Native Nginx Log Replay
Stream-parse Common or Combined Nginx logs natively without writing complex Lua scripts. Supports sequential, shuffled, and random replay modes.
--access-log access.log --replay-order shuffle --seed 42
RUST-NATIVE HTTP TRAFFIC REPLAY
Replay reality.
Find the limit.
rload is a Rust HTTP load generator with wrk-compatible CLI semantics. Versioned YAML workloads, typed CI assertions, and deterministic offline reports pair with Nginx and JSONL replay for reproducible load validation.
●●●
rload v0.3.1
$ rload -t 4 -c 200 -d 30s \
--log-replay access.log \
--log-order shuffle --seed 42 \
http://staging-api.internal
Running 30s test @ http://staging-api.internal
4 threads and 200 connections
Thread Stats Avg Max +/- Stdev
Latency 2.18ms 18.74ms 84.32%
Req/Sec 12.45k 15.92k 76.10%
1482701 requests in 30.00s, 3.24GB read
Requests/sec: 49,423.36
Average latency: 2.18ms
99% Latency: 8.12ms
Socket errors: connect 0, read 0, write 0, timeout 0
● log replay complete (shuffled, seed: 42)
WRK SEMANTICS PARITY
·
NGINX ACCESS LOG REPLAY
·
STRUCTURED JSONL REPLAY
·
COORDINATED OMISSION PREVENTION
·
STABLE STATIC BINARY
01 / Capabilities
From static benchmarking
to production-grade traffic replay.
Static single-endpoint benchmarks fail to trigger real-world caching or queueing bottlenecks. rload ensures every request and sequence is derived from real environments, 100% reproducible.
Structured JSONL Replay
Replay structured HTTP payloads (headers, methods, bodies) out-of-the-box. Gracefully mimic relative inter-arrival delays using high-precision timers.
--request-file queries.jsonl --replay-timestamps
wrk-Parity Footprint & Speed
Built on bare-metal mio with thread-affinity. Maintains a constant ~3.5MB RSS footprint. Delivers near-identical throughput and latency distribution (MAE < 1% vs. wrk) with zero scheduler noise.
Pacing & Anti-Coordinated Omission
Prevents Coordinated Omission by tracking latency against scheduled send times. Program custom load profiles (step, burst, ramp-up) via flexible stage triggers.
--replay-stages 10s:100,5s:1000
02 / Why rload?
wrk Semantics Parity.
Replay modern reality.
rload strikes the perfect balance between raw bare-metal execution speed and feature-rich traffic reproduction:
Feature / Dimension
wrk 4.2.0 (C)
k6 (Go / JS)
rload v0.3.1 (Rust)
Static HTTP Load
✓ Extreme
✓ Heavy Footprint
✓ Extreme (Parity)
Memory Footprint (RSS)
~3.5 MiB
> 100 MiB
~3.55 MiB (Parity)
Nginx Access Log Replay
✗ No (Requires Lua)
✗ Memory Prohibitive
✓ Native & Streamed
JSONL Custom Replay
✗ No (Requires Lua)
✓ Yes
✓ Native & Streamed
Rate Pacing
✗ No
✓ Yes
✓ High-Precision Native
Anti-Coordinated Omission (CO)
✗ No (Requires wrk2)
✗ No
✓ Native & Precise
CI/CD Runner Friendliness
✗ Requires compile/deps
✓ Yes
🏆 Excellent (Static Binary)
03 / Installation
One-second install.
Ready out of the box.
We provide first-class cross-platform support with single, dependency-free pre-compiled static binaries. Pick the method that fits your environment:
# Install via Homebrew (macOS & Linux)
brew install wenhaozhao/rload/rload
# Verify installation
rload --help# Install via cargo (for Rust toolchains)
cargo install rload
# Ensure ~/.cargo/bin is in your PATH
rload --help# For container & minimal server environments
# 1. Download the archive for your architecture from GitHub Releases
# 2. Extract and move the binary into your PATH, e.g.:
tar -xvf rload-v0.3.0-x86_64-unknown-linux-gnu.tar.gz
sudo mv rload /usr/local/bin/
# 3. Verify execution
rload --help04 / Practical Commands
Common engineering scenarios,
copy and run instantly.
🔥 Scenario A: Static HTTP benchmarking (wrk parity)
Bench a target with 4 threads and 200 connections for 30 seconds, tracking latency distribution:
rload -t 4 -c 200 -d 30s https://api.yourdomain.com📂 Scenario B: High-fidelity Nginx log shuffle replay
Replay an Nginx access log in a shuffled order using a deterministic seed for reproducible results:
rload -c 100 -d 60s --access-log ./access.log --replay-order shuffle --seed 12345 http://staging.internal⏳ Scenario C: Log replay with relative timestamp pacing
Reproduce real temporal gaps based on access-log timestamps at double speed (2.0x pacing):
rload -c 50 --access-log ./access.log --replay-timestamps --replay-speed 2.0 http://staging.internal📝 Scenario D: Structured JSONL request sequence replay
Sequentially replay structured query templates with custom headers, methods, and payloads:
rload -c 100 -d 5m --request-file ./requests.jsonl http://staging.internal