Debugging subgraphs with a local Graph Node


So your subgraph is now ready, and you deployed it to Chainstack Subgraphs!
You were careful in writing your schemas and mappings, so your subgraph should work, right? Well, maybe not.

Despite all the due diligence you might employ, it is possible that your subgraph encounters an error while indexing data. If this happens, the subgraph will not index any more data, and you might even be unable to query already indexed data.

This article will teach you how to spin up a local Graph Node using Docker Compose, and how to configure the logs of that local node to pinpoint the error that is plaguing your code.

Before we begin

  • This tutorial isn’t a beginner-friendly introduction to subgraph development. If you have never created a subgraph before, we have a beginner’s guide to subgraph development on our docs.

  • In my experience, it is really difficult to manually spin up a Graph Node with Cargo, IPFS, and Postgres. You need to install a whole bunch of dependencies, and even then the source code may not compile due to OS-specific package dependency issues. If you want to try your hand at setting up a local Graph Node without Docker, you can refer to the repo’s README file.

  • The author of this post, however, recommends you save yourselves a whole lot of hair-scratching and head-banging by setting up Docker on your machine. Docker is a tool that allows you to develop your applications in an isolated environment called a container. These containers can be replicated across different machines, ensuring a consistent and easy-to-set-up dev environment.

  • Docker Compose, more specifically, can spin up a multi-container dev environment from a single YAML file that contains all the required configuration data.

  • Before moving on with the tutorial, you need to install Docker Engine, Docker Desktop, and Docker Compose. This might seem tedious at first, but it will save you a lot of time and effort in the long run. With Docker set up, you don’t need to worry if a particular application isn’t well-suited to run on your OS.

What will we be doing?

The Bored Apes Yacht Club (BAYC) subgraph that we code along in the beginner’s guide works perfectly. However, that wasn’t the initial case when I had a subgraph that failed soon after deployment. Not knowing where exactly I was making an error, I had to redeploy the subgraph many times after minor tweaks to nail down the bug.

I will take you through the debugging process I followed, hopefully teaching you the basics of debugging subgraphs along the way.

In this tutorial, we will:

  1. Create a buggy subgraph.
  2. Spin up a local Graph Node using Docker Compose.
  3. Deploy a buggy subgraph on the said node and check for data availability.
  4. Configure the environment variables in the YAML configuration file to tweak the exact data that is logged by our local node to check for errors.
  5. Finally, deploy the debugged subgraph to Chainstack Subgraphs for long-term indexing.

Creating our subgraph

This tutorial won’t dive deep into the details of this subgraph. I recommend you check out the beginner’s guide for that, linked above. I will, however, try to incorporate brief explanations.

  1. Install graph-cli on your machine.

  2. Create a new directory and initialize a project in it by running:

    graph init 
  3. Set up a subgraph with the following parameters:

    Protocol: Ethereum
    Product for which to initialize: subgraph-studio
    Subgraph Slug: BAYC
    Directory to create the subgraph in: ChainstackSubgraph
    Ethereum network: mainnet
    Contract address: 0xBC4CA0EdA7647A8aB7C2061c2E118A18a936f13D
    Contract Name: BAYC
    Index contract events as entities: true
  4. Paste the following code in schema.graphql:

    type Transfer @entity(immutable: true) {
      id: Bytes!
      from: Bytes!
      to: Bytes!
      tokenId: BigInt! 
      blockNumber: BigInt!
      transactionHash: Bytes!
    type BoredApe @entity {
      id: ID!
      creator: Bytes!
      newOwner: Bytes!
      tokenURI: String!
      blockNumber: BigInt!
    type Property @entity {
      id: ID!
      image: String!
      background: String!
      clothes: String!
      earring: String!
      eyes: String!
      fur: String!
      hat: String!
      mouth: String!

    Explaining schema.graphql:

    • The Transfer entity will index data every time a Bored Ape NFT is transferred. It will log the transaction data, as well as the recipient and receiver addresses.
    • The BoredApe entity will index all the major information about an ape, including the original creator and the current owner.
    • The Property entity will index the IPFS metadata for all the apes. Each instance of this entity can be identified using the id field, which is nothing but the tokenID of the ape.
  5. Delete everything inside subgraph.yaml, and paste the following code:

    specVersion: 0.0.5
    description: A subgraph to index data on the Bored Apes contract
      - ipfsOnEthereumContracts
      file: ./schema.graphql
      - kind: ethereum
        name: BAYC
        network: mainnet
          address: "0xBC4CA0EdA7647A8aB7C2061c2E118A18a936f13D"
          abi: BAYC
          startBlock: 12287507
          kind: ethereum/events
          apiVersion: 0.0.7
          language: wasm/assemblyscript
            - Transfer
            - BoredApe
            - Property
            - name: BAYC
              file: ./abis/BAYC.json
            - event: Transfer(indexed address,indexed address,indexed uint256)
              handler: handleTransfer
          file: ./src/bayc.ts

    Explaining subgraph.yaml:

    • In here, we declared the data source (address of the BAYC smart contract), the entities we want to index, and the events we want to look out for. For our use case, tracking the Transfer event from the smart contract will suffice.
    • For subgraphs that want to index IPFS data, ipfsOnEthereumContracts must be declared under features.
  6. Run graph codegen in your terminal. You must run this command every time you edit the schema or YAML file.

  7. Go to src/bayc.ts and delete everything inside it. Paste the following code:

    import {
      Transfer as TransferEvent,
    } from "../generated/BAYC/BAYC"
    import {
    } from "../generated/schema"
    import { ipfs, json, JSONValue, log } from '@graphprotocol/graph-ts'
    export function handleTransfer(event: TransferEvent): void {
      let transfer = new Transfer(event.transaction.hash.concatI32(event.logIndex.toI32()))
      transfer.from = event.params.from =
      transfer.tokenId = event.params.tokenId
      transfer.blockNumber = event.block.number
      transfer.transactionHash = event.transaction.hash
      let contractAddress = BAYC.bind(event.address);
    let boredApe = BoredApe.load(event.params.tokenId.toString());
      boredApe = new BoredApe(event.params.tokenId.toString());;
      const ipfshash = "QmeSjSinHpPnmXmspMjwiXyN6zS4E9zccariGR3jxcaWtq"
      let tokenURI = "/" + event.params.tokenId.toString();
      let property = Property.load(event.params.tokenId.toString());
      if (property == null) {
        property = new Property(event.params.tokenId.toString());
        let fullURI = ipfshash + tokenURI;
        let ipfsData =;
        if (ipfsData) {
          let ipfsValues = json.fromBytes(ipfsData);
          let ipfsValuesObject = ipfsValues.toObject();
          if (ipfsValuesObject) {
            let image = ipfsValuesObject.get('image');
            let attributes = ipfsValuesObject.get('attributes');
            let attributeArray: JSONValue[];
            if (image) {
              property.image = image.toString();
            if (attributes) {
              attributeArray = attributes.toArray();
              for (let i = 0; i < attributeArray.length; i++) {
                let attributeObject = attributeArray[i].toObject();
                let trait_type = attributeObject.get('trait_type');
                let value_type = attributeObject.get('value');
                let trait: string;
                let value: string;
                if (trait_type && value_type) {
                  trait = trait_type.toString();
                  value = value_type.toString();
                  if (trait && value) {
                    if (trait == "Background") {
                      property.background = value;
                    if (trait == "Clothes") {
                      property.clothes = value;
                    if (trait == "Earring") {
                      property.earring = value;
                    if (trait == "Eyes") {
                      property.eyes = value;
                    if (trait == "Fur") {
                      property.fur = value;
                    if (trait == "Hat") {
                      property.hat = value;
                    if (trait == "Mouth") {
                      property.mouth = value;

    Explaining src/bayc.ts:

    • We write the mapping logic for our entities under the handleTransfer function. This function will be automatically triggered every time the Transfer event is triggered in the BAYC smart contract. We configured this behavior inside the YAML file by creating the event-handler pair.
    • We get the data points we need for the Transfer and BoredApe entities straight from the logs emitted by the smart contract.
    • To index IPFS metadata, however, we use the []( method from the graph-ts library by passing it the IPFS hash and the token URI for a specific ape.
    • We need to include a null check at every step while parsing the IPFS metadata, since an IPFS node may be unavailable at a given time, or the data we are querying may not exist at all.

And that’s it. Run graph build to compile your subgraph. Technically, you can even deploy this code to Chainstack, and it will index data without failing explicitly.

When I tried querying this subgraph, however, I realized I could only query data for the Transfer and BoredApe entities. The Property entity didn’t return any data at all.

But how do you know that the fault definitely lies with your subgraph? What if Chainstack is running a buggy indexer? Within your Chainstack Subgraphs console, you can query all 4 levels of logs that are defined in The Graph’s logging API. You can always refer to these to try and debug your subgraphs. You can use these logs to debug your subgraph, and then there is no need to set up a local Graph Node at all.

Let us recreate this entire scenario on our machine.

Running a local Graph Node

  1. If you have Docker and Docker Compose installed on your machine, the easiest way to set up a Graph Node locally:

    git clone

    This will clone a repo onto your machine that just contains the YAML configuration file we need. Make sure that Docker Desktop is running on your machine.

  2. Open the cloned directory, and then the docker-compose.yaml file in a code editor. Even the old and rusty Notepad++ will do. Docker will download the required dependencies using official images uploaded to it. Go to the image under IPFS, and change the line to the following:

    image: ipfs/go-ipfs:v0.17.0

    This will result in Docker fetching a relatively recent version of IPFS. You can always check Docker hub for the latest available versions of an image. You can also skip specifying a particular version if you wish to work with the latest available versions.

  3. To index subgraphs that index IPFS data using the .cat method, we need to configure an extra environment variable. Inside the environment object, add an environment variable like this:

  4. Lastly, we need to pass an RPC URL for an Ethereum archive mainnet node to the YAML file. You can set up an archive node using Chainstack. If you don’t know how to set up a node with Chainstack, refer to our blog. Edit the ethereum environment variable like this:

    ethereum: 'mainnet:YOUR_CHAINSTACK_ENDPOINT'
  5. Save the YAML file. Open a new terminal inside the cloned repo, and run:

    docker-compose up

This command is all you need. Open Docker Desktop and you should see a container named graph-node-local running. This step might take a while to execute the first time.

See how powerful this is? We didn’t have to download or set up a single dependency, and Docker automatically set up everything on our local endpoints. You can stop this container, and hence the Graph Node, by simply halting your terminal.

Finally, make sure the latest_block_head in your terminal corresponds to the latest block on Etherscan. Otherwise, it means your node is lagging behind.

Deploying to a local Graph Node

If you set up a subgraph project using graph init, the graph-cli creates a few commands inside the package.json file that can be executed using npm run.

  1. Open a new terminal at the root of the subgraph project. To compile your subgraph, run:

    graph codegen && graph build
  2. To deploy your subgraph to the local Graph Node, run the following command by passing any random version number to the terminal:

    npm run create-local && npm run deploy-local

Your subgraph is now deployed on a Graph Node running locally on your machine. Open the previous terminal where you ran the docker-compose up command. It should now be indexing data by requesting data from the Ethereum archive node we passed to it.

While the subgraph won’t be able to show us data from the recently added blocks, we can still query it. The graph-cli provides you with a UI interface URL and a GraphQL query endpoint in the terminal. Either one of those can be used to query the subgraph. In this tutorial, we will be using the UI interface. The Graph’s query API is discussed in detail in the beginner’s guide. Feel free to refer to that if you feel lost here.

In the next section, we will start debugging our subgraph.

First debugging iteration

  1. Open the UI URL in your browser. For me, it is http://localhost:8000/subgraphs/name/BAYC/graphql.
    This URL may be different for you.

  2. Run the following query in the interface:

      transfers(first:3) {
      boredApes(first:3) {
      properties(first:3) {

This query will return the first three instances of each entity from our subgraph’s data store. This is what I get back:

The Property entity returns no data

Clearly, there’s something wrong. When I saw that the Property entity had indexed no data, my first guess was that I passed the wrong IPFS path to the method. To double-check that, I used The Graph’s logging API.

This is where the fun starts.
Let us do this!

  1. Go to the terminal in your subgraph’s directory, and run:

    npm run remove-local

    This command will remove the buggy subgraph from your node’s data store.

  2. Go to the terminal running the local node, and stop the terminal (ctrl+c on my system, may be different on yours).

  3. Go to src/bayc.ts in your subgraph code, and add this line to the file right before we invoke the method:'The fullURI is: {} ', [fullURI]);
  4. Save the file.

  5. Restart the Docker container using docker-compose up and re-deploy to the local node from the subgraph’s directory.

With this change, you will notice that the Graph Node logs out the fullURI every time the handler function runs.
When I go to ipfs://<FULL_URI>, I can see all the metadata for the specific NFT. So this is not it. We’re missing something else.

Back to square 1.


Alternative link

If ipfs://<FULL_URI> doesn’t work on your machine, try accessing IPFS with<FULL_URI>.

Second debugging iteration

At this point, it finally hit me. The Property entity wasn’t returning null in the data fields. That would imply that the IPFS data is not being indexed correctly.

No. The Property entity isn’t returning anything at all. As in, the entity doesn’t have a single instance in our subgraph’s data store. This means we aren’t correctly storing the new instances generated every time the handleTransfer function is triggered.

I scan the code carefully once more, armed with this knowledge. And I finally catch it.

Check out the code for the src/bayc.ts file again. Specifically, notice the save() method. Every time we generate and fill out a new instance of an entity, we need to SAVE it to the subgraph’s data store. That is what the save() method does.

While writing the mapping function, I never saved any instance of the Property entity to the data store.

To solve this error, add the following expression to the handlerTransfer right before it ends:

As before, build the subgraph again, and deploy it to the local node.

Third debugging iteration

It is always a good moment when you have finally solved all the errors in your code.
This was not one of those moments.

As soon as my subgraph started indexing, it failed with the following error:

Entity Property[0]: missing value for non-nullable field `hat`

First instance of the entity failed to index

Notice Property[0] in the error log. This means that the very first instance of our entity failed to index. The field hat apparently can’t be null.

And again, it hit me.

Check out the code in the schema.graphql file. All of our entities are suffixed by the ! symbol.

Each entity field marked by ! has to contain a value, it can’t be null. Now, this makes sense for fields like blockNumber and transactionHash, since those values will never be null.

It is, however, possible, that a particular ape might not have all the 7 traits. In my experience, this is true more often than not.
The very first Bored Ape, with tokenID 0, does not have a “hat”. Quite literally, it doesn’t have a hat. Hence, the metadata for this NFT returns null, when we try to query it for the value of the property hat. Check it out on OpenSea.

Likewise, other apes will lack one or more of the properties that some other apes have. Thus, we need to edit the Property entity to look like this:

type Property @entity {
  id: ID!
  image: String
  background: String
  clothes: String
  earring: String
  eyes: String
  fur: String
  hat: String
  mouth: String

Run the build and deploy commands again, hopefully for the last time.

Open the GraphQL URL again and query the Property entity like this:


This time around, I can successfully query my subgraph for the IPFS metadata. Hopefully, it works for you too.

At this point, I suggest you wait for a while before pushing it to production. Your subgraph may fail while indexing data from blocks further down the line.


Pro tip

If your subgraph fails way down the line, it might take a long while to re-index all the data after tweaking the code. In that case, note the block number where the subgraph fails, and set it as the startBlock in the YAML file.
If you are confident that the changed code won’t cause the subgraph to fail on any of the blocks that were successfully indexed during the previous run, it makes sense to test out the subgraph on only the remaining blocks. This can save you time and resources.

Having successfully debugged the subgraph, we are now ready to deploy it to Chainstack Subgraphs.

This really is the simplest part. Run the codegen and build commands as before. Instead of deploying the subgraph to the local node, copy the deployment command from your Chainstack Subgraph’s console, and run it in a terminal at the root of your subgraph project.


In this tutorial, you learned how to set up a local Graph Node on your machine and use it to debug your subgraph code. This is a good point to mention that The Graph node’s behavior can be tweaked in a variety of ways by configuring its environment variables. You can read about it in the official documentation.

Chainstack Subgraphs also generates logs for you. You can always start by using those.

Practically speaking, it only makes sense to go through the whole debugging process if you need really detailed logs to pinpoint the error.

Again, you can configure the generated logs by adding or editing the environment variables in the docker-compose.yaml file.

Create an account on Chainstack today, to get access to super-fast, enterprise-grade blockchain indexers. Or maybe just hang around and check out our Web3 [De]Coded section for cool Web3 tutorials and expert-level articles.


See also

About the author

Priyank Gupta

🥑 Developer Advocate @ Chainstack
🛠️ BUIDLs on Ethereum, zkEVMs, The Graph protocol, and IPFS
🦀 Part-time Rust aficionado
Priyank Gupta | GitHub Priyank Gupta | Twitter Priyank Gupta | LinkedIN