Binary * operation.

Binary + operation.

Binary - operation.

Binary / operation.

Binary != operation.

Binary < operation.

Binary <= operation.

Binary == operation.

Binary > operation.

Binary >= operation.

Subtracts the given interval from the current time in UTC.

The current time in UTC is retrieved from Elixir and not from the database.

See Intervals for supported interval values.

Examples

from p in Post, where: p.published_at > ago(3, "month")

Evaluates whether all values returned from the provided subquery match in a comparison operation.

from p in Post, where: p.visits >= all(
  from(p in Post, select: avg(p.visits), group_by: [p.category_id])
)

For a post to match in the above example it must be visited at least as much as the average post in all categories.

from p in Post, where: p.visits == all(
  from(p in Post, select: max(p.visits))
)

The above example matches all the posts which are tied for being the most visited.

Both any and all must be given a subquery as an argument, and they must be used on the right hand side of a comparison. Both can be used with every comparison operator: ==, !=, >, >=, <, <=.

Binary and operation.

Tests whether one or more values returned from the provided subquery match in a comparison operation.

from p in Product, where: p.id == any(
  from(li in LineItem, select: [li.product_id], where: li.created_at > ^since and li.qty >= 10)
)

A product matches in the above example if a line item was created since the provided date where the customer purchased at least 10 units.

Both any and all must be given a subquery as an argument, and they must be used on the right hand side of a comparison. Both can be used with every comparison operator: ==, !=, >, >=, <, <=.

Refer to a named atom binding.

See the "Named bindings" section in Ecto.Query for more information.

Calculates the average for the given entry.

from p in Payment, select: avg(p.value)

Takes the first value which is not null, or null if they both are.

In SQL, COALESCE takes any number of arguments, but in ecto it only takes two, so it must be chained to achieve the same effect.

from p in Payment, select: p.value |> coalesce(p.backup_value) |> coalesce(0)

Counts the entries in the table.

from p in Post, select: count()

Counts the given entry.

from p in Post, select: count(p.id)

Counts the distinct values in given entry.

from p in Post, select: count(p.id, :distinct)

Adds a given interval to a date.

See datetime_add/3 for more information.

See Intervals for supported interval values.

Adds a given interval to a datetime.

The first argument is a datetime, the second one is the count for the interval, which may be either positive or negative and the interval value:

# Get all items published since the last month
from p in Post, where: p.published_at >
                       datetime_add(^NaiveDateTime.utc_now(), -1, "month")

In the example above, we used datetime_add/3 to subtract one month from the current datetime and compared it with the p.published_at. If you want to perform operations on date, date_add/3 could be used.

See Intervals for supported interval values.

Evaluates to true if the provided subquery returns 1 or more rows.

from p in Post,
  as: :post,
  where:
    exists(
      from(
        c in Comment,
        where: parent_as(:post).id == c.post_id and c.replies_count > 5,
        select: 1
      )
    )

This is best used in conjunction with parent_as to correlate the subquery with the parent query to test some condition on related rows in a different table. In the above example the query returns posts which have at least one comment that has more than 5 replies.

Allows a field to be dynamically accessed.

def at_least_four(doors_or_tires) do
  from c in Car,
    where: field(c, ^doors_or_tires) >= 4
end

In the example above, both at_least_four(:doors) and at_least_four(:tires) would be valid calls as the field is dynamically generated.

Applies the given expression as a FILTER clause against an aggregate. This is currently only supported by Postgres.

from p in Payment, select: filter(avg(p.value), p.value > 0 and p.value < 100)

from p in Payment, select: avg(p.value) |> filter(p.value < 0)

Send fragments directly to the database.

It is not possible to represent all possible database queries using Ecto's query syntax. When such is required, it is possible to use fragments to send any expression to the database:

def unpublished_by_title(title) do
  from p in Post,
    where: is_nil(p.published_at) and
           fragment("lower(?)", p.title) == ^title
end

Every occurrence of the ? character will be interpreted as a place for parameters, which must be given as additional arguments to fragment. If the literal character ? is required as part of the fragment, it can be escaped with \\? (one escape for strings, another for fragment).

In the example above, we are using the lower procedure in the database to downcase the title column.

It is very important to keep in mind that Ecto is unable to do any type casting when fragments are used. Therefore it may be necessary to explicitly cast parameters via type/2:

fragment("lower(?)", p.title) == type(^title, :string)

Literals

Sometimes you need to interpolate a literal value into a fragment, instead of a parameter. For example, you may need to pass a table name or a collation, such as:

collation = "es_ES"
fragment("? COLLATE ?", ^name, ^collation)

The example above won't work because collation will be passed as a parameter, while it has to be a literal part of the query.

You can address this by telling Ecto that variable is a literal:

fragment("? COLLATE ?", ^name, literal(^collation))

Ecto will then escape it and make it part of the query.

Literals and query caching

Because literals are made part of the query, each interpolated literal will generate a separate query, with its own cache.

Splicing

Sometimes you may need to interpolate a variable number of arguments into the same fragment. For example, when overriding Ecto's default where behaviour for Postgres:

from p in Post, where: fragment("? in (?, ?)", p.id, val1, val2)

The example above will only work if you know the number of arguments upfront. If it can vary, the above will not work.

You can address this by telling Ecto to splice a list argument into the fragment:

from p in Post, where: fragment("? in (?)", p.id, splice(^val_list))

This will let Ecto know it should expand the values of the list into separate fragment arguments. For example:

from p in Post, where: fragment("? in (?)", p.id, splice(^[1, 2, 3]))

would be expanded into

from p in Post, where: fragment("? in (?,?,?)", p.id, ^1, ^2, ^3)

Defining custom functions using macros and fragment

You can add a custom Ecto query function using macros. For example to expose SQL's coalesce function you can define this macro:

defmodule CustomFunctions do
  defmacro coalesce(left, right) do
    quote do
      fragment("coalesce(?, ?)", unquote(left), unquote(right))
    end
  end
end

To have coalesce/2 available, just import the module that defines it.

import CustomFunctions

The only downside is that it will show up as a fragment when inspecting the Elixir query. Other than that, it should be equivalent to a built-in Ecto query function.

Keyword fragments

In order to support databases that do not have string-based queries, like MongoDB, fragments also allow keywords to be given:

from p in Post,
    where: fragment(title: ["$eq": ^some_value])

Adds the given interval to the current time in UTC.

The current time in UTC is retrieved from Elixir and not from the database.

See Intervals for supported interval values.

Examples

from a in Account, where: a.expires_at < from_now(3, "month")

Searches for search in string in a case insensitive fashion.

from p in Post, where: ilike(p.body, "Chapter%")

Translates to the underlying SQL ILIKE query. This operation is only available on PostgreSQL.

Checks if the left-value is included in the right one.

from p in Post, where: p.id in [1, 2, 3]

The right side may either be a literal list, an interpolated list, any struct that implements the Enumerable protocol, or even a column in the database with array type:

from p in Post, where: "elixir" in p.tags

Additionally, the right side may also be a subquery, which should return a single column:

from c in Comment, where: c.post_id in subquery(
  from(p in Post, where: p.created_at > ^since, select: p.id)
)

Checks if the given value is nil.

from p in Post, where: is_nil(p.published_at)

To check if a given value is not nil use:

from p in Post, where: not is_nil(p.published_at)

Returns value from the json_field pointed to by path.

from(post in Post, select: json_extract_path(post.meta, ["author", "name"]))

The path can be dynamic:

path = ["author", "name"]
from(post in Post, select: json_extract_path(post.meta, ^path))

And the field can also be dynamic in combination with it:

path = ["author", "name"]
from(post in Post, select: json_extract_path(field(post, :meta), ^path))

The query can be also rewritten as:

from(post in Post, select: post.meta["author"]["name"])

Path elements can be integers to access values in JSON arrays:

from(post in Post, select: post.meta["tags"][0]["name"])

Any element of the path can be dynamic:

field = "name"
from(post in Post, select: post.meta["author"][^field])

Warning: indexes on PostgreSQL

PostgreSQL supports indexing on jsonb columns via GIN indexes. Whenever comparing the value of a jsonb field against a string or integer, Ecto will use the containment operator @> which is optimized. You can even use the more efficient jsonb_path_ops GIN index variant. For more information, consult PostgreSQL's docs on JSON indexing.

Warning: return types

The underlying data in the JSON column is returned without any additional decoding. This means "null" JSON values are not the same as SQL's "null". For example, the Repo.all operation below returns an empty list because p.meta["author"] returns JSON's null and therefore is_nil does not succeed:

Repo.insert!(%Post{meta: %{author: nil}})
Repo.all(from(post in Post, where: is_nil(p.meta["author"])))

Similarly, other types, such as datetimes, are returned as strings. This means conditions like post.meta["published_at"] > from_now(-1, "day") may return incorrect results or fail as the underlying database tries to compare incompatible types. You can, however, use type/2 to force the types on the database level.

Searches for search in string.

from p in Post, where: like(p.body, "Chapter%")

Translates to the underlying SQL LIKE query, therefore its behaviour is dependent on the database. In particular, PostgreSQL will do a case-sensitive operation, while the majority of other databases will be case-insensitive. For performing a case-insensitive like in PostgreSQL, see ilike/2.

You should be very careful when allowing user sent data to be used as part of LIKE query, since they allow to perform LIKE-injections.

Allows a literal identifier to be injected into a fragment:

collation = "es_ES"
fragment("? COLLATE ?", ^name, literal(^collation))

The example above will inject collation into the query as a literal identifier instead of a query parameter. Note that each different value of collation will emit a different query, which will be independently prepared and cached.

Used in select to specify which fields should be returned as a map.

For example, if you don't need all fields to be returned or neither need a struct, you can use map/2 to achieve both:

from p in Post,
  select: map(p, [:title, :body])

map/2 can also be used to dynamically select fields:

fields = [:title, :body]
from p in Post, select: map(p, ^fields)

If the same source is selected multiple times with a map, the fields are merged in order to avoid fetching multiple copies from the database. In other words, the expression below:

from(city in City, preload: :country,
     select: {map(city, [:country_id]), map(city, [:name])})

is expanded to:

from(city in City, preload: :country,
     select: {map(city, [:country_id, :name]), map(city, [:country_id, :name])})

For preloads, the selected fields may be specified from the parent:

from(city in City, preload: :country,
     select: map(city, [:country_id, :name, country: [:id, :population]]))

It's also possible to select a struct from one source but only a subset of fields from one of its associations:

from(city in City, preload: :country,
     select: %{city | country: map(country: [:id, :population])})

IMPORTANT: When filtering fields for associations, you MUST include the foreign keys used in the relationship, otherwise Ecto will be unable to find associated records.

Calculates the maximum for the given entry.

from p in Payment, select: max(p.value)

Merges the map on the right over the map on the left.

If the map on the left side is a struct, Ecto will check all of the field on the right previously exist on the left before merging.

from(city in City, select: merge(city, %{virtual_field: "some_value"}))

This function is primarily used by Ecto.Query.select_merge/3 to merge different select clauses.

Calculates the minimum for the given entry.

from p in Payment, select: min(p.value)

Unary not operation.

It is used to negate values in :where. It is also used to match the assert the opposite of in/2, is_nil/1, and exists/1. For example:

from p in Post, where: p.id not in [1, 2, 3]

from p in Post, where: not is_nil(p.title)

# Retrieve all the posts that doesn't have comments.
from p in Post,
  as: :post,
  where:
    not exists(
      from(
        c in Comment,
        where: parent_as(:post).id == c.post_id
      )
    )

Binary or operation.

Refer to a named atom binding in the parent query.

This is available only inside subqueries.

See the "Named bindings" section in Ecto.Query for more information.

Refer to an alias of a selected value.

This can be used to refer to aliases created using selected_as/2. If the alias hasn't been created using selected_as/2, an error will be raised.

Each database has its own rules governing which clauses can reference these aliases. If an error is raised mentioning an unknown column, most likely the alias is being referenced somewhere that is not allowed. Consult the documentation for the database to ensure the alias is being referenced correctly.

Creates an alias for the given selected value.

When working with calculated values, an alias can be used to simplify the query. Otherwise, the entire expression would need to be copied when referencing it outside of select statements.

This comes in handy when, for instance, you would like to use the calculated value in Ecto.Query.group_by/3 or Ecto.Query.order_by/3:

from p in Post,
  select: %{
    posted: selected_as(p.posted, :date),
    sum_visits: p.visits |> coalesce(0) |> sum() |> selected_as(:sum_visits)
  },
  group_by: selected_as(:date),
  order_by: selected_as(:sum_visits)

The name of the alias must be an atom and it can only be used in the outer most select expression, otherwise an error is raised. Please note that the alias name does not have to match the key when select returns a map, struct or keyword list.

Using this in conjunction with selected_as/1 is recommended to ensure only defined aliases are referenced.

Subqueries and CTEs

Subqueries and CTEs automatically alias the selected fields, for example, one can write:

# Subquery
s = from p in Post, select: %{visits: coalesce(p.visits, 0)}
from(s in subquery(s), select: s.visits)

# CTE
cte_query = from p in Post, select: %{visits: coalesce(p.visits, 0)}
Post |> with_cte("cte", as: ^cte_query) |> join(:inner, [p], c in "cte") |> select([p, c], c.visits)

However, one can also use selected_as to override the default naming:

# Subquery
s = from p in Post, select: %{visits: coalesce(p.visits, 0) |> selected_as(:num_visits)}
from(s in subquery(s), select: s.num_visits)

# CTE
cte_query = from p in Post, select: %{visits: coalesce(p.visits, 0) |> selected_as(:num_visits)}
Post |> with_cte("cte", as: ^cte_query) |> join(:inner, [p], c in "cte") |> select([p, c], c.num_visits)

The name given to selected_as/2 can also be referenced in selected_as/1, as in regular queries.

Allows a list argument to be spliced into a fragment.

from p in Post, where: fragment("? in (?)", p.id, splice(^[1, 2, 3]))

The example above will be transformed at runtime into the following:

from p in Post, where: fragment("? in (?,?,?)", p.id, ^1, ^2, ^3)

You may only splice runtime values. For example, this would not work because query bindings are compile-time constructs:

from p in Post, where: fragment("concat(?)", splice(^[p.count, " ", "count"])

Used in select to specify which struct fields should be returned.

For example, if you don't need all fields to be returned as part of a struct, you can filter it to include only certain fields by using struct/2:

from p in Post,
  select: struct(p, [:title, :body])

struct/2 can also be used to dynamically select fields:

fields = [:title, :body]
from p in Post, select: struct(p, ^fields)

As a convenience, select allows developers to take fields without an explicit call to struct/2:

from p in Post, select: [:title, :body]

Or even dynamically:

fields = [:title, :body]
from p in Post, select: ^fields

For preloads, the selected fields may be specified from the parent:

from(city in City, preload: :country,
     select: struct(city, [:country_id, :name, country: [:id, :population]]))

If the same source is selected multiple times with a struct, the fields are merged in order to avoid fetching multiple copies from the database. In other words, the expression below:

from(city in City, preload: :country,
     select: {struct(city, [:country_id]), struct(city, [:name])})

is expanded to:

from(city in City, preload: :country,
     select: {struct(city, [:country_id, :name]), struct(city, [:country_id, :name])})

IMPORTANT: When filtering fields for associations, you MUST include the foreign keys used in the relationship, otherwise Ecto will be unable to find associated records.

Calculates the sum for the given entry.

from p in Payment, select: sum(p.value)

Casts the given value to the given type at the database level.

Most of the times, Ecto is able to proper cast interpolated values due to its type checking mechanism. In some situations though, you may want to tell Ecto that a parameter has some particular type:

type(^title, :string)

It is also possible to say the type must match the same of a column:

type(^title, p.title)

Or a parameterized type, which must be previously initialized with Ecto.ParameterizedType.init/2:

@my_enum Ecto.ParameterizedType.init(Ecto.Enum, values: [:foo, :bar, :baz])
type(^title, ^@my_enum)

Ecto will ensure ^title is cast to the given type and enforce such type at the database level. If the value is returned in a select, Ecto will also enforce the proper type throughout.

When performing arithmetic operations, type/2 can be used to cast all the parameters in the operation to the same type:

from p in Post,
  select: type(p.visits + ^a_float + ^a_integer, :decimal)

Inside select, type/2 can also be used to cast fragments:

type(fragment("NOW"), :naive_datetime)

Or to type fields from schemaless queries:

from p in "posts", select: type(p.cost, :decimal)

Or to type aggregation results:

from p in Post, select: type(avg(p.cost), :integer)
from p in Post, select: type(filter(avg(p.cost), p.cost > 0), :integer)

Or to type comparison expression results:

from p in Post, select: type(coalesce(p.cost, 0), :integer)

Or to type fields from a parent query using parent_as/1:

child = from c in Comment, where: type(parent_as(:posts).id, :string) == c.text
from Post, as: :posts, inner_lateral_join: c in subquery(child), select: c.text

type vs fragment

type/2 is all about Ecto types. Therefore, you can perform type(expr, :string) but not type(expr, :text), because :text is not an actual Ecto type. If you want to perform casting exclusively at the database level, you can use fragment. For example, in PostgreSQL, you might do fragment("?::text", p.column).

Creates a values list/constant table.

A values list can be used as a source in a query, both in Ecto.Query.from/2 and Ecto.Query.join/5.

The first argument is a list of maps representing the values of the constant table. Each entry in the list must have exactly the same fields or an error is raised.

The second argument is a map of types corresponding to the fields in the first argument. Each field must be given a type or an error is raised. Any type that can be specified in a schema may be used.

Queries using a values list are not cacheable by Ecto.

Select example

values = [%{id: 1, text: "abc"}, %{id: 2, text: "xyz"}]
types = %{id: :integer, text: :string}

query =
  from v1 in values(values, types),
    join: v2 in values(values, types),
    on: v1.id == v2.id

Repo.all(query)

Delete example

values = [%{id: 1, text: "abc"}, %{id: 2, text: "xyz"}]
types = %{id: :integer, text: :string}

query =
  from p in Post,
    join: v in values(values, types),
    on: p.id == v.id,
    where: p.counter == ^0

Repo.delete_all(query)

Update example

values = [%{id: 1, text: "abc"}, %{id: 2, text: "xyz"}]
types = %{id: :integer, text: :string}

query =
  from p in Post,
    join: v in values(values, types),
    on: p.id == v.id,
    update: [set: [text: v.text]]

Repo.update_all(query, [])