Google Cloud Platform

Datastore Queries

A Datastore query retrieves entities from the App Engine Datastore that meet a specified set of conditions. The query operates on entities of a given kind; it can specify filters on the entities' property values, keys, and ancestors, and can return zero or more entities as results. A query can also specify sort orders to sequence the results by their property values. The results include all entities that have at least one (possibly null) value for every property named in the filters and sort orders, and whose property values meet all the specified filter criteria. The query can return entire entities, projected entities, or just entity keys.

A typical query includes the following:

  • An entity kind to which the query applies
  • Zero or more filters based on the entities' property values, keys, and ancestors
  • Zero or more sort orders to sequence the results
When executed, the query retrieves all entities of the given kind that satisfy all of the given filters, sorted in the specified order. Queries execute as read-only.

Note: To conserve memory and improve performance, a query should, whenever possible, specify a limit on the number of results returned.

Every Datastore query computes its results using one or more indexes, which contain entity keys in a sequence specified by the index's properties and, optionally, the entity's ancestors. The indexes are updated incrementally to reflect any changes the application makes to its entities, so that the correct results of all queries are available with no further computation needed.

Note: The index-based query mechanism supports a wide range of queries and is suitable for most applications. However, it does not support some kinds of query common in other database technologies: in particular, joins and aggregate queries aren't supported within the Datastore query engine. See Restrictions on Queries, below, for limitations on Datastore queries.


  1. Python query interface
  2. Query structure
    1. Filters
      1. Property filters
      2. Key filters
      3. Ancestor filters
    2. Sort orders
    3. Special query types
      1. Kindless queries
      2. Ancestor queries
      3. Kindless ancestor queries
      4. Keys-only queries
      5. Projection queries
  3. Restrictions on queries
  4. Retrieving results
  5. Offsets versus cursors
  6. Query cursors
    1. Limitations of cursors
    2. Cursors and data updates
  7. Data consistency

Python query interface

The Python Datastore API provides two classes for preparing and executing queries:

  • Query uses method calls to prepare the query.
  • GqlQuery uses a SQL-like query language called GQL to prepare the query from a query string.

    class Person(db.Model):
      first_name = db.StringProperty()
      last_name = db.StringProperty()
      city = db.StringProperty()
      birth_year = db.IntegerProperty()
      height = db.IntegerProperty()
    # Query interface constructs a query using instance methods
    q = Person.all()
    q.filter("last_name =", "Smith")
    q.filter("height <=", max_height)
    # GqlQuery interface constructs a query using a GQL query string
    q = db.GqlQuery("SELECT * FROM Person " +
                    "WHERE last_name = :1 AND height <= :2 " +
                    "ORDER BY height DESC",
                    "Smith", max_height)
    # Query is not executed until results are accessed
    for p in
      print "%s %s, %d inches tall" % (p.first_name, p.last_name, p.height)

Query structure

A query can specify an entity kind, zero or more filters, and zero or more sort orders.


A query's filters set constraints on the properties, keys, and ancestors of the entities to be retrieved.

Property filters

A property filter specifies

  • A property name
  • A comparison operator
  • A property value
For example:

q = Person.all()
q.filter("height <=", max_height)

The property value must be supplied by the application; it cannot refer to or be calculated in terms of other properties. An entity satisfies the filter if it has a property of the given name whose value compares to the value specified in the filter in the manner described by the comparison operator.

The comparison operator can be any of the following:

Operator Meaning
= Equal to
< Less than
<= Less than or equal to
> Greater than
>= Greater than or equal to
!= Not equal to
IN Member of (equal to any of the values in a specified list)

The not-equal (!=) operator actually performs two queries: one in which all other filters are unchanged and the not-equal filter is replaced with a less-than (<) filter, and one where it is replaced with a greater-than (>) filter. The results are then merged, in order. A query can have no more than one not-equal filter, and a query that has one cannot have any other inequality filters.

The IN operator also performs multiple queries: one for each item in the specified list, with all other filters unchanged and the IN filter replaced with an equality (=) filter. The results are merged in order of the items in the list. If a query has more than one IN filter, it is performed as multiple queries, one for each possible combination of values in the IN lists.

A single query containing not-equal (!=) or IN operators is limited to no more than 30 subqueries.

Key filters

To filter on the value of an entity's key, use the special property __key__:

q = Person.all()
q.filter('__key__ >', last_seen_key)

When comparing for inequality, keys are ordered by the following criteria, in order:

  1. Ancestor path
  2. Entity kind
  3. Identifier (key name or numeric ID)

Elements of the ancestor path are compared similarly: by kind (string), then by key name or numeric ID. Kinds and key names are strings and are ordered by byte value; numeric IDs are integers and are ordered numerically. If entities with the same parent and kind use a mix of key name strings and numeric IDs, those with numeric IDs precede those with key names.

Queries on keys use indexes just like queries on properties and require custom indexes in the same cases, with a couple of exceptions: inequality filters or an ascending sort order on the key do not require a custom index, but a descending sort order on the key does. As with all queries, the development web server creates appropriate entries in the index configuration file when a query that needs a custom index is tested.

Ancestor filters

You can filter your Datastore queries to a specified ancestor, so that the results returned will include only entities descended from that ancestor:

q = Person.all()

Sort orders

A query sort order specifies

  • A property name
  • A sort direction (ascending or descending)

In Python, descending sort order is denoted by a hyphen (-) preceding the property name; omitting the hyphen specifies ascending order by default. For example:

# Order alphabetically by last name:
q = Person.all()

# Order by height, tallest to shortest:
q = Person.all()

If a query includes multiple sort orders, they are applied in the sequence specified. The following example sorts first by ascending last name and then by descending height:

q = Person.all()

If no sort orders are specified, the results are returned in the order they are retrieved from the Datastore.

Note: Because of the way the App Engine Datastore executes queries, if a query specifies inequality filters on a property and sort orders on other properties, the property used in the inequality filters must be ordered before the other properties.

Special query types

Some specific types of query deserve special mention:

Kindless queries

A query with no kind and no ancestor filter retrieves all of the entities of an application from the Datastore. This includes entities created and managed by other App Engine features, such as statistics entities and Blobstore metadata entities (if any). Such kindless queries cannot include filters or sort orders on property values. They can, however, filter on entity keys by specifying __key__ as the property name:

q = db.Query()
q.filter('__key__ >', last_seen_key)

In Python, every entity returned by the query must have a corresponding model class defined for the entity's kind. To define the model classes for the statistics entity kinds, you must import the stats package:

from google.appengine.ext.db import stats

If your application has a Blobstore value, you must add the following code to get the query API to recognize the __BlobInfo__ entity kind. (Importing the Blobstore API does not define this class.)

from google.appengine.ext import db

class BlobInfo(db.Expando):
  def kind(cls):
    return '__BlobInfo__'

Ancestor queries

A query with an ancestor filter limits its results to the specified entity and its descendants:

tom = Person(key_name='Tom')

wedding_photo = Photo(parent=tom)

baby_photo = Photo(parent=tom)

dance_photo = Photo(parent=tom)

camping_photo = Photo()

photo_query = Photo.all()

# This returns wedding_photo, baby_photo, and dance_photo,
# but not camping_photo, because tom is not an ancestor
for photo in
  # Do something with photo

Kindless ancestor queries

A kindless query that includes an ancestor filter will retrieve the specified ancestor and all of its descendants, regardless of kind. This type of query does not require custom indexes. Like all kindless queries, it cannot include filters or sort orders on property values, but can filter on the entity's key:

q = db.Query()
q.filter('__key__ >', last_seen_key)

To perform a kindless ancestor query using GQL (either in the App Engine Administration Console or using the GqlQuery class), omit the FROM clause:

q = db.GqlQuery('SELECT * WHERE ANCESTOR IS :1 AND __key__ > :2',

The following example illustrates how to retrieve all entities descended from a given ancestor:

tom = Person(key_name='Tom')

wedding_photo = Photo(parent=tom)

wedding_video = Video(parent=tom)

# The following query returns both weddingPhoto and weddingVideo,
# even though they are of different entity kinds
media_query = db.query_descendants(tom)
for media in
  # Do something with media

Keys-only queries

A keys-only query returns just the keys of the result entities instead of the entities themselves, at lower latency and cost than retrieving entire entities:

q = Person.all(keys_only=True)

It is often more economical to do a keys-only query first, and then fetch a subset of entities from the results, rather than executing a general query which may fetch more entities than you actually need.

Projection queries

Sometimes all you really need from the results of a query are the values of a few specific properties. In such cases, you can use a projection query to retrieve just the properties you're actually interested in, at lower latency and cost than retrieving the entire entity; see the Projection Queries page for details.

Restrictions on queries

The nature of the index query mechanism imposes certain restrictions on what a query can do:

Entities lacking a property named in the query are ignored

Entities of the same kind need not have the same properties. To be eligible as a query result, an entity must possess a value (possibly null) for every property named in the query's filters and sort orders. If not, the entity is omitted from the indexes used to execute the query and consequently will not be included in the query's results.

Filtering on unindexed properties returns no results

A query can't find property values that aren't indexed, nor can it sort on such properties. See the Datastore Indexes page for a detailed discussion of unindexed properties.

Inequality filters are limited to at most one property

To avoid having to scan the entire index, the query mechanism relies on all of a query's potential results being adjacent to one another in the index. To satisfy this constraint, a single query may not use inequality comparisons (<, <=, >, >=, !=) on more than one property across all of its filters. For example, the following query is valid, because both inequality filters apply to the same property:

SELECT * FROM Person WHERE birth_year >= :min_birth_year
                       AND birth_year <= :max_birth_year

However, this query is not valid, because it uses inequality filters on two different properties:

SELECT * FROM Person WHERE birth_year >= :max_birth_year
                       AND height <= :max_height          # ERROR

Note that a query can combine equality (=) filters for different properties, along with one or more inequality filters on a single property. Thus the following is a valid query:

SELECT * FROM Person WHERE last_name = :target_last_name
                       AND city = :target_city
                       AND birth_year >= :min_birth_year
                       AND birth_year <= :max_birth_year

Ordering of query results is undefined when no sort order is specified

When a query does not specify a sort order, the results are returned in the order they are retrieved. As the Datastore implementation evolves (or if an application's indexes change), this order may change. Therefore, if your application requires its query results in a particular order, be sure to specify that sort order explicitly in the query.

Sort orders are ignored on properties with equality filters

Queries that include an equality filter for a given property ignore any sort order specified for that property. This is a simple optimization to save needless processing for single-valued properties, since all results have the same value for the property and so no further sorting is needed. Multiple-valued properties, however, may have additional values besides the one matched by the equality filter. Because this use case is rare and applying the sort order would be expensive and require extra indexes, the Datastore query planner simply ignores the sort order even in the multiple-valued case. This may cause query results to be returned in a different order than the sort order appears to imply.

Properties used in inequality filters must be sorted first

To retrieve all results that match an inequality filter, a query scans the index for the first row matching the filter, then scans forward until it encounters a nonmatching row. For the consecutive rows to encompass the complete result set, they must be ordered by the property used in the inequality filter before any other properties. Thus if a query specifies one or more inequality filters along with one or more sort orders, the first sort order must refer to the same property named in the inequality filters. The following is a valid query:

SELECT * FROM Person WHERE birth_year >= :min_birth_year
                     ORDER BY birth_year, last_name

This query is not valid, because it doesn't sort on the property used in the inequality filter:

SELECT * FROM Person WHERE birth_year >= :min_birth_year
                     ORDER BY last_name                   # ERROR

Similarly, this query is not valid because the property used in the inequality filter is not the first one sorted:

SELECT * FROM Person WHERE birth_year >= :min_birth_year
                     ORDER BY last_name, birth_year       # ERROR

Properties with multiple values can behave in surprising ways

Because of the way they're indexed, entities with multiple values for the same property can sometimes interact with query filters and sort orders in unexpected and surprising ways.

If a query has multiple inequality filters on a given property, an entity will match the query only if at least one of its individual values for the property satisfies all of the filters. For example, if an entity of kind Widget has values 1 and 2 for property x, it will not match the query:

SELECT * FROM Widget WHERE x > 1
                       AND x < 2

Each of the entity's x values satisfies one of the filters, but neither single value satisfies both. Note that this does not apply to equality filters. For example, the same entity will satisfy the query

SELECT * FROM Widget WHERE x = 1
                       AND x = 2

even though neither of the entity's individual x values satisfies both filter conditions.

The not-equal (!=) operator works as a "value is other than" test. So, for example, the query

SELECT * FROM Widget WHERE x != 1

matches any Widget entity with an x value other than 1.

Similarly, the sort order for multiple-valued properties is unusual. Because such properties appear once in the index for each unique value, the first value seen in the index determines an entity's sort order:

  • If the query results are sorted in ascending order, the smallest value of the property is used for ordering.
  • If the results are sorted in descending order, the greatest value is used for ordering.
  • Other values do not affect the sort order, nor does the number of values.

This has the unusual consequence that an entity with property values 1 and 9 precedes one with values 4, 5, 6, and 7 in both ascending and descending order.

Queries inside transactions must include ancestor filters

Datastore transactions operate only on entities belonging to the same entity group (descended from a common ancestor). To preserve this restriction, all queries performed within a transaction must include an ancestor filter specifying an ancestor in the same entity group as the other operations in the transaction.

Retrieving results

After constructing a query, you can specify a number of retrieval options to further control the results it returns.

To retrieve just a single entity matching your query, use the method Query.get() (or GqlQuery.get()):

q = Person.all()
q.filter("last_name =", target_last_name)

result = q.get()

This returns the first result found in the index that matches the query.

To retrieve only selected properties of an entity rather than the entire entity, use a projection query. This type of query runs faster and costs less than one that returns complete entities.

Similarly, a keys-only query saves time and resources by returning just the keys to the entities it matches, rather than the full entities themselves. To create this type of query, set keys_only=True when constructing the query object:

q = Person.all(keys_only=True)

You can specify a limit for your query to control the maximum number of results returned in one batch. The following example retrieves the five tallest people from the Datastore:

q = Person.all()

for p in
  print "%s %s, %d inches tall" % (p.first_name, p.last_name, p.height)

When iterating through the results of a query using the run() method of a Query or GqlQuery object, the Datastore retrieves the results in batches. By default each batch contains 20 results, but you can change this value using the method's batch_size parameter. You can continue iterating through query results until all are returned or the request times out.

Offsets versus cursors

Although Datastore supports integer offsets, you should avoid using them. Instead, use cursors. Using an offset only avoids returning the skipped entities to your application, but these entities are still retrieved internally. The skipped entities do affect the latency of the query, and your application is billed for the read operations required to retrieve them. Using cursors instead of offsets lets you avoid all these costs.

Query cursors

Query cursors allow an application to retrieve a query's results in convenient batches without incurring the overhead of a query offset. After performing a retrieval operation, the application can obtain a cursor, which is an opaque base64-encoded string marking the index position of the last result retrieved. The application can save this string (for instance in the Datastore, in Memcache, in a Task Queue task payload, or embedded in a web page as an HTTP GET or POST parameter), and can then use the cursor as the starting point for a subsequent retrieval operation to obtain the next batch of results from the point where the previous retrieval ended. A retrieval can also specify an end cursor, to limit the extent of the result set returned.

In Python, an application obtains a cursor after retrieving query results by calling the Query object's cursor() method. To retrieve additional results from the point of the cursor, the application prepares a similar query (with the same entity kind, filters, and sort orders), and passes the cursor to the query's with_cursor() method before performing the retrieval:

from google.appengine.api import memcache
from google.appengine.ext import db

# class Person(db.Model): ...
# Start a query for all Person entities
people = Person.all()
# If the application stored a cursor during a previous request, use it
person_cursor = memcache.get('person_cursor')
if person_cursor:
# Iterate over the results
for person in people:
  # Do something
# Get updated cursor and store it for next time
person_cursor = people.cursor()
memcache.set('person_cursor', person_cursor)

Limitations of cursors

Cursors are subject to the following limitations:

  • A cursor can be used only by the same application that performed the original query, and only to continue the same query. To use the cursor in a subsequent retrieval operation, you must reconstitute the original query exactly, including the same entity kind, ancestor filter, property filters, and sort orders. It is not possible to retrieve results using a cursor without setting up the same query from which it was originally generated.
  • Because the != and IN operators are implemented with multiple queries, queries that use them do not support cursors.
  • Cursors don't always work as expected with a query that uses an inequality filter or a sort order on a property with multiple values. The de-duplication logic for such multiple-valued properties does not persist between retrievals, possibly causing the same result to be returned more than once.
  • New App Engine releases may change internal implementation details, invalidating cursors that depend on them. If an application attempts to use a cursor that is no longer valid, the Datastore raises a BadRequestError exception.

Cursors and data updates

The cursor's position is defined as the location in the result list after the last result returned. A cursor is not a relative position in the list (it's not an offset); it's a marker to which the Datastore can jump when starting an index scan for results. If the results for a query change between uses of a cursor, the query notices only changes that occur in results after the cursor. If a new result appears before the cursor's position for the query, it will not be returned when the results after the cursor are fetched. Similarly, if an entity is no longer a result for a query but had appeared before the cursor, the results that appear after the cursor do not change. If the last result returned is removed from the result set, the cursor still knows how to locate the next result.

When retrieving query results, you can use both a start cursor and an end cursor to return a continuous group of results from the Datastore. When using a start and end cursor to retrieve the results, you are not guaranteed that the size of the results will be the same as when you generated the cursors. Entities may be added or deleted from the Datastore between the time the cursors are generated and when they are used in a query.

Data consistency

Datastore queries can deliver their results at either of two consistency levels:

In an eventually consistent query, the indexes used to gather the results are also accessed with eventual consistency. Consequently, such queries may sometimes return entities that no longer match the original query criteria, while strongly consistent queries are always transactionally consistent. See the article Transaction Isolation in App Engine for more information on how entities and indexes are updated.

Queries return their results with different levels of consistency guarantee, depending on the nature of the query:

  • Ancestor queries (those within an entity group) are strongly consistent by default, but can instead be made eventually consistent by setting the Datastore read policy (see below).
  • Non-ancestor queries are always eventually consistent.

To improve performance, you can set the Datastore read policy so that all reads and queries are eventually consistent. (The API also allows you to explicitly set a strong consistency policy, but this setting will have no practical effect, since non-ancestor queries are always eventually consistent regardless of policy.)

You can also set the Datastore call deadline: the maximum time, in seconds, that the application will wait for the Datastore to return a result before aborting with an error. The default deadline is 60 seconds; it is not currently possible to set it higher, but you can adjust it downward to ensure that a particular operation fails quickly (for instance, to return a faster response to the user).

To set the Datastore read policy and call deadline in Python, you pass them as arguments to the run(), get(), fetch(), and count() methods of class Query or GqlQuery. For example:

for result in Employee.all().run(limit=5,
  # Body of iterative loop