Spaces:
Running
Running
File size: 8,750 Bytes
8a841b2 | 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 | """
Locust load test file for InferRoute gateway.
Scenarios:
- NormalUser: Mixed blocking + streaming at realistic rates
- BurstUser: High-concurrency burst to stress routing/CB
- CostUser: Cost-optimized requests (forces local backends)
- ChaosUser: Invalid keys, oversized payloads, bad JSON
Run:
locust -f tests/locustfile.py --headless -u 50 -r 5 -t 60s --host http://localhost:8080
locust -f tests/locustfile.py # interactive UI at http://localhost:8089
"""
import json
import random
import time
from locust import HttpUser, task, between, events, constant_throughput
# Test credentials
VALID_KEY = "sk-inferroute-demo"
INVALID_KEY = "sk-invalid-key-chaos"
# Sample prompts of varying complexity
PROMPTS = [
"What is machine learning?",
"Explain the difference between REST and GraphQL APIs.",
"Write a Python function to calculate the nth Fibonacci number recursively.",
"What are the SOLID principles in software engineering? Give brief examples.",
"How does Redis handle cache eviction? Describe the LRU algorithm in detail with a concrete example.",
"Summarize the key differences between SQL and NoSQL databases for a technical audience.",
"Explain CAP theorem and how it applies to distributed system design.",
]
SYSTEM_PROMPT = "You are a helpful AI assistant. Keep answers concise and accurate."
def _chat_body(prompt: str, model: str = "edge/auto", stream: bool = False, **routing_opts) -> dict:
body = {
"model": model,
"messages": [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": prompt},
],
"stream": stream,
"max_output_tokens": 150,
}
if routing_opts:
body["routing"] = routing_opts
return body
class NormalUser(HttpUser):
"""
Simulates a typical application user: mostly blocking requests,
occasional streaming, realistic think time.
"""
wait_time = between(1, 3)
weight = 60 # 60% of users
@task(7)
def blocking_completion(self):
prompt = random.choice(PROMPTS)
with self.client.post(
"/v1/chat/completions",
json=_chat_body(prompt),
headers={"Authorization": f"Bearer {VALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [blocking]",
) as resp:
if resp.status_code == 200:
data = resp.json()
if "choices" not in data:
resp.failure(f"Missing choices in response: {data}")
elif resp.status_code in (429, 503):
resp.success() # expected under load
else:
resp.failure(f"Unexpected status {resp.status_code}: {resp.text[:200]}")
@task(2)
def streaming_completion(self):
prompt = random.choice(PROMPTS)
start_time = time.time()
ttft_ms = None
chunk_count = 0
with self.client.post(
"/v1/chat/completions",
json=_chat_body(prompt, stream=True),
headers={
"Authorization": f"Bearer {VALID_KEY}",
"Accept": "text/event-stream",
},
stream=True,
catch_response=True,
name="/v1/chat/completions [streaming]",
) as resp:
if resp.status_code == 200:
for line in resp.iter_lines():
if line.startswith(b"data: "):
payload = line[6:].strip()
if payload == b"[DONE]":
break
try:
chunk = json.loads(payload)
chunk_count += 1
if ttft_ms is None:
choices = chunk.get("choices", [])
if choices and choices[0].get("delta", {}).get("content"):
ttft_ms = (time.time() - start_time) * 1000.0
except json.JSONDecodeError:
pass
resp.success()
elif resp.status_code in (429, 503):
resp.success()
else:
resp.failure(f"Stream error {resp.status_code}")
@task(1)
def check_health(self):
self.client.get("/healthz", name="/healthz")
@task(1)
def check_routing_status(self):
self.client.get(
"/v1/routing/status",
name="/v1/routing/status",
)
class BurstUser(HttpUser):
"""
High-frequency burst user to stress circuit breakers and fallback logic.
"""
wait_time = between(0.1, 0.5)
weight = 20 # 20% of users
@task(10)
def rapid_blocking(self):
prompt = random.choice(PROMPTS[:3]) # short prompts for speed
with self.client.post(
"/v1/chat/completions",
json=_chat_body(prompt, max_output_tokens=50),
headers={"Authorization": f"Bearer {VALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [burst]",
) as resp:
if resp.status_code in (200, 429, 503):
resp.success()
else:
resp.failure(f"Burst error {resp.status_code}")
class CostUser(HttpUser):
"""
Cost-optimized user: forces local-only routing to minimize API costs.
"""
wait_time = between(2, 5)
weight = 15 # 15% of users
@task(5)
def cost_optimized_request(self):
with self.client.post(
"/v1/chat/completions",
json=_chat_body(
random.choice(PROMPTS),
policy="cost",
allow_local=True,
allow_cloud=False,
),
headers={"Authorization": f"Bearer {VALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [cost-policy]",
) as resp:
if resp.status_code in (200, 429, 502, 503):
resp.success()
else:
resp.failure(f"Cost request error {resp.status_code}")
@task(2)
def cloud_only_request(self):
with self.client.post(
"/v1/chat/completions",
json=_chat_body(
random.choice(PROMPTS),
policy="latency",
allow_local=False,
allow_cloud=True,
),
headers={"Authorization": f"Bearer {VALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [cloud-only]",
) as resp:
if resp.status_code in (200, 429, 502, 503):
resp.success()
else:
resp.failure(f"Cloud request error {resp.status_code}")
class ChaosUser(HttpUser):
"""
Chaos user: invalid auth keys, malformed payloads, oversized prompts.
Verifies that the gateway correctly rejects and rate-limits bad requests.
"""
wait_time = between(3, 8)
weight = 5 # 5% of users
@task(3)
def invalid_api_key(self):
with self.client.post(
"/v1/chat/completions",
json=_chat_body("hello"),
headers={"Authorization": f"Bearer {INVALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [invalid-auth]",
) as resp:
if resp.status_code == 401:
resp.success() # expected
else:
resp.failure(f"Expected 401, got {resp.status_code}")
@task(2)
def malformed_json(self):
with self.client.post(
"/v1/chat/completions",
data="not valid json at all {{{",
headers={
"Authorization": f"Bearer {VALID_KEY}",
"Content-Type": "application/json",
},
catch_response=True,
name="/v1/chat/completions [bad-json]",
) as resp:
if resp.status_code in (400, 422):
resp.success() # expected
else:
resp.failure(f"Expected 400/422, got {resp.status_code}")
@task(1)
def oversized_prompt(self):
huge_prompt = "Explain this concept in extreme detail. " * 500
with self.client.post(
"/v1/chat/completions",
json=_chat_body(huge_prompt),
headers={"Authorization": f"Bearer {VALID_KEY}"},
catch_response=True,
name="/v1/chat/completions [oversized]",
) as resp:
if resp.status_code in (200, 400, 413, 422, 429, 502):
resp.success() # any handled response is acceptable
else:
resp.failure(f"Oversized request unexpected {resp.status_code}")
|