Skip to main content
Version: FireSquid

Step 4: Optimization

This is the fourth part of the tutorial where 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 it in a database, and serves it over a GraphQL API. In the first three parts (1, 2, 3), we created a squid that does all the above but performs many IO operations sequentially, resulting in a long sync time. In this part, we discuss strategies for mitigating that shortcoming. We also discuss an alternative metadata fetching strategy that reduces redundant fetches and handles the changes in metadata of "cold" (i.e., not involved in any transfers) tokens more effectively.

Pre-requisites: Node.js, Subsquid CLI, Docker, a project folder with the code from the third part (this commit).

Using Multicall for aggregating state queries

We begin by introducing batch processing wherever possible, and our first step is to replace individual contract state queries with batch calls to a MakerDAO multicall contract. Retrieve the multicall contract ABI by re-running squid-evm-typegen with --multicall option:

npx squid-evm-typegen --multicall src/abi 0xbc4ca0eda7647a8ab7c2061c2e118a18a936f13d#bayc

This adds a Typescript ABI interface at src/abi/multicall.ts. Let us use it in a rewrite of completeTokens():

import { Multicall } from './abi/multicall'

const MULTICALL_ADDRESS = '0x5ba1e12693dc8f9c48aad8770482f4739beed696'

// ...

async function completeTokens(
ctx: Context,
partialTokensMap: Map<string, PartialToken>
): Promise<Map<string, Token>> {

let partialTokens: PartialToken[] = [...partialTokensMap.values()]

let tokens: Map<string, Token> = new Map()
if (partialTokens.length === 0) return tokens

let lastBatchBlockHeader = ctx.blocks[ctx.blocks.length-1].header
let contract = new Multicall(ctx, lastBatchBlockHeader, MULTICALL_ADDRESS)

let tokenURIs = await contract.aggregate(
CONTRACT_ADDRESS, => [BigNumber.from(t.tokenId)]),
MULTICALL_BATCH_SIZE // paginating to avoid RPC timeouts

for (let [i, ptoken] of partialTokens.entries()) {
let uri = tokenURIs[i]
let metadata: TokenMetadata | undefined = await fetchTokenMetadata(ctx, uri)
tokens.set(, new Token({

return tokens

// ...

Here we replaced the direct calls to tokenURI() of the BAYC token contract with aggreaged batches of 100 state calls and called aggregate() of the multicall contract for each batch. Multicall.aggregate() takes care of splitting the set of state calls into chunks and merging the results together. The number 100 for the batch size was chosen to be as large as possible while not triggering any response size limits on the public RPC endpoint we use. For private RPC endpoints one would try to further increase the batch size.

You can find the full code after this optimization at this commit. In our test this optimization reduced the time required for retrieving the contract state for all 10000 tokens once from 114 minutes to 94 seconds.

Retrieving metadata from HTTPS concurrently

Next, we make our metadata retrieval requests concurrent.

Many metadata URIs point to the same HTTPS server or to IPFS that we are accessing through a single gateway. When retrieving data from just a few servers, it is a common courtesy to not send all the requests at once; often, this is enforced by rate limits. Our code takes that into account, limiting the rate of the outgoing requests.

We implement HTTPS batching at src/metadata.ts:

import { asyncSleep, splitIntoBatches } from './util'

const MAX_REQ_SEC = 10

export async function fetchTokenMetadatasConcurrently(
ctx: Context,
uris: string[]
): Promise<(TokenMetadata | undefined)[]> {

let metadatas: (TokenMetadata | undefined)[] = []
for (let batch of splitIntoBatches(uris, MAX_REQ_SEC)) {
let m = await Promise.all(, index) => {
// spread out the requests evenly within a second interval
let sleepMs = Math.ceil(1000*(index+1)/MAX_REQ_SEC)
return asyncSleep(sleepMs).then(() => fetchTokenMetadata(ctx, uri))
return metadatas

then we call the function from completeTokens() and use its output to populate metadata fields of Token entity instances. Utility functions asyncSleep() and splitIntoBathches() are implemented here. Full code is available at this commit.

On the first processor batch, when we retrieve metadata from for all 10000 tokens, this optimization reduced the required time from 45 to 21 minutes.

Retrieving immutable metadata once

Another glaring inefficiency in our current code is the fact that we often retrieve metadata from each URI more than once. This is understandable when we work with HTTPS links that point to mutable data; however, the data IPFS links point to is immutable, so there is no need to retrieve it more than once.

To avoid these repeated retrievals we find the already known tokens in the database. If a token metadata URI is an immutable IPFS link (that is, does not point to MFS) and metadata is already available, we skip the retrieval. The implementation is at src/metadata.ts:

export async function selectivelyUpdateMetadata(
ctx: Context,
tokens: Map<string, Token>
): Promise<Map<string, Token>> {

let knownTokens: Map<string, Token> = await
{id: In([...tokens.keys()])}
.then(ts => new Map( => [, t])))

let updatedTokens: Map<string, Token> = new Map()
let tokensToBeUpdated: Token[] = []
for (let [id, t] of tokens) {
let ktoken: Token | undefined = knownTokens.get(id)
if (ktoken != null &&
ktoken.image != null && ktoken.attributes != null &&
ktoken.uri === t.uri && uriPointsToImmutable(t.uri)) {`Repeated retrieval from ${t.uri} skipped`)
updatedTokens.set(id, ktoken)
else {`Re-retrieving from ${t.uri}`)

let metadatas: (TokenMetadata | undefined)[] = await fetchTokenMetadatasConcurrently(
ctx, => t.uri)

for (let [i, t] of tokensToBeUpdated.entries()) {
let m = metadatas[i]
if (m != null) {
t.image = m.image
t.attributes = m.attributes
updatedTokens.set(, t)

return updatedTokens

uriPointsToImmutable() used to classify metadata URIs rejects any non-IPFS links and IPFS links that may point to MFS:

export function uriPointsToImmutable(uri: string): boolean {
return uri.startsWith('ipfs://') && !uri.includes('ipns')

Full code is available at this commit. In our test this optimization reduced the total sync time from about 4.1 to 1.5 hours.

Alternative: Post-sync retrieval of metadata

Despite all the optimizations, our squid still takes 1.5 hours to sync instead of 11 minutes it needed before we introduced metadata retrieval. This is the cost of maintaining a fully populated database at all stages of the sync, which is rarely a requirement. An alternative is to begin retrieving metadata only after the rest of the data has been fully synced, and the squid has caught up with the blockchain. We do that by reading Token entities from the database after the initial sync, retrieving their metadata and persisting them.

This approach has another advantage: metadata URIs can change without notice, and our old retrieval strategy would only pick up the changes when the token has been transferred. If our goal is to keep token metadata as up-to-date as possible, we have to constantly renew it even for tokens that are not involved in any recent transfers. This is easy to implement with our new metadata retrieval strategy: simply add a field to the Token entity that tracks the block height of the most recent metadata update and select Tokens that were updated some fixed number of blocks ago for metadata updates.

Here is how the new metadata retrieval strategy reflects in the batch handler code: TypeormDatabase(), async (ctx) => {
let rawTransfers: RawTransfer[] = getRawTransfers(ctx)

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

+ if (ctx.isHead) {
+ let updatedTokens = await updateTokensWithOutdatedMetadata(ctx)
+ await
+`Updated metadata for ${updatedTokens.length} tokens`)
+ }

Full implementation requires changes to the schema, replacement of completeTokens() with updateTokensWithOutdatedMetadata() and rewrites of createTokens() and selectivelyUpdateMetadata(). It is available in this branch. The resulting squid took 10 minutes for the initial sync, then 50 minutes more to retrieve metadata at least once for every token.

Extra: Using Multicall for metadata exploration

In part 3 of this tutorial we explored metadata URIs by running tokenURI() directly on the BAYC token contract. This process took several hours. Replacing the exploration code with its [multicall]-based equivalent, we can reduce that time to about 17 minutes. Starting with the code as it was at the end of part two, get src/abi/multicall.ts by running

npx squid-evm-typegen --multicall src/abi 0xbc4ca0eda7647a8ab7c2061c2e118a18a936f13d#bayc

then supply a RPC endpoint to the processor and add the code for batch URI retrieval to the batch handler:

 let processor = new EvmBatchProcessor()
archive: lookupArchive('eth-mainnet'),
+ chain: '',

# ...

+import { BigNumber } from 'ethers'
+import { Multicall } from './abi/multicall'

+const MULTICALL_ADDRESS = '0x5ba1e12693dc8f9c48aad8770482f4739beed696'
+const MUTLTICALL_BATCH_SIZE = 100 TypeormDatabase(), async (ctx) => {
let tokens: Map<string, Token> = createTokens(rawTransfers, owners)
let transfers: Transfer[] = createTransfers(rawTransfers, owners, tokens)

+ let lastBatchBlockHeader = ctx.blocks[ctx.blocks.length-1].header
+ let contract = new Multicall(ctx, lastBatchBlockHeader, MULTICALL_ADDRESS)
+ let tokenURIs = await contract.aggregate(
+ bayc.functions.tokenURI,
+ [...tokens.values()].map(t => [BigNumber.from(t.tokenId)]),
+ )
+ for (let uri of tokenURIs) {
+`Retrieved a metadata URI: ${uri}`)
+ }