PR: PR comments addressed Part 1

# Conflicts:
#	eth/tracers/firehose.go
#	eth/tracers/internal/tracetest/firehose/firehose_test.go
This commit is contained in:
David Zhou 2025-05-13 15:01:56 -04:00
parent a053ac76fd
commit 326a3ee583
3 changed files with 46 additions and 104 deletions

View file

@ -144,13 +144,14 @@ func (c *FirehoseConfig) ForcedBackwardCompatibility() bool {
type Firehose struct {
// Global state
outputBuffer *bytes.Buffer
initSent *atomic.Bool
config *FirehoseConfig
chainConfig *params.ChainConfig
hasher crypto.KeccakState // Keccak256 hasher instance shared across tracer needs (non-concurrent safe)
hasherBuf common.Hash // Keccak256 hasher result array shared across tracer needs (non-concurrent safe)
tracerID string
outputBuffer *bytes.Buffer
initSent *atomic.Bool
config *FirehoseConfig
chainConfig *params.ChainConfig
hasher crypto.KeccakState // Keccak256 hasher instance shared across tracer needs (non-concurrent safe)
hasherBuf common.Hash // Keccak256 hasher result array shared across tracer needs (non-concurrent safe)
tracerID string
closeChannels sync.Once
// The FirehoseTracer is used in multiple chains, some for which were produced using a legacy version
// of the whole tracing infrastructure. This legacy version had many small bugs here and there that
// we must "reproduce" on some chain to ensure that the FirehoseTracer produces the same output
@ -173,6 +174,8 @@ type Firehose struct {
blockReorderOrdinalSnapshot uint64
blockReorderOrdinalOnce sync.Once
blockIsGenesis bool
blockPrintQueue chan *blockPrintJob
blockFlushDone sync.WaitGroup
// Transaction state
evm *tracing.VMContext
@ -190,13 +193,6 @@ type Firehose struct {
// Testing state, only used in tests and private configs
testingBuffer *bytes.Buffer
testingIgnoreGenesisBlock bool
// Worker queue for block printing
blockPrintQueue chan *blockPrintJob
blockOutputQueue chan *blockOutput
flushDone sync.WaitGroup
flushOutputDone sync.WaitGroup
closeOnce sync.Once
}
const FirehoseProtocolVersion = "3.0"
@ -260,38 +256,11 @@ func NewFirehose(config *FirehoseConfig) *Firehose {
}
if config.ConcurrentBlockFlushing {
log.Info("Concurrent block flushing enabled: starting goroutine...")
const numWorkers = 10 // TODO: This becomes a parameter
firehose.blockPrintQueue = make(chan *blockPrintJob, 100) // TODO: Optimal buffer size tbd
firehose.blockOutputQueue = make(chan *blockOutput, 100)
for i := 0; i < numWorkers; i++ {
firehose.flushDone.Add(1)
go firehose.blockPrintWorker()
}
// Output channel to order the flushing linearly
firehose.flushOutputDone.Add(1)
go func() {
defer firehose.flushOutputDone.Done()
expected := uint64(0)
buffer := make(map[uint64][]byte)
for result := range firehose.blockOutputQueue {
buffer[result.blockNum] = result.data
for {
data, ok := buffer[expected]
if !ok {
break
}
firehose.flushToFirehose(data)
delete(buffer, expected)
expected++
}
}
}()
log.Info("Firehose concurrent block flushing enabled, starting block " +
"print worker goroutine")
firehose.blockPrintQueue = make(chan *blockPrintJob, 100)
firehose.blockFlushDone.Add(1)
go firehose.blockPrintWorker()
}
return firehose
@ -660,9 +629,8 @@ func (f *Firehose) reorderCallOrdinals(call *pbeth.Call, ordinalBase uint64) (or
}
func (f *Firehose) OnClose() {
log.Info("Firehose closing...")
if f.concurrentBlockFlushing {
log.Info("Closing channel: flushing the remaining blocks to firehose")
log.Info("Firehose closing, flushing queued blocks to standard output")
f.CloseBlockPrintQueue()
}
}
@ -1857,6 +1825,9 @@ func (f *Firehose) isChainOneOf(chainIDs ...*big.Int) bool {
}
func isChainIDOneOf(actualChainID *big.Int, chainIDs ...*big.Int) bool {
if actualChainID == nil {
return false
}
for _, chainID := range chainIDs {
if actualChainID.Cmp(chainID) == 0 {
return true
@ -1885,13 +1856,14 @@ func (f *Firehose) panicInvalidState(msg string, callerSkip int) string {
// printBlockToFirehose is a helper function to print a block to Firehose protocl format.
func (f *Firehose) printBlockToFirehose(block *pbeth.Block, finalityStatus *FinalityStatus) {
var buf bytes.Buffer
marshalled, err := proto.Marshal(block)
if err != nil {
panic(fmt.Errorf("failed to marshal block: %w", err))
}
// TODO: If multiple goroutine, this becomes shared resource
f.outputBuffer.Reset()
previousHash := block.PreviousID()
previousNum := 0
if block.Number > 0 {
@ -1910,9 +1882,9 @@ func (f *Firehose) printBlockToFirehose(block *pbeth.Block, finalityStatus *Fina
}
// **Important* The final space in the Sprintf template is mandatory!
buf.WriteString(fmt.Sprintf("FIRE BLOCK %d %s %d %s %d %d ", block.Number, hex.EncodeToString(block.Hash), previousNum, previousHash, libNum, block.MustTime().UnixNano()))
f.outputBuffer.WriteString(fmt.Sprintf("FIRE BLOCK %d %s %d %s %d %d ", block.Number, hex.EncodeToString(block.Hash), previousNum, previousHash, libNum, block.MustTime().UnixNano()))
encoder := base64.NewEncoder(base64.StdEncoding, &buf)
encoder := base64.NewEncoder(base64.StdEncoding, f.outputBuffer)
if _, err = encoder.Write(marshalled); err != nil {
panic(fmt.Errorf("write to encoder should have been infaillible: %w", err))
}
@ -1921,16 +1893,9 @@ func (f *Firehose) printBlockToFirehose(block *pbeth.Block, finalityStatus *Fina
panic(fmt.Errorf("closing encoder should have been infaillible: %w", err))
}
buf.WriteString("\n")
f.outputBuffer.WriteString("\n")
if f.concurrentBlockFlushing {
f.blockOutputQueue <- &blockOutput{
blockNum: block.Number,
data: buf.Bytes(),
}
} else {
f.flushToFirehose(buf.Bytes())
}
f.flushToFirehose(f.outputBuffer.Bytes())
}
// printToFirehose is an easy way to print to Firehose format, it essentially
@ -2871,25 +2836,20 @@ type blockPrintJob struct {
}
func (f *Firehose) blockPrintWorker() {
defer f.flushDone.Done()
defer f.blockFlushDone.Done()
for job := range f.blockPrintQueue {
f.printBlockToFirehose(job.block, job.finality)
}
}
// CloseBlockPrintQueue signals block printing goroutines to shut down and waits for them.
// It blocks until all concurrent block flushing operations are completed, ensuring a clean
// shutdown of the printing pipeline.
func (f *Firehose) CloseBlockPrintQueue() {
if f.concurrentBlockFlushing {
f.closeOnce.Do(func() {
f.closeChannels.Do(func() {
close(f.blockPrintQueue)
f.flushDone.Wait()
close(f.blockOutputQueue)
f.flushOutputDone.Wait()
f.blockFlushDone.Wait()
})
}
}
type blockOutput struct {
blockNum uint64
data []byte
}

View file

@ -84,6 +84,12 @@ The implementation was validated at multiple levels to ensure correctness, confi
## Section 4: Analysis
The specifications of the operating system used for testing are as follows:
Model: Macbook Air \
Processor: Apple M1 chip \
Memory: 8 GB
### 4.1 Results
The following outlines the results for metric three:
@ -92,18 +98,14 @@ The following outlines the results for metric three:
**No concurrency**
Run 1: 71.49s user 15.56s system 56% cpu 2:34.28 total
Run 2: 69.77s user 15.60s system 54% cpu 2:37.73 total
Run 1: 71.49s user 15.56s system 56% cpu 2:34.28 total\
Run 2: 69.77s user 15.60s system 54% cpu 2:37.73 total\
Run 3: 68.81s user 15.12s system 53% cpu 2:38.18 total
**Concurrency**
Run 1: 69.61s user 14.97s system 55% cpu 2:32.85 total
Run 2: 67.94s user 15.13s system 54% cpu 2:33.59 total
Run 1: 69.61s user 14.97s system 55% cpu 2:32.85 total\
Run 2: 67.94s user 15.13s system 54% cpu 2:33.59 total\
Run 3: 68.60s user 16.37s system 48% cpu 2:54.22 total (Not sure what happened here)
**Until Block 100000**
@ -133,8 +135,7 @@ Run 3: 68.60s user 16.37s system 48% cpu 2:54.22 total (Not sure what happened h
### 4.2 Discussion
Run 3 with concurrency seems to be an outlier. Without it, the general trend would be that every block saves around 0.0006 second, or 0.6 millisecond.
Run 3 with concurrency seems to be an outlier. Without it, the general trend would be that every block saves around 0.0006 second, or 0.6 millisecond. \
User seems slightly lower, whereas system and cpu are relatively the same.
## Section 5: Conclusion

View file

@ -94,7 +94,9 @@ func TestFirehose_BlocksPrintToFirehose_MultipleBlocksInOrder(t *testing.T) {
"Expected %d blocks in output, found %d", blockCount, len(extractedBlocks))
// Verify blocks in order
verifyBlockSequence(t, extractedBlocks, baseBlockNum)
for i, block := range extractedBlocks {
require.Equal(t, baseBlockNum+uint64(i), block.number, "Blocks out of order at position %d", i)
}
// Verify block hashes
for _, block := range extractedBlocks {
@ -139,24 +141,3 @@ func extractBlocksFromOutput(t *testing.T, output string) []extractedBlock {
return blocks
}
func verifyBlockSequence(t *testing.T, blocks []extractedBlock, baseBlockNum uint64) {
t.Helper()
// First block should be the base block number
require.Equal(t, baseBlockNum, blocks[0].number,
"First block should be %d, got %d", baseBlockNum, blocks[0].number)
// Last block should be base + count - 1
expectedLast := baseBlockNum + uint64(len(blocks)) - 1
require.Equal(t, expectedLast, blocks[len(blocks)-1].number,
"Last block should be %d, got %d", expectedLast, blocks[len(blocks)-1].number)
// Verify sequence
for i := 0; i < len(blocks)-1; i++ {
current := blocks[i].number
next := blocks[i+1].number
require.Equal(t, current+1, next,
"Blocks out of order at position %d: %d followed by %d", i, current, next)
}
}