rapidclient

A multi-mode e2e test utility image. The mode is selected by the MODE environment variable; an unset MODE defaults to client (the original behaviour). See DESIGN.md for the full contract and rationale.

Modes

server mode

Listens for TCP HTTP and UDP echo on the same port (PORT, default 5000), dual-stack:

• GET /length/{N} — response body of exactly N whitespace-free bytes.
• POST /post — returns the number of bytes received (GET returns help text).
• GET / — static sanity string.
• UDP — echoes each datagram back verbatim.

docker run --rm -e MODE=server -p 5000:5000 quay.io/tigeradev/rapidclient

client mode

A simple HTTP client tool that forces source port reuse by bypassing TIME_WAIT state, designed for testing Maglev consistent hashing and load balancer behavior.

Features

• Source Port Reuse: Forces connections to use a specific source port, bypassing TIME_WAIT
• New TCP Connection: Each call establishes a fresh TCP connection (no connection reuse)
• Socket Options: Sets SO_REUSEADDR for rapid port reuse
• Simple Output: Sends a single request and prints the raw response
• Configurable: Command-line options for URL, source port, and timeout

Usage

Docker Run

# Basic Docker usage with host networking
docker run --rm --network host rapidclient:latest -url "http://10.103.218.83:8080/shell?cmd=hostname"

# With verbose output
docker run --rm --network host rapidclient:latest -url "http://10.103.218.83:8080/shell?cmd=hostname" -v

# With custom source port
docker run --rm --network host rapidclient:latest -url "http://10.103.218.83:8080/shell?cmd=hostname" -port 54321

# With custom timeout
docker run --rm --network host rapidclient:latest -url "http://10.103.218.83:8080/shell?cmd=hostname" -timeout 10s

Basic Usage

# Send a single request to a service
./rapidclient -url "http://10.96.0.1:8080/shell?cmd=hostname"

# Use a specific source port
./rapidclient -url "http://10.96.0.1:8080/shell?cmd=hostname" -port 12345

# Verbose output with status information
./rapidclient -url "http://10.96.0.1:8080/shell?cmd=hostname" -v

Command Line Options

• -url string: Target URL to send request to (required)
• -port int: Source port to use for connection (default: 12345)
• -timeout duration: Request timeout (default: 30s)
• -v: Verbose logging (shows status and debug info)

Examples

# Test Maglev load balancing with consistent source port
./rapidclient -url "http://service-ip:8080/shell?cmd=hostname" -port 12345

# Test with custom timeout
./rapidclient -url "http://service-ip:8080/api/health" -timeout 10s

# Verbose mode for debugging
./rapidclient -url "http://service-ip:8080/shell?cmd=hostname" -port 12345 -v

Technical Details

• Uses SO_REUSEADDR socket option to allow rapid port reuse
• Disables HTTP keep-alive to force new connections
• Custom dialer with specific source port binding
• Handles both IPv4 and IPv6 connections

Output

Normal Mode

$ ./rapidclient -url "http://10.96.0.1:8080/shell?cmd=hostname"
{"output":"backend-pod-5\n"}

Verbose Mode

$ ./rapidclient -url "http://10.96.0.1:8080/shell?cmd=hostname" -v
2025/09/04 10:30:00 Sending request to: http://10.96.0.1:8080/shell?cmd=hostname
2025/09/04 10:30:00 Using source port: 12345
2025/09/04 10:30:00 Timeout: 30s
Status: 200 OK
Response: {"output":"backend-pod-5\n"}

How the e2e tests get this image

The tests reference the image via images.RapidClientImage() (e2e/pkg/utils/images/images.go), which is env-driven:

• PR CI on gcp-kubeadm: PR builds have no registry push credential, so .semaphore/end-to-end/scripts/phases/load_images.sh builds this image from the PR source and imports it straight into each node's containerd (and the external node's docker), then exports RAPIDCLIENT_TAG (e.g. pr-13105). The tests pin that tag with ImagePullPolicy: Never.
• Everything else (other providers, scheduled runs, local dev): RAPIDCLIENT_TAG is unset and the tests use the published quay.io/tigeradev/rapidclient:latest. If you change this image and want a non-gcp-kubeadm run to use your build, publish it (post-merge push-images/e2e-test.yml promotion) or set RAPIDCLIENT_TAG yourself to a tag the cluster can pull.

Integration with Tests

This tool can replace curl commands in tests for better reliability:

# Instead of:
curl --local-port 12345 -s http://service:8080/shell?cmd=hostname

# Use:
./rapidclient -url "http://service:8080/shell?cmd=hostname" -port 12345