple times.

select also accepts a list of atoms where each atom refers to a field in the source to be selected.

Keywords examples

from(c in City, select: c) # returns the schema as a struct
from(c in City, select: {c.name, c.population})
from(c in City, select: [c.name, c.county])
from(c in City, select: %{n: c.name, answer: 42})
from(c in City, select: %{c | alternative_name: c.name})
from(c in City, select: %Data{name: c.name})

It is also possible to select a struct and limit the returned fields at the same time:

from(City, select: [:name])

The syntax above is equivalent to:

from(city in City, select: struct(city, [:name]))

You can also write:

from(city in City, select: map(city, [:name]))

If you want a map with only the selected fields to be returned.

For more information, read the docs for Ecto.Query.API.struct/2 and Ecto.Query.API.map/2.

Expressions examples

City |> select([c], c)
City |> select([c], {c.name, c.country})
City |> select([c], %{"name" => c.name})
City |> select([:name])
City |> select([c], struct(c, [:name]))
City |> select([c], map(c, [:name]))

Mergeable select query expression.

This macro is similar to select/3 except it may be specified multiple times as long as every entry is a map. This is useful for merging and composing selects. For example:

query = from p in Post, select: %{}

query =
  if include_title? do
    from p in query, select_merge: %{title: p.title}
  else
    query
  end

query =
  if include_visits? do
    from p in query, select_merge: %{visits: p.visits}
  else
    query
  end

In the example above, the query is built little by little by merging into a final map. If both conditions above are true, the final query would be equivalent to:

from p in Post, select: %{title: p.title, visits: p.visits}

If :select_merge is called and there is no value selected previously, it will default to the source, p in the example above.

The argument given to :select_merge must always be a map. The value being merged on must be a struct or a map. If it is a struct, the fields merged later on must be part of the struct, otherwise an error is raised.

If the argument to :select_merge is a constructed struct (Ecto.Query.API.struct/2) or map (Ecto.Query.API.map/2) where the source to struct or map may be a nil value (as in an outer join), the source will be returned unmodified.

query =
  Post
  |> join(:left, [p], t in Post.Translation,
    on: t.post_id == p.id and t.locale == ^"en"
  )
  |> select_merge([_p, t], map(t, ^~w(title summary)a))

If there is no English translation for the post, the untranslated post title will be returned and summary will be nil. If there is, both title and summary will be the value from Post.Translation.

select_merge cannot be used to set fields in associations, as associations are always loaded later, overriding any previous value.

Converts a query into a subquery.

If a subquery is given, returns the subquery itself. If any other value is given, it is converted to a query via Ecto.Queryable and wrapped in the Ecto.SubQuery struct.

subquery is supported in from, join, and where, in the form p.x in subquery(q).

Examples

# Get the average salary of the top 10 highest salaries
query = from Employee, order_by: [desc: :salary], limit: 10
from e in subquery(query), select: avg(e.salary)

A prefix can be specified for a subquery, similar to standard repo operations:

query = from Employee, order_by: [desc: :salary], limit: 10
from e in subquery(query, prefix: "my_prefix"), select: avg(e.salary)

Subquery can also be used in a join expression.

UPDATE posts
  SET sync_started_at = $1
  WHERE id IN (
    SELECT id FROM posts
      WHERE synced = false AND (sync_started_at IS NULL OR sync_started_at < $1)
      LIMIT $2
  )

We can write it as a join expression:

set = from(p in Post,
  where: p.synced == false and
           (is_nil(p.sync_started_at) or p.sync_started_at < ^min_sync_started_at),
  limit: ^batch_size
)

Repo.update_all(
  from(p in Post, join: s in subquery(set), on: s.id == p.id),
  set: [sync_started_at: NaiveDateTime.utc_now()]
)

Or as a where condition:

subset = from(p in subset, select: p.id)
Repo.update_all(
  from(p in Post, where: p.id in subquery(subset)),
  set: [sync_started_at: NaiveDateTime.utc_now()]
)

If you need to refer to a parent binding which is not known when writing the subquery, you can use parent_as as shown in the examples under "Named bindings" in this module doc.

A union query expression.

Combines result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Union expression returns only unique rows as if each query returned distinct results. This may cause a performance penalty. If you need to combine multiple result sets without removing duplicate rows consider using union_all/2.

Note that the operations order_by, limit and offset of the current query apply to the result of the union.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, union: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> union(^supplier_query)

A union all query expression.

Combines result sets of multiple queries. The select of each query must be exactly the same, with the same types in the same order.

Note that the operations order_by, limit and offset of the current query apply to the result of the union.

Keywords example

supplier_query = from s in Supplier, select: s.city
from c in Customer, select: c.city, union_all: ^supplier_query

Expressions example

supplier_query = Supplier |> select([s], s.city)
Customer |> select([c], c.city) |> union_all(^supplier_query)

An update query expression.

Updates are used to update the filtered entries. In order for updates to be applied, Ecto.Repo.update_all/3 must be invoked.

Keywords example

from(u in User, update: [set: [name: "new name"]])

Expressions examples

User |> update([u], set: [name: "new name"])
User |> update(set: [name: "new name"])

Interpolation

new_name = "new name"
from(u in User, update: [set: [name: ^new_name]])

new_name = "new name"
from(u in User, update: [set: [name: fragment("upper(?)", ^new_name)]])

Operators

The update expression in Ecto supports the following operators:

• set - sets the given field in the table to the given value

  from(u in User, update: [set: [name: "new name"]])

• inc - increments (or decrements if the value is negative) the given field in the table by the given value

  from(u in User, update: [inc: [accesses: 1]])

• push - pushes (appends) the given value to the end of the array field

  from(u in User, update: [push: [tags: "cool"]])

• pull - pulls (removes) the given value from the array field

  from(u in User, update: [pull: [tags: "not cool"]])

An AND where query expression.

where expressions are used to filter the result set. If there is more than one where expression, they are combined with an and operator. All where expressions have to evaluate to a boolean value.

where also accepts a keyword list where the field given as key is going to be compared with the given value. The fields will always refer to the source given in from.

Keywords example

from(c in City, where: c.country == "Sweden")
from(c in City, where: [country: "Sweden"])

It is also possible to interpolate the whole keyword list, allowing you to dynamically filter the source:

filters = [country: "Sweden"]
from(c in City, where: ^filters)

Expressions examples

City |> where([c], c.country == "Sweden")
City |> where(country: "Sweden")

Defines windows which can be used with Ecto.Query.WindowAPI.

Receives a keyword list where keys are names of the windows and values are a keyword list with window expressions.

Examples

# Compare each employee's salary with the average salary in his or her department
from e in Employee,
  select: {e.depname, e.empno, e.salary, over(avg(e.salary), :department)},
  windows: [department: [partition_by: e.depname]]

In the example above, we get the average salary per department. :department is the window name, partitioned by e.depname and avg/1 is the window function. For more information on windows functions, see Ecto.Query.WindowAPI.

Window expressions

The following keys are allowed when specifying a window.

:partition_by

A list of fields to partition the window by, for example:

windows: [department: [partition_by: e.depname]]

A list of atoms can also be interpolated for dynamic partitioning:

fields = [:depname, :year]
windows: [dynamic_window: [partition_by: ^fields]]

:order_by

A list of fields to order the window by, for example:

windows: [ordered_names: [order_by: e.name]]

It works exactly as the keyword query version of order_by/3.

:frame

A fragment which defines the frame for window functions.

Examples

# Compare each employee's salary for each month with his average salary for previous 3 months
from p in Payroll,
  select: {p.empno, p.date, p.salary, over(avg(p.salary), :prev_months)},
  windows: [prev_months: [partition_by: p.empno, order_by: p.date, frame: fragment("ROWS 3 PRECEDING EXCLUDE CURRENT ROW")]]

A common table expression (CTE) also known as WITH expression.

name must be a compile-time literal string that is being used as the table name to join the CTE in the main query or in the recursive CTE.

IMPORTANT! Beware of using CTEs. In raw SQL, CTEs can be used as a mechanism to organize queries, but said mechanism has no purpose in Ecto since Ecto queries are composable by definition. In other words, if you need to break a large query into parts, use all of the functionality in Elixir and in this module to structure your code. Furthermore, breaking a query into CTEs can negatively impact performance, as the database may not optimize efficiently across CTEs. The main use case for CTEs in Ecto is to provide recursive definitions, which we outline in the following section. Non-recursive CTEs can often be written as joins or subqueries, which provide better performance.

Options