Skip to main content
Version: Old ArrowSquid docs

Step 2: Deriving owners and tokens

This is the second part of the tutorial in which we build a squid that indexes Bored Ape Yacht Club NFTs, their transfers, and owners from the Ethereum blockchain, fetches the metadata from IPFS and regular HTTP URLs, stores all the data in a database, and serves it over a GraphQL API. In the first part we created a simple squid that scraped Transfer events emitted by the BAYC token contract. Here, we go a step further and derive separate entities for the NFTs and their owners from the transfers. The new entities will reference the corresponding Transfer entities. It will be automatically translated into primary key-foreign key references in the new database schema, and enable efficient cross-entity GraphQL queries.

Prerequisites: Node.js, Subsquid CLI, Docker, a project folder with the code from the first part (this commit).

Writing schema.graphql

Start the process by adding new entities to the schema.graphql file:

# any unique string can be used as id
type Owner @entity {
id: ID! # owner address

type Token @entity {
id: ID! # string form of tokenId
tokenId: BigInt!

Next, add entity relations. Let us begin with adding a simple relation linking tokens to their owners:

 type Token @entity {
id: ID! # string form of tokenId
tokenId: BigInt!
+ owner: Owner!

Now, Token is considered an owning entity in relation to Owner. As a result,

  • On the database side: the token table that maps to the Token entity gains a foreign key column owner_id holding primary keys of the owner table. The column is automatically indexed - no need to add @index.
  • On the Typeorm side: the Token entity gains an owner field decorated with @ManyToOne. To create a well-formed Token entity instance in processor code, we now will have to first get a hold of an appropriate Owner entity instance and populate the owner field of a new Token with a reference to it:
    let newOwner: Owner = new Owner({
    id: '0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045'
    let newToken: Token = new Token({
    id: '1',
    tokenId: 1,
    owner: newOwner // the whole entity instance
  • On the GraphQL side: queries to token can now select owner and any of its subfields (id is the only one available now).

Introduce more entity relations by replacing the from, to and tokenId fields of the Transfer entity with fields from the new entity types:

 type Transfer @entity {
id: ID!
- tokenId: BigInt! @index
- from: String! @index
- to: String! @index
+ token: Token!
+ from: Owner!
+ to: Owner!
timestamp: DateTime!
blockNumber: Int!
txHash: String! @index

Lastly, include the virtual (i.e., not mapped to a column in the database schema) reverse lookup fields:

 type Owner @entity {
id: ID! # owner address
+ ownedTokens: [Token!]! @derivedFrom(field: "owner")

type Token @entity {
id: ID! # string form of tokenId
tokenId: BigInt!
+ transfers: [Transfer!]! @derivedFrom(field: "token")

This addition doesn't create any new database columns, but it makes the ownedTokens and transfers fields accessible through GraphQL and Typeorm.

You can find the final version of schema.graphql here. Once you're finished, regenerate the Typeorm entity code with the following command:

sqd codegen

We also need to regenerate the database migrations to match the new schema. However, we'll postpone this step for now: it requires recompiling the squid code and it is not possible until we fix the entity creation code.

Creating the entities

Note how the entities we define form an acyclic dependency graph:

  • Owner entity instances can be made straight from the raw events data;
  • Tokens require the raw data plus the Owners;
  • Transfer entities require all the above.

As a consequence, the creation of entity instances must proceed in a particular order. Squids usually use small graphs like this one, and in these the order can be easily found manually (e.g. Owners then Tokens then Transfers in this case). We will assume that it can be hardcoded by the programmer.

Further, at each step we will process the data for the whole batch instead of handling the items individually. This is crucial for achieving a good syncing performance.

With all that in mind, let's create a batch processor that generates and persists all of our entities:

import {Owner, Token} from './model' TypeormDatabase(), async (ctx) => {
let rawTransfers: RawTransfer[] = getRawTransfers(ctx)

let owners: Map<string, Owner> = createOwners(rawTransfers)
let tokens: Map<string, Token> = createTokens(rawTransfers, owners)
let transfers: Transfer[] = createTransfers(rawTransfers, owners, tokens)



interface RawTransfer {
id: string
tokenId: bigint
from: string
to: string
timestamp: Date
blockNumber: number
txHash: string

is an interface very similar to that of the Transfer entity as it was at the beginning of this part of the tutorial. This allows us to reuse most of the code of the old batch handler in getRawTransfers():

import {Context} from './processor'

function getRawTransfers(ctx: Context): RawTransfer[] {
let transfers: RawTransfer[] = []

for (let block of ctx.blocks) {
for (let log of block.logs) {
if (log.address === CONTRACT_ADDRESS && log.topics[0] === {
let {from, to, tokenId} =
timestamp: new Date(block.header.timestamp),
blockNumber: block.header.height,
txHash: log.transactionHash,

return transfers

The next step involves creating Owner entity instances. We will need these to create both Tokens and Transfers. In both scenarios, we'll have the IDs of the owners (i.e., their addresses) prepared. To simplify future lookups, we choose to return the Owner instances as a Map<string, Owner>:

function createOwners(rawTransfers: RawTransfer[]): Map<string, Owner> {
let owners: Map<string, Owner> = new Map()
for (let t of rawTransfers) {
owners.set(t.from, new Owner({id: t.from}))
owners.set(, new Owner({id:}))
return owners

Similarly, Token instances will also need to be looked up later, so we return them as a Map<string, Token>. To identify the most recent owner of each token, we traverse all the transfers in the order they appear on the blockchain and assign the owner of any involved tokens to their recipient:

function createTokens(
rawTransfers: RawTransfer[],
owners: Map<string, Owner>
): Map<string, Token> {

let tokens: Map<string, Token> = new Map()
for (let t of rawTransfers) {
let tokenIdString = `${t.tokenId}`
tokens.set(tokenIdString, new Token({
id: tokenIdString,
tokenId: t.tokenId,
owner: owners.get(
return tokens

Some Token and Owner instances might have been created in previous batches, so we use to store these instances while updating any existing ones.


In some circumstances we might have had to retrieve the old entity instances from the database before updating, but here we have all the required fields populated, so we simply overwrite the whole entity with

Finally, we create an array of Transfer entity instances through a simple mapping:

function createTransfers(
rawTransfers: RawTransfer[],
owners: Map<string, Owner>,
tokens: Map<string, Token>
): Transfer[] {

return => new Transfer({
token: tokens.get(`${t.tokenId}`),
from: owners.get(t.from),
to: owners.get(,
timestamp: t.timestamp,
blockNumber: t.blockNumber,
txHash: t.txHash

Since Transfers are unique, we can safely use to persist them.

At this point, the squid has accomplished everything planned for this part of the tutorial. The only remaining task is to drop and recreate the database (if it's running) and regenerate the migrations:

sqd down
sqd up
sqd migration:generate

Full code can be found at this commit.

To test it, start the processor and the GraphQL server by running sqd process and sqd serve in separate terminals. Then, visit the GraphiQL playground:

BAYC GraphiQL at step two

The new entities should be displayed in the query schema.

Thanks to the added entity relations, we can now execute more complex nested queries. For example, the one displayed in the screenshot selects a transfer, retrieves its token, looks up its owner, and finds out which tokens are currently owned by them.