File size: 10,756 Bytes
dc4e6da | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 | """
Performance tests — latency and throughput
==========================================
Measures:
2.1 Lightweight endpoint latency (GET / and GET /health, N sequential samples)
2.2 POST /generate/pdf input-validation latency (schema rejection path — fast)
2.3 POST /generate/async input-validation latency (schema rejection path — fast)
2.4 GET /jobs/user/{id} latency (Supabase read — lightweight)
2.5 Sequential throughput across lightweight GETs
2.6 Concurrent lightweight GET requests (ThreadPoolExecutor)
NOTE: The generation endpoints (/generate/pdf, /generate/async) are NOT exercised
with real requests because they require a valid Supabase request_id and call
expensive downstream services (Claude API, PDF rendering). Instead, we measure
the *input-validation* (422 / 404) paths which are still real network round-trips
to the deployed API and reflect cold-path overhead.
All timings are collected into _perf_results and saved to artifacts/perf_metrics.json
at the end of the session so compile_results.py can embed them in the report.
"""
import json
import time
import pathlib
import statistics
import pytest
import requests
from concurrent.futures import ThreadPoolExecutor, as_completed
from tests.conftest import (
BASE_URL, TIMEOUT, SEED_IMAGE_URL,
NONEXISTENT_REQUEST_ID, NONEXISTENT_USER_ID,
MINIMAL_GENERATE_PAYLOAD,
)
ARTIFACTS = pathlib.Path(__file__).parent.parent / "artifacts"
ARTIFACTS.mkdir(exist_ok=True)
# ---------------------------------------------------------------------------
# Thresholds (loose — HuggingFace Spaces can have cold-start jitter)
# ---------------------------------------------------------------------------
MAX_HEALTH_MEAN_S = 10.0 # mean response time for /health
MAX_HEALTH_P95_S = 20.0 # 95th-percentile for /health
MAX_SCHEMA_REJECT_S = 5.0 # 422 responses should always be fast
MAX_USER_JOBS_S = 10.0 # Supabase read should be fast when warm
N_SAMPLES = 5 # sequential sample count for latency stats
N_CONCURRENT = 4 # worker count for concurrent test
# Shared results dict — populated by the test methods
_perf_results: dict = {}
# ---------------------------------------------------------------------------
# Session-end fixture — persist metrics to JSON for compile_results.py
# ---------------------------------------------------------------------------
@pytest.fixture(scope="session", autouse=True)
def _persist_perf_metrics():
"""Yield during the test session; write metrics JSON afterwards."""
yield
try:
(ARTIFACTS / "perf_metrics.json").write_text(
json.dumps(_perf_results, indent=2)
)
except Exception as e:
print(f"Warning: could not save perf metrics: {e}")
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def _timeit(fn) -> float:
t0 = time.perf_counter()
fn()
return time.perf_counter() - t0
def _stats(samples: list) -> dict:
return {
"n": len(samples),
"min_s": round(min(samples), 4),
"mean_s": round(statistics.mean(samples), 4),
"median_s": round(statistics.median(samples), 4),
"max_s": round(max(samples), 4),
"p95_s": round(sorted(samples)[int(len(samples) * 0.95)], 4)
if len(samples) >= 2 else round(max(samples), 4),
}
# ---------------------------------------------------------------------------
# 2.1 Lightweight endpoint latency
# ---------------------------------------------------------------------------
class TestLightweightEndpointLatency:
"""Sequential GET / and GET /health — N samples each."""
def test_root_latency_under_threshold(self, http):
samples = []
for _ in range(N_SAMPLES):
samples.append(_timeit(lambda: http.get(f"{BASE_URL}/", timeout=TIMEOUT)))
st = _stats(samples)
_perf_results["root_latency"] = st
print(f"\n GET / — {st}")
assert st["mean_s"] < MAX_HEALTH_MEAN_S, (
f"GET / mean latency {st['mean_s']:.3f}s exceeds {MAX_HEALTH_MEAN_S}s"
)
def test_health_latency_under_threshold(self, http):
samples = []
for _ in range(N_SAMPLES):
samples.append(_timeit(lambda: http.get(f"{BASE_URL}/health", timeout=TIMEOUT)))
st = _stats(samples)
_perf_results["health_latency"] = st
print(f"\n GET /health — {st}")
assert st["mean_s"] < MAX_HEALTH_MEAN_S, (
f"GET /health mean latency {st['mean_s']:.3f}s exceeds {MAX_HEALTH_MEAN_S}s"
)
def test_user_jobs_latency_under_threshold(self, http):
url = f"{BASE_URL}/jobs/user/{NONEXISTENT_USER_ID}"
samples = []
for _ in range(N_SAMPLES):
samples.append(_timeit(lambda: http.get(url, timeout=TIMEOUT)))
st = _stats(samples)
_perf_results["user_jobs_latency"] = st
print(f"\n GET /jobs/user/{{id}} — {st}")
assert st["mean_s"] < MAX_USER_JOBS_S, (
f"GET /jobs/user mean latency {st['mean_s']:.3f}s exceeds {MAX_USER_JOBS_S}s"
)
# ---------------------------------------------------------------------------
# 2.2 Input-validation (422) path latency for /generate/pdf
# ---------------------------------------------------------------------------
class TestGeneratePdfValidationLatency:
"""422 responses are pure FastAPI work (no DB / LLM calls)."""
def test_schema_rejection_is_fast(self, http):
bad_payload = {} # missing required request_id → immediate 422
samples = []
for _ in range(N_SAMPLES):
samples.append(_timeit(
lambda: http.post(f"{BASE_URL}/generate/pdf", json=bad_payload, timeout=TIMEOUT)
))
st = _stats(samples)
_perf_results["pdf_validation_latency"] = st
print(f"\n POST /generate/pdf (422 path) — {st}")
assert st["mean_s"] < MAX_SCHEMA_REJECT_S, (
f"422 path mean {st['mean_s']:.3f}s exceeds {MAX_SCHEMA_REJECT_S}s"
)
# ---------------------------------------------------------------------------
# 2.3 Input-validation (422) path latency for /generate/async
# ---------------------------------------------------------------------------
class TestGenerateAsyncValidationLatency:
def test_schema_rejection_is_fast(self, http):
bad_payload = {}
samples = []
for _ in range(N_SAMPLES):
samples.append(_timeit(
lambda: http.post(f"{BASE_URL}/generate/async", json=bad_payload, timeout=TIMEOUT)
))
st = _stats(samples)
_perf_results["async_validation_latency"] = st
print(f"\n POST /generate/async (422 path) — {st}")
assert st["mean_s"] < MAX_SCHEMA_REJECT_S
# ---------------------------------------------------------------------------
# 2.4 Sequential throughput — GET /health
# ---------------------------------------------------------------------------
class TestSequentialThroughput:
"""How many /health requests can the API serve sequentially per second?"""
def test_health_sequential_throughput(self, http):
n = N_SAMPLES
t_start = time.perf_counter()
statuses = []
for _ in range(n):
r = http.get(f"{BASE_URL}/health", timeout=TIMEOUT)
statuses.append(r.status_code)
wall = time.perf_counter() - t_start
ok = statuses.count(200)
req_per_min = round(ok / wall * 60, 2)
result = {
"requests": n, "ok": ok, "failures": n - ok,
"wall_s": round(wall, 3),
"mean_per_req_s": round(wall / n, 3),
"req_per_min": req_per_min,
}
_perf_results["sequential_throughput"] = result
print(f"\n Sequential throughput — {result}")
assert ok == n, f"Expected all {n} requests to succeed, got {ok}"
# ---------------------------------------------------------------------------
# 2.5 Concurrent GET /health
# ---------------------------------------------------------------------------
class TestConcurrentRequests:
"""Fire N concurrent GET /health requests and measure wall-clock time."""
def _run_concurrent(self, n_workers: int, http_session: requests.Session):
url = f"{BASE_URL}/health"
results = []
wall_start = time.perf_counter()
def _fetch():
t0 = time.perf_counter()
r = requests.get(url, timeout=TIMEOUT)
elapsed = time.perf_counter() - t0
return r.status_code, elapsed
with ThreadPoolExecutor(max_workers=n_workers) as pool:
futures = [pool.submit(_fetch) for _ in range(n_workers)]
for f in as_completed(futures):
results.append(f.result())
wall = time.perf_counter() - wall_start
statuses = [r[0] for r in results]
per_req = [r[1] for r in results]
ok_count = statuses.count(200)
return {
"concurrency": n_workers,
"ok": ok_count, "fail": n_workers - ok_count,
"wall_s": round(wall, 3),
"min_req_s": round(min(per_req), 3),
"mean_req_s": round(statistics.mean(per_req), 3),
"max_req_s": round(max(per_req), 3),
}
def test_concurrent_2_health_requests(self, http):
result = self._run_concurrent(2, http)
_perf_results["concurrent_2"] = result
print(f"\n Concurrent (2) — {result}")
assert result["ok"] == 2, f"Expected 2/2 successes, got {result}"
def test_concurrent_4_health_requests(self, http):
result = self._run_concurrent(4, http)
_perf_results["concurrent_4"] = result
print(f"\n Concurrent (4) — {result}")
assert result["ok"] == 4, f"Expected 4/4 successes, got {result}"
def test_concurrent_wall_less_than_serial(self, http):
"""
Wall-clock for N concurrent requests should be less than
N × mean single-request time (i.e., some parallelism is achieved).
"""
single_latency = _timeit(lambda: http.get(f"{BASE_URL}/health", timeout=TIMEOUT))
result = self._run_concurrent(N_CONCURRENT, http)
serial_estimate = single_latency * N_CONCURRENT
# Allow a generous 80 % of serial time as "acceptable parallelism"
assert result["wall_s"] < serial_estimate * 0.95 or result["wall_s"] < 30, (
f"Concurrent wall={result['wall_s']:.2f}s not better than "
f"serial estimate={serial_estimate:.2f}s"
)
|