EVM RPC read benchmark script for state store load testing#3004
EVM RPC read benchmark script for state store load testing#3004
Conversation
Hammers a node endpoint with debug_traceBlockByNumber, debug_traceTransaction, eth_getBalance, eth_getStorageAt, eth_getCode, and eth_getTransactionCount across 5 phases (sequential, concurrent, per-method, mixed workload). Reports p50/p95/p99 latencies and rps. Made-with: Cursor
|
The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).
|
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #3004 +/- ##
==========================================
- Coverage 58.76% 58.59% -0.17%
==========================================
Files 2090 2091 +1
Lines 172323 172841 +518
==========================================
+ Hits 101257 101269 +12
- Misses 62090 62597 +507
+ Partials 8976 8975 -1
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| for _, s := range stats { | ||
| s.Duration = elapsed | ||
| } |
Check warning
Code scanning / CodeQL
Iteration over map Warning
| if len(s.Latencies) == 0 { | ||
| return 0 | ||
| } | ||
| idx := int(float64(len(s.Latencies)) * pct) |
Check notice
Code scanning / CodeQL
Floating point arithmetic Note
| } | ||
| return s.Latencies[idx] | ||
| } | ||
| rps := float64(s.Total) / s.Duration.Seconds() |
Check notice
Code scanning / CodeQL
Floating point arithmetic Note
| go func() { | ||
| defer wg.Done() | ||
| for idx := range work { | ||
| method, lat, err := workFn(idx) | ||
| record(method, lat, err) | ||
| } | ||
| }() |
Check notice
Code scanning / CodeQL
Spawning a Go routine Note
| totalDuration = stats[k].Duration | ||
| } | ||
| } | ||
| rps := float64(totalReqs) / totalDuration.Seconds() |
Check notice
Code scanning / CodeQL
Floating point arithmetic Note
Each RPC method is defined once in a single registry slice. Phases derive from heavy/light classification. Adding a new method is one line. Use -methods flag to run a subset. Made-with: Cursor
Traces a configurable number of txs (-trace-discover, default 5) during discovery to extract real (address, slot) pairs. eth_getStorageAt now queries actual populated slots instead of random ones. Falls back to random slots when discovery is disabled or finds nothing. Made-with: Cursor
| for addr := range addrSet { | ||
| info.Addresses = append(info.Addresses, addr) | ||
| } |
Check warning
Code scanning / CodeQL
Iteration over map Warning
| for k, v := range s { | ||
| lightStats[k] = v | ||
| } |
Check warning
Code scanning / CodeQL
Iteration over map Warning
| for addr, acct := range prestate { | ||
| for slot := range acct.Storage { | ||
| key := addr + "|" + slot | ||
| if !seen[key] { | ||
| seen[key] = true | ||
| slots = append(slots, storageSlot{Address: addr, Slot: slot}) | ||
| } | ||
| } | ||
| } |
Check warning
Code scanning / CodeQL
Iteration over map Warning
| for slot := range acct.Storage { | ||
| key := addr + "|" + slot | ||
| if !seen[key] { | ||
| seen[key] = true | ||
| slots = append(slots, storageSlot{Address: addr, Slot: slot}) | ||
| } | ||
| } |
Check warning
Code scanning / CodeQL
Iteration over map Warning
sei-db/tools/rpc_bench/main.go
Outdated
| if err != nil { | ||
| return nil, elapsed, err | ||
| } | ||
| defer resp.Body.Close() |
Phase 1 now traces every discovered block once and prints a table: BLOCK TXS LATENCY 196571121 97 2.5s 196571111 1 120ms Made-with: Cursor
Made-with: Cursor
| } | ||
| avgGasPerTx := 0.0 | ||
| if len(info.Transactions) > 0 { | ||
| avgGasPerTx = float64(info.GasUsed) / float64(len(info.Transactions)) |
Check notice
Code scanning / CodeQL
Floating point arithmetic Note
sei-db/tools/rpc_bench/main.go
Outdated
| } | ||
| avgGasPerTx := 0.0 | ||
| if len(b.Transactions) > 0 { | ||
| avgGasPerTx = float64(b.GasUsed) / float64(len(b.Transactions)) |
Check notice
Code scanning / CodeQL
Floating point arithmetic Note
Kbhat1
changed the title
This keeps the benchmark workload aligned with the selected methods, avoids extra hidden trace work, and fixes concurrent random request selection. Made-with: Cursor
Resolve the go.mod conflict by keeping main's shared dependency updates and preserving rpc_bench's x/image dependency so the benchmark tool still builds. Made-with: Cursor
Check file-handle and parse errors explicitly and tighten plot path handling so golangci-lint passes on the benchmark tool. Made-with: Cursor
Drop the unused x/exp typeparams entry and stale x/text checksums so go mod tidy stays clean on the benchmark branch. Made-with: Cursor
- Adds an EVM RPC read benchmark tool for e2e node debug_trace and state-store load testing - Hits heavy trace endpoints plus common state reads, including `eth_getLogs`, and reports latency/RPS across per-method and mixed workloads - Includes lightweight per-block trace summaries and optional PNG plots to help spot trace performance patterns. - Verified fully on node Example: <img width="1400" height="900" alt="latency_vs_gas" src="https://github.com/user-attachments/assets/4a46bd97-fef6-49e4-b87d-85c48e253c56" />
3 participants
Describe your changes and provide context
Adds an EVM RPC read benchmark tool for e2e node debug_trace and state-store load testing
Hits heavy trace endpoints plus common state reads, including
eth_getLogs, and reports latency/RPS across per-method and mixed workloadsIncludes lightweight per-block trace summaries and optional PNG plots to help spot trace performance patterns.
Testing performed to validate your change
Example: