feat(fd): event cache for Geth L1s (#3950)

Introduces a new event cache to the fault detector. Requires for running
the fault detector against Geth as the L1 node and with certain L1 node
providers.

.changeset/shaggy-donkeys-matter.md

+---
+'@eth-optimism/fault-detector': minor
+---
+
+Includes a new event caching mechanism for running the fault detector against Geth.

packages/fault-detector/src/helpers.ts

-import { Contract, ethers } from 'ethers'
+import { Contract } from 'ethers'
+
+/**
+ * Partial event interface, meant to reduce the size of the event cache to avoid
+ * running out of memory.
+ */
+export interface PartialEvent {
+  blockNumber: number
+  transactionHash: string
+  args: any
+}
+
+// Event caching is necessary for the fault detector to work properly with Geth.
+const caches: {
+  [contractAddress: string]: {
+    highestBlock: number
+    eventCache: Map<string, PartialEvent>
+  }
+} = {}
+
+/**
+ * Retrieves the cache for a given address.
+ *
+ * @param address Address to get cache for.
+ * @returns Address cache.
+ */
+const getCache = (
+  address: string
+): {
+  highestBlock: number
+  eventCache: Map<string, PartialEvent>
+} => {
+  if (!caches[address]) {
+    caches[address] = {
+      highestBlock: 0,
+      eventCache: new Map(),
+    }
+  }
+
+  return caches[address]
+}
+
+/**
+ * Updates the event cache for the SCC.
+ *
+ * @param scc The State Commitment Chain contract.
+ */
+export const updateStateBatchEventCache = async (
+  scc: Contract
+): Promise<void> => {
+  const cache = getCache(scc.address)
+  let currentBlock = cache.highestBlock
+  const endingBlock = await scc.provider.getBlockNumber()
+  let step = endingBlock - currentBlock
+  let failures = 0
+  while (currentBlock < endingBlock) {
+    try {
+      const events = await scc.queryFilter(
+        scc.filters.StateBatchAppended(),
+        currentBlock,
+        currentBlock + step
+      )
+      for (const event of events) {
+        cache.eventCache[event.args._batchIndex.toNumber()] = {
+          blockNumber: event.blockNumber,
+          transactionHash: event.transactionHash,
+          args: event.args,
+        }
+      }
+
+      // Update the current block and increase the step size for the next iteration.
+      currentBlock += step
+      step = Math.ceil(step * 2)
+    } catch {
+      // Might happen if we're querying too large an event range.
+      step = Math.floor(step / 2)
+
+      // When the step gets down to zero, we're pretty much guaranteed that range size isn't the
+      // problem. If we get three failures like this in a row then we should just give up.
+      if (step === 0) {
+        failures++
+      } else {
+        failures = 0
+      }
+
+      // We've failed 3 times in a row, we're probably stuck.
+      if (failures >= 3) {
+        throw new Error('failed to update event cache')
+      }
+    }
+  }
+
+  // Update the highest block.
+  cache.highestBlock = endingBlock
+}
 
 /**
  * Finds the Event that corresponds to a given state batch by index.
@@ -10,20 +104,23 @@ import { Contract, ethers } from 'ethers'
 export const findEventForStateBatch = async (
   scc: Contract,
   index: number
-): Promise<ethers.Event> => {
-  const events = await scc.queryFilter(scc.filters.StateBatchAppended(index))
+): Promise<PartialEvent> => {
+  const cache = getCache(scc.address)
 
-  // Only happens if the batch with the given index does not exist yet.
-  if (events.length === 0) {
-    throw new Error(`unable to find event for batch`)
+  // Try to find the event in cache first.
+  if (cache.eventCache[index]) {
+    return cache.eventCache[index]
   }
 
-  // Should never happen.
-  if (events.length > 1) {
-    throw new Error(`found too many events for batch`)
+  // Update the event cache if we don't have the event.
+  await updateStateBatchEventCache(scc)
+
+  // Event better be in cache now!
+  if (cache.eventCache[index] === undefined) {
+    throw new Error(`unable to find event for batch ${index}`)
   }
 
-  return events[0]
+  return cache.eventCache[index]
 }
 
 /**
@@ -45,7 +142,7 @@ export const findFirstUnfinalizedStateBatchIndex = async (
   while (lo !== hi) {
     const mid = Math.floor((lo + hi) / 2)
     const event = await findEventForStateBatch(scc, mid)
-    const block = await event.getBlock()
+    const block = await scc.provider.getBlock(event.blockNumber)
 
     if (block.timestamp + fpw < latestBlock.timestamp) {
       lo = mid + 1

packages/fault-detector/src/service.ts

@@ -14,6 +14,8 @@ import { version } from '../package.json'
 import {
   findFirstUnfinalizedStateBatchIndex,
   findEventForStateBatch,
+  updateStateBatchEventCache,
+  PartialEvent,
 } from './helpers'
 
 type Options = {
@@ -95,6 +97,10 @@ export class FaultDetector extends BaseServiceV2<Options, Metrics, State> {
     this.state.scc = this.state.messenger.contracts.l1.StateCommitmentChain
     this.state.fpw = (await this.state.scc.FRAUD_PROOF_WINDOW()).toNumber()
 
+    // Populate the event cache.
+    this.logger.info(`warming event cache, this might take a while...`)
+    await updateStateBatchEventCache(this.state.scc)
+
     // Figure out where to start syncing from.
     if (this.options.startBatchIndex === -1) {
       this.logger.info(`finding appropriate starting height`)
@@ -165,7 +171,7 @@ export class FaultDetector extends BaseServiceV2<Options, Metrics, State> {
       latestIndex: latestBatchIndex,
     })
 
-    let event: ethers.Event
+    let event: PartialEvent
     try {
       event = await findEventForStateBatch(
         this.state.scc,
@@ -187,7 +193,9 @@ export class FaultDetector extends BaseServiceV2<Options, Metrics, State> {
 
     let batchTransaction: Transaction
     try {
-      batchTransaction = await event.getTransaction()
+      batchTransaction = await this.options.l1RpcProvider.getTransaction(
+        event.transactionHash
+      )
     } catch (err) {
       this.logger.error(`got error when connecting to node`, {
         error: err,

packages/fault-detector/test/helpers.spec.ts

@@ -92,30 +92,6 @@ describe('helpers', () => {
         ).to.eventually.be.rejectedWith('unable to find event for batch')
       })
     })
-
-    describe('when more than one event exists', () => {
-      beforeEach(async () => {
-        await StateCommitmentChain.appendStateBatch(
-          [hre.ethers.constants.HashZero],
-          0
-        )
-        await hre.ethers.provider.send('hardhat_setStorageAt', [
-          ChainStorageContainer.address,
-          '0x2',
-          hre.ethers.constants.HashZero,
-        ])
-        await StateCommitmentChain.appendStateBatch(
-          [hre.ethers.constants.HashZero],
-          0
-        )
-      })
-
-      it('should throw an error', async () => {
-        await expect(
-          findEventForStateBatch(StateCommitmentChain, 0)
-        ).to.eventually.be.rejectedWith('found too many events for batch')
-      })
-    })
   })
 
   describe('findFirstUnfinalizedIndex', () => {