Skip to main content
Version: Next

Globally Unique Object Identification ("id" / "nodeId")

The GraphQL Global Object Identification Specification is one of the best practices in GraphQL, it gives clients a way to uniquely identify each object in the schema and to fetch these objects by their IDs.

By default, the postgraphile/presets/amber preset implements this specification. The amber preset assigns the unique identifier to every table with a primary key. The preset exposes the unique identifier as an attribute named id.

It is common in database design to use the column name id for primary keys. For this reason, if there is an attribute named id that is already on the GraphQL type, the amber preset renames that attribute to rowId. If you want to use the amber preset but you do not like this behavior, you have several options:

  1. You can use the postgraphile/presets/relay preset which makes several changes including removing rowId entirely.
  2. You can use a V4 preset that mimics the behavior and settings of PostGraphile V4 which used nodeId to represent the unique identifier.
  3. You can create your plugin similar to the following:
const IdToNodeIdPlugin: GraphileConfig.Plugin = {
name: "IdToNodeIdPlugin",
version: "1.0.0",
inflection: {
replace: {
nodeIdFieldName() {
return "nodeId";
},
attribute(previous, options, details) {
if (!previous) {
throw new Error("There was no 'attribute' inflector to replace?!");
}
const name = previous(details);
if (name === "rowId") {
return "id";
}
return name;
},
},
},
};

One common use case for the unique id field is as the cache key for your client library, e.g. with Apollo Client's dataIdFromObject:

import ApolloClient from "apollo-client";
import { HttpLink } from "apollo-link-http";
import { InMemoryCache } from "apollo-cache-inmemory";

const cache = new InMemoryCache({
dataIdFromObject: (object) => object.id || null,
// Or if you renamed 'id' to 'nodeId' then:
// dataIdFromObject: (object) => object.nodeId || null,
});

export const client = new ApolloClient({
link: new HttpLink(),
cache,
});

Disabling the Global Object Identifier

The global object identifier is added by the amber preset. If you use the amber preset but you want to disable the global object identifier throughout your API, you can do so by disabling NodePlugin:

graphile.config.mjs
export default {
// ...
disablePlugins: ["NodePlugin"],
};

Ensure that you have a good way of generating cache identifiers for your GraphQL client though!

(Note: the GraphQL Global Object Identification Specification was previously known as the Relay Global Object Identification Specification, but it is not specific to Relay and is a general best practice for GraphQL APIs.)

More On the Relay Preset

If having both id: ID! and rowId: Int! in your schema bothers you (as it should!), you should consider using the postgraphile/presets/relay preset. This preset will hide raw primary keys from most of the schema, and will use global object identifiers instead - not just in the query schema but also in mutations and filtering (and, with a little guidance, in function inputs).

graphile.config.mjs
import { PostGraphileAmberPreset } from "postgraphile/presets/amber";
import { PostGraphileRelayPreset } from "postgraphile/presets/relay";

export default {
extends: [
PostGraphileAmberPreset,
PostGraphileRelayPreset,
//...
],
// ...
};

Globally Unique ID Structure

In GraphQL a globally unique ID should be treated as an "opaque" value: you should not extract values from inside it in your application. Though the globally unique ID is stable for the same object, when new objects are created there is no guarantee that their new ID will conform to the same encoding.

That said, it is generally easy to extract details from PostGraphile's globally unique IDs. Take for example the unique ID WyJQb3N0IiwxXQ==. By base64 decoding this value, we can see the data in it is ["Post",1]. This states that it is for the Post GraphQL type, and the associated primary key value is 1. (If you are using the V4 preset then nodeIds will use the table name (or a derivative thereof) rather than the GraphQL type name.)

Thus, using globally unique IDs does not make your primary keys unobtainable, and doing so is not a goal of globally unique IDs. Should you need your primary keys to be meaningless, you should use an approach like UUIDv4 or a Feistel cipher.