postgraphile-plugin-connection-filter

Adds a powerful suite of filtering capabilities to a PostGraphile schema.

Warning: Use of this plugin with the default options may make it astoundingly trivial for a malicious actor (or a well-intentioned application that generates complex GraphQL queries) to overwhelm your database with expensive queries. See the Performance and Security section below for details.

Usage

Requires PostGraphile v4.5.0 or higher.

Install with:

yarn add postgraphile postgraphile-plugin-connection-filter

CLI usage via --append-plugins:

postgraphile --append-plugins postgraphile-plugin-connection-filter -c postgres://localhost/my_db ...

Library usage via appendPlugins:

import ConnectionFilterPlugin from "postgraphile-plugin-connection-filter";
// or: const ConnectionFilterPlugin = require("postgraphile-plugin-connection-filter");

const middleware = postgraphile(DATABASE_URL, SCHEMAS, {
  appendPlugins: [ConnectionFilterPlugin],
});

Performance and Security

By default, this plugin:

Exposes a large number of filter operators, including some that can perform expensive pattern matching.
Allows filtering on computed columns, which can result in expensive operations.
Allows filtering on functions that return setof, which can result in expensive operations.
Allows filtering on List fields (Postgres arrays), which can result in expensive operations.

To protect your server, you can:

Use the connectionFilterAllowedFieldTypes and connectionFilterAllowedOperators options to limit the filterable fields and operators exposed through GraphQL.
Set connectionFilterComputedColumns: false to prevent filtering on computed columns.
Set connectionFilterSetofFunctions: false to prevent filtering on functions that return setof.
Set connectionFilterArrays: false to prevent filtering on List fields (Postgres arrays).

Also see the Production Considerations page of the official PostGraphile docs, which discusses query whitelisting.

Features

This plugin supports filtering on almost all PostgreSQL types, including complex types such as domains, ranges, arrays, and composite types. For details on the specific operators supported for each type, see docs/operators.md.

Handling `null` and empty objects

By default, this plugin will throw an error when null literals or empty objects ({}) are included in filter input objects. This prevents queries with ambiguous semantics such as filter: { field: null } and filter: { field: { equalTo: null } } from returning unexpected results. For background on this decision, see https://github.com/graphile-contrib/postgraphile-plugin-connection-filter/issues/58.

To allow null and {} in inputs, use the connectionFilterAllowNullInput and connectionFilterAllowEmptyObjectInput options documented under Plugin Options. Please note that even with connectionFilterAllowNullInput enabled, null is never interpreted as a SQL NULL; fields with null values are simply ignored when resolving the query.

Plugin Options

When using PostGraphile as a library, the following plugin options can be passed via graphileBuildOptions:

connectionFilterAllowedOperators

Restrict filtering to specific operators:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterAllowedOperators: [
      "isNull",
      "equalTo",
      "notEqualTo",
      "distinctFrom",
      "notDistinctFrom",
      "lessThan",
      "lessThanOrEqualTo",
      "greaterThan",
      "greaterThanOrEqualTo",
      "in",
      "notIn",
    ],
  },
});

connectionFilterAllowedFieldTypes

Restrict filtering to specific field types:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterAllowedFieldTypes: ["String", "Int"],
  },
});

The available field types will depend on your database schema.

connectionFilterArrays

Enable/disable filtering on PostgreSQL arrays:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterArrays: false, // default: true
  },
});

connectionFilterComputedColumns

Enable/disable filtering by computed columns:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterComputedColumns: false, // default: true
  },
});

Consider setting this to false and using @filterable smart comments to selectively enable filtering:

create function app_public.foo_computed(foo app_public.foo)
  returns ... as $$ ... $$ language sql stable;

comment on function app_public.foo_computed(foo app_public.foo) is E'@filterable';

connectionFilterOperatorNames

Use alternative names (e.g. eq, ne) for operators:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterOperatorNames: {
      equalTo: "eq",
      notEqualTo: "ne",
    },
  },
});

connectionFilterRelations

Enable/disable filtering on related fields:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterRelations: true, // default: false
  },
});

connectionFilterSetofFunctions

Enable/disable filtering on functions that return setof:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterSetofFunctions: false, // default: true
  },
});

Consider setting this to false and using @filterable smart comments to selectively enable filtering:

create function app_public.some_foos()
  returns setof ... as $$ ... $$ language sql stable;

comment on function app_public.some_foos() is E'@filterable';

connectionFilterLogicalOperators

Enable/disable filtering with logical operators (and/or/not):

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterLogicalOperators: false, // default: true
  },
});

connectionFilterAllowNullInput

Allow/forbid null literals in input:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterAllowNullInput: true, // default: false
  },
});

When false, passing null as a field value will throw an error. When true, passing null as a field value is equivalent to omitting the field.

connectionFilterAllowEmptyObjectInput

Allow/forbid empty objects ({}) in input:

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterAllowEmptyObjectInput: true, // default: false
  },
});

When false, passing {} as a field value will throw an error. When true, passing {} as a field value is equivalent to omitting the field.

connectionFilterUseListInflectors

When building the "many" relationship filters, if this option is set true then we will use the "list" field names rather than the "connection" field names when naming the fields in the filter input. This would be desired if you have simpleCollection set to "only" or "both" and you've simplified your inflection to omit the -list suffix, e.g. using @graphile-contrib/pg-simplify-inflector's pgOmitListSuffix option. Use this if you see Connection added to your filter field names.

postgraphile(pgConfig, schema, {
  graphileBuildOptions: {
    connectionFilterUseListInflectors: true, // default: false
  },
});

Examples

query {
  allPosts(filter: {
    createdAt: { greaterThan: "2021-01-01" }
  }) {
    ...
  }
}

For an extensive set of examples, see docs/examples.md.

Development

To establish a test environment, create an empty PostgreSQL database with C collation (required for consistent ordering of strings) and set a TEST_DATABASE_URL environment variable with your database connection string.

createdb graphile_test_c --template template0 --lc-collate C
export TEST_DATABASE_URL=postgres://localhost:5432/graphile_test_c
yarn
yarn test