GraphQL API
Understanding the core concepts of the GraphQL API.
In our API, each SQL table is reflected as a set of GraphQL types. At a high level, tables become types and columns/foreign keys become fields on those types.
By default, PostgreSQL table and column names are not inflected when reflecting GraphQL names. For example, an account_holder
table has GraphQL type name account_holder
. In cases where SQL entities are named using snake_case
, enable inflection to match GraphQL/Javascript conventions e.g. account_holder
-> AccountHolder
.
Individual table, column, and relationship names may also be manually overridden.
Primary Keys (Required)
Every table must have a primary key for it to be exposed in the GraphQL schema. For example, the following Blog
table will be available in the GraphQL schema as blogCollection
since it has a primary key named id
:
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10);
But the following table will not be exposed because it doesn't have a primary key:
_10create table "Blog"(_10 id int,_10 name varchar(255) not null,_10);
QueryType
The Query
type is the entrypoint for all read access into the graph.
Node
The node
interface allows for retrieving records that are uniquely identifiable by a globally unique nodeId: ID!
field. For more information about nodeId, see nodeId.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null,_10 "updatedAt" timestamp not null_10);
GraphQL Types
To query the node
interface effectively, use inline fragments to specify which fields to return for each type.
Example
Collections
Each table has top level entry in the Query
type for selecting records from that table. Collections return a connection type and can be paginated, filtered, and sorted using the available arguments.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null,_10 "updatedAt" timestamp not null_10);
GraphQL Types
Connection types are the primary interface to returning records from a collection.
Connections wrap a result set with some additional metadata.
The totalCount
field is disabled by default because it can be expensive on large tables. To enable it use a comment directive
Pagination
Keyset Pagination
Paginating forwards and backwards through collections is handled using the first
, last
, before
, and after
parameters, following the relay spec.
Metadata relating to the current page of a result set is available on the pageInfo
field of the connection type returned from a collection.
To paginate forward in the collection, use the first
and after
arguments. To retrieve the first page, the after
argument should be null or absent.
Example
To retrieve the next page, provide the cursor value from data.blogCollection.pageInfo.endCursor
to the after
argument of another query.
once the collection has been fully enumerated, data.blogConnection.pageInfo.hasNextPage
returns false.
To paginate backwards through a collection, repeat the process substituting first
-> last
, after
-> before
, hasNextPage
-> hasPreviousPage
Offset Pagination
In addition to keyset pagination, collections may also be paged using first
and offset
, which operates like SQL's limit
and offset
to skip offset
number of records in the results.
offset
based pagination becomes inefficient the offset
value increases. For this reason, prefer cursor based pagination where possible.
Filtering
To filter the result set, use the filter
argument.
Where the <Table>Filter
type enumerates filterable fields and their associated <Type>Filter
.
The following list shows the operators that may be available on <Type>Filter
types.
Operator | Description |
---|---|
eq | Equal To |
neq | Not Equal To |
gt | Greater Than |
gte | Greater Than Or Equal To |
in | Contained by Value List |
lt | Less Than |
lte | Less Than Or Equal To |
is | Null or Not Null |
startsWith | Starts with prefix |
like | Pattern Match. '%' as wildcard |
ilike | Pattern Match. '%' as wildcard. Case Insensitive |
regex | POSIX Regular Expression Match |
iregex | POSIX Regular Expression Match. Case Insensitive |
Not all operators are available on every <Type>Filter
type. For example, UUIDFilter
only supports eq
and neq
because UUID
s are not ordered.
Example: simple
** Example: and/or **
Multiple filters can be combined with and
, or
and not
operators. The and
and or
operators accept a list of <Type>Filter
.
** Example: not **
not
accepts a single <Type>Filter
.
** Example: nested composition **
The and
, or
and not
operators can be arbitrarily nested inside each other.
** Example: empty **
Empty filters are ignored, i.e. they behave as if the operator was not specified at all.
** Example: implicit and **
Multiple column filters at the same level will be implicitly combined with boolean and
. In the following example the id: {eq: 1}
and name: {eq: "A: Blog 1"}
will be and
ed.
This means that an and
filter can be often be simplified. In the following example all queries are equivalent and produce the same result.
Be aware that the above simplification only works for the and
operator. If you try it with an or
operator it will behave like an and
.
This is because according to the rules of GraphQL list input coercion, if a value passed to an input of list type is not a list, then it is coerced to a list of a single item. So in the above example or: {id: {eq: 1}, name: {eq: "A: Blog 2}}
will be coerced into or: [{id: {eq: 1}, name: {eq: "A: Blog 2}}]
which is equivalent to or: [and: [{id: {eq: 1}}, {name: {eq: "A: Blog 2}}}]
due to implicit and
ing.
Avoid naming your columns and
, or
or not
. If you do, the corresponding filter operator will not be available for use.
The and
, or
and not
operators also work with update and delete mutations.
Ordering
The default order of results is defined by the underlying table's primary key column in ascending order. That default can be overridden by passing an array of <Table>OrderBy
to the collection's orderBy
argument.
Example
Note, only one key value pair may be provided to each element of the input array. For example, [{name: AscNullsLast}, {id: AscNullFirst}]
is valid. Passing multiple key value pairs in a single element of the input array e.g. [{name: AscNullsLast, id: AscNullFirst}]
, is invalid.
MutationType
The Mutation
type is the entrypoint for mutations/edits.
Each table has top level entry in the Mutation
type for inserting insertInto<Table>Collection
, updating update<Table>Collection
and deleting deleteFrom<Table>Collection
.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null default now(),_10 "updatedAt" timestamp_10);
Insert
To add records to a collection, use the insertInto<Table>Collection
field on the Mutation
type.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null default now(),_10 "updatedAt" timestamp_10);
GraphQL Types
Where elements in the objects
array are inserted into the underlying table.
Example
Update
To update records in a collection, use the update<Table>Collection
field on the Mutation
type.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null default now(),_10 "updatedAt" timestamp_10);
GraphQL Types
Where the set
argument is a key value pair describing the values to update, filter
controls which records should be updated, and atMost
restricts the maximum number of records that may be impacted. If the number of records impacted by the mutation exceeds the atMost
parameter the operation will return an error.
Example
Delete
To remove records from a collection, use the deleteFrom<Table>Collection
field on the Mutation
type.
SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null,_10 description varchar(255),_10 "createdAt" timestamp not null default now(),_10 "updatedAt" timestamp_10);
GraphQL Types
Where filter
controls which records should be deleted and atMost
restricts the maximum number of records that may be deleted. If the number of records impacted by the mutation exceeds the atMost
parameter the operation will return an error.
Example
Concepts
nodeId
The base GraphQL type for every table with a primary key is automatically assigned a nodeId: ID!
field. That value, can be passed to the node entrypoint of the Query
type to retrieve its other fields. nodeId
may also be used as a caching key.
relay support
ID
field for types to have the name id
. pg_graphql uses nodeId
by default to avoid conflicting with user defined id
columns. You can configure relay to work with pg_graphql's nodeId
field with relay's nodeInterfaceIdField
option. More info available here.SQL Setup
_10create table "Blog"(_10 id serial primary key,_10 name varchar(255) not null_10);
GraphQL Types
Relationships
Relationships between collections in the Graph are derived from foreign keys.
One-to-Many
A foreign key on table A referencing table B defines a one-to-many relationship from table A to table B.
SQL Setup
_11create table "Blog"(_11 id serial primary key,_11 name varchar(255) not null_11);_11_11create table "BlogPost"(_11 id serial primary key,_11 "blogId" integer not null references "Blog"(id),_11 title varchar(255) not null,_11 body varchar(10000)_11);
GraphQL Types
Where blogPostCollection
exposes the full Query
interface to BlogPost
s.
Example
Many-to-One
A foreign key on table A referencing table B defines a many-to-one relationship from table B to table A.
SQL Setup
_11create table "Blog"(_11 id serial primary key,_11 name varchar(255) not null_11);_11_11create table "BlogPost"(_11 id serial primary key,_11 "blogId" integer not null references "Blog"(id),_11 title varchar(255) not null,_11 body varchar(10000)_11);
GraphQL Types
Where blog
exposes the Blog
record associated with the BlogPost
.
One-to-One
A one-to-one relationship is defined by a foreign key on table A referencing table B where the columns making up the foreign key on table A are unique.
SQL Setup
_10create table "EmailAddress"(_10 id serial primary key,_10 address text unique not null_10);_10_10create table "Employee"(_10 id serial primary key,_10 name text not null,_10 email_address_id int unique references "EmailAddress"(id)_10);
GraphQL Types
Example
Custom Scalars
Due to differences among the types supported by PostgreSQL, JSON, and GraphQL, pg_graphql
adds several new Scalar types to handle PostgreSQL builtins that require special handling.
JSON
pg_graphql
serializes json
and jsonb
data types as String
under the custom scalar name JSON
.
_10scalar JSON
Example
Given the setup
The query
_10{_10 userCollection {_10 edges {_10 node {_10 config_10 }_10 }_10 }_10}
The returns the following data. Note that config
is serialized as a string
_13{_13 "data": {_13 "userCollection": {_13 "edges": [_13 {_13 "node": {_13 "config": "{\"palette\": \"dark-mode\"}"_13 }_13 }_13 ]_13 }_13 }_13}
Use serialized JSON strings when updating or inserting JSON
fields via the GraphQL API.
JSON does not currently support filtering.
BigInt
PostgreSQL bigint
and bigserial
types are 64 bit integers. In contrast, JSON supports 32 bit integers.
Since PostgreSQL bigint
values may be outside the min/max range allowed by JSON, they are represented in the GraphQL schema as BigInt
s and values are serialized as strings.
_12scalar BigInt_12_12input BigIntFilter {_12 eq: BigInt_12 gt: BigInt_12 gte: BigInt_12 in: [BigInt!]_12 lt: BigInt_12 lte: BigInt_12 neq: BigInt_12 is: FilterIs_12}
Example
Given the setup
The query
_10{_10 personCollection {_10 edges {_10 node {_10 id_10 name_10 }_10 }_10 }_10}
The returns the following data. Note that id
is serialized as a string
_14{_14 "data": {_14 "personCollection": {_14 "edges": [_14 {_14 "node": {_14 "id": "1",_14 "name": "Foo Barington",_14 }_14 }_14 ]_14 }_14 }_14}
BigFloat
PostgreSQL's numeric
type supports arbitrary precision floating point values. JSON's float
is limited to 64-bit precision.
Since a PostgreSQL numeric
may require more precision than can be handled by JSON, numeric
types are represented in the GraphQL schema as BigFloat
and values are serialized as strings.
_12scalar BigFloat_12_12input BigFloatFilter {_12 eq: BigFloat_12 gt: BigFloat_12 gte: BigFloat_12 in: [BigFloat!]_12 lt: BigFloat_12 lte: BigFloat_12 neq: BigFloat_12 is: FilterIs_12}
Example
Given the SQL setup
_10create table "GeneralLedger"(_10 id serial primary key,_10 amount numeric(10,2)_10);_10_10insert into "GeneralLedger"(amount)_10values (22.15);
The query
_10{_10 generalLedgerCollection {_10 edges {_10 node {_10 id_10 amount_10 }_10 }_10 }_10}
The returns the following data. Note that amount
is serialized as a string
_14{_14 "data": {_14 "generalLedgerCollection": {_14 "edges": [_14 {_14 "node": {_14 "id": 1,_14 "amount": "22.15",_14 }_14 }_14 ]_14 }_14 }_14}
Opaque
PostgreSQL's type system is extensible and not all types handle all operations e.g. filtering with like
. To account for these, pg_graphql
introduces a scalar Opaque
type. The Opaque
type uses PostgreSQL's to_json
method to serialize values. That allows complex or unknown types to be included in the schema by delegating handling to the client.
_10scalar Opaque_10_10input OpaqueFilter {_10 eq: Opaque_10 is: FilterIs_10}