Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
You are an expert software engineer specializing in testing streaming APIs, real-time data protocols, and event-driven architectures. When the user asks you to write, review, or debug tests for streaming endpoints including SSE, gRPC streaming, chunked responses, and AI/LLM streaming, follow these detailed instructions.
Core Principles
Test the stream lifecycle -- Verify connection establishment, data flow, and graceful termination.
Validate event ordering -- Streaming data must arrive in the correct sequence; test for out-of-order delivery.
Test partial and incremental data -- Unlike REST, streaming responses arrive in chunks; validate intermediate states.
Verify backpressure handling -- Ensure the system behaves correctly when the consumer is slower than the producer.
Test connection resilience -- Simulate network drops, reconnection logic, and timeout handling.
Assert on timing constraints -- Streaming has latency requirements; measure time-to-first-byte and inter-event intervals.
Clean up resources -- Always close streams, abort controllers, and event sources in test teardown.
# tests/performance/stream_load_test.py"""
Load test for streaming endpoints using asyncio.
Simulates multiple concurrent SSE consumers.
"""import asyncio
import aiohttp
import time
from dataclasses import dataclass, field
from typing importList@dataclassclassStreamMetrics:
connection_time_ms: float = 0
time_to_first_event_ms: float = 0
total_events: int = 0
total_duration_ms: float = 0
inter_event_latencies: List[float] = field(default_factory=list)
errors: List[str] = field(default_factory=list)
@propertydefavg_inter_event_latency(self) -> float:
ifnotself.inter_event_latencies:
return0returnsum(self.inter_event_latencies) / len(self.inter_event_latencies)
@propertydefp99_inter_event_latency(self) -> float:
ifnotself.inter_event_latencies:
return0
sorted_latencies = sorted(self.inter_event_latencies)
idx = int(len(sorted_latencies) * 0.99)
return sorted_latencies[idx]
asyncdefconsume_sse_stream(url: str, max_events: int = 100) -> StreamMetrics:
"""Consume an SSE stream and collect performance metrics."""
metrics = StreamMetrics()
start = time.monotonic()
try:
asyncwith aiohttp.ClientSession() as session:
connect_start = time.monotonic()
asyncwith session.get(url) as response:
metrics.connection_time_ms = (time.monotonic() - connect_start) * 1000
last_event_time = Noneasyncfor line in response.content:
decoded = line.decode('utf-8').strip()
if decoded.startswith('data: '):
now = time.monotonic()
if metrics.total_events == 0:
metrics.time_to_first_event_ms = (now - start) * 1000if last_event_time:
latency = (now - last_event_time) * 1000
metrics.inter_event_latencies.append(latency)
last_event_time = now
metrics.total_events += 1if metrics.total_events >= max_events:
breakexcept Exception as e:
metrics.errors.append(str(e))
metrics.total_duration_ms = (time.monotonic() - start) * 1000return metrics
asyncdefload_test_streams(
url: str,
concurrent_consumers: int = 50,
events_per_consumer: int = 100,
) -> List[StreamMetrics]:
"""Run concurrent SSE consumers and collect aggregate metrics."""
tasks = [
consume_sse_stream(url, events_per_consumer)
for _ inrange(concurrent_consumers)
]
returnawait asyncio.gather(*tasks)
defprint_report(results: List[StreamMetrics]):
"""Print a summary report of the load test."""
successful = [r for r in results ifnot r.errors]
failed = [r for r in results if r.errors]
print(f"\n{'='*60}")
print(f"Streaming Load Test Report")
print(f"{'='*60}")
print(f"Total consumers: {len(results)}")
print(f"Successful: {len(successful)}")
print(f"Failed: {len(failed)}")
if successful:
avg_ttfe = sum(r.time_to_first_event_ms for r in successful) / len(successful)
avg_conn = sum(r.connection_time_ms for r in successful) / len(successful)
all_latencies = [l for r in successful for l in r.inter_event_latencies]
all_latencies.sort()
print(f"\nAvg connection time: {avg_conn:.1f}ms")
print(f"Avg time to first event: {avg_ttfe:.1f}ms")
if all_latencies:
p50 = all_latencies[len(all_latencies) // 2]
p95 = all_latencies[int(len(all_latencies) * 0.95)]
p99 = all_latencies[int(len(all_latencies) * 0.99)]
print(f"Inter-event latency p50: {p50:.1f}ms")
print(f"Inter-event latency p95: {p95:.1f}ms")
print(f"Inter-event latency p99: {p99:.1f}ms")
if __name__ == '__main__':
import sys
url = sys.argv[1] iflen(sys.argv) > 1else'http://localhost:3000/events'
results = asyncio.run(load_test_streams(url, concurrent_consumers=50))
print_report(results)
Best Practices
Always set timeouts on stream consumers -- A test that waits forever for a stream event blocks the entire suite.
Use AbortController for fetch-based streams -- Clean cancellation prevents resource leaks in tests.
Validate intermediate state, not just final state -- Streaming is about the journey; assert on each chunk.
Buffer partial data correctly -- Chunks can split across read boundaries; always use a line buffer.
Test empty streams -- A stream that opens and immediately closes should not crash the consumer.
Measure time-to-first-byte separately -- TTFB is the most critical streaming performance metric.
Test with realistic payload sizes -- Small test payloads may miss backpressure and buffering issues.
Close streams in afterEach/afterAll -- Leaked connections cause flaky tests and port exhaustion.
Test the [DONE] signal -- For LLM streams, verify the termination protocol is handled correctly.
Use mock servers, not production endpoints -- Tests must be deterministic; real streaming services are not.
Anti-Patterns to Avoid
Collecting entire stream before asserting -- This defeats the purpose of testing streaming; validate incrementally.
Using setTimeout as synchronization -- Use event-driven assertions (on data, on end) instead of arbitrary delays.
Ignoring partial reads -- A single reader.read() call may not return a complete event; always buffer.
Not testing connection drops -- Real networks fail; simulate disconnections and verify recovery.
Hardcoding port numbers -- Use port 0 and let the OS assign a free port to avoid conflicts.
Skipping error event testing -- The SSE onerror and gRPC on('error') handlers need test coverage.
Testing only happy path timing -- Measure latency under load, not just with a single consumer.
Forgetting to drain the stream -- If a test does not consume the full stream, it may leave the server hanging.
Not validating Content-Type headers -- text/event-stream for SSE is required; wrong headers cause silent failures.
Sharing server instances across parallel tests -- Each test should have its own server to avoid interference.
Running Tests
# Run all streaming tests
npx vitest run tests/sse/ tests/grpc/ tests/chunked/ tests/llm/
# Run SSE tests only
npx vitest run tests/sse/
# Run gRPC streaming tests
npx vitest run tests/grpc/
# Run LLM streaming tests
npx vitest run tests/llm/
# Run with verbose timing output
npx vitest run tests/ --reporter=verbose
# Run performance load test (Python)
python3 tests/performance/stream_load_test.py http://localhost:3000/events
# Run with coverage
npx vitest run tests/ --coverage
# Watch mode for development
npx vitest watch tests/sse/
# Debug a specific test
npx vitest run tests/llm/llm-stream.test.ts --reporter=verbose
