Files
mtusi2026lab4-fork/scripts/check_grade4.py
2026-11-04 21:27:54 +03:00

356 lines
14 KiB
Python

"""Проверяет инфраструктуру, SIGTERM и обновление реального стенда."""
import asyncio
from collections import Counter
from concurrent.futures import ThreadPoolExecutor
import hashlib
import json
from pathlib import Path
import subprocess
from threading import Event
import time
from uuid import uuid4
import asyncpg
from scripts.deployment import ComposeProject, ROOT, deployment_lock
from scripts.failover_all import exercise, failure_kind, history
from scripts.rolling_update import update
from tests.client import Client, Failure, arguments, check, run
PROBE = """import json, sys, urllib.request, urllib.error
payload = json.load(sys.stdin)
results = []
for item in payload:
data = json.dumps(item['body']).encode() if 'body' in item else None
request = urllib.request.Request(item['url'], data=data, headers=item.get('headers', {}), method=item.get('method', 'GET'))
try:
response = urllib.request.urlopen(request, timeout=7)
except urllib.error.HTTPError as error:
response = error
with response:
results.append({'status': response.status, 'body': json.loads(response.read())})
print(json.dumps(results))
"""
def probe(project, runner, requests):
"""Обращается к внутреннему API, передавая токен только через stdin."""
result = subprocess.run(
["docker", "exec", "-i", runner, "python", "-c", PROBE],
input=json.dumps(requests),
text=True,
capture_output=True,
timeout=10,
)
check(result.returncode == 0, "internal HTTP probe failed")
return json.loads(result.stdout)
def audit(project):
"""Проверяет фактические лимиты, порты и область доступа к секретам."""
allowed = {
"postgres": {"db_password"},
"redis": {"redis_password"},
"migrate": {"db_password"},
"api": {"db_password", "redis_password", "s3_access_key_id", "s3_secret_access_key"},
"minio": {"minio_root_user", "minio_root_password"},
"minio-bootstrap": {"minio_root_user", "minio_root_password", "s3_access_key_id", "s3_secret_access_key"},
"cleanup": {"db_password", "s3_access_key_id", "s3_secret_access_key"},
"nginx": set(),
}
ids = project.run("ps", "--all", "--quiet").splitlines()
containers = json.loads(project.docker("inspect", *ids))
for item in containers:
service = item["Config"]["Labels"]["com.docker.compose.service"]
check(service in allowed, "unexpected service")
host = item["HostConfig"]
check(
host["Memory"] > 0 and host["NanoCpus"] > 0,
service + ": missing CPU/RAM limit",
)
env_names = {entry.split("=", 1)[0] for entry in item["Config"]["Env"]}
check(
not env_names
& {
"DB_PASSWORD",
"REDIS_PASSWORD",
"POSTGRES_PASSWORD",
"REDISCLI_AUTH",
"DATABASE_URL",
"MINIO_ROOT_USER", "MINIO_ROOT_PASSWORD", "S3_ACCESS_KEY_ID", "S3_SECRET_ACCESS_KEY",
},
service + ": password in environment",
)
secrets = {
Path(mount["Destination"]).name
for mount in item["Mounts"]
if mount["Destination"].startswith("/run/secrets/")
}
check(secrets == allowed[service], service + ": incorrect secret scope")
if service in {"api", "minio"}:
check(not host["PortBindings"], service + " publishes ports")
for mount in item["Mounts"]:
if mount["Destination"].startswith("/run/secrets/"):
check(not mount["RW"], "writable secret mount")
check(
"container_name" not in project.config["services"][service],
"fixed container name",
)
check(
len(containers) == len(allowed) - 1 + len(project.replicas()),
"unexpected container count",
)
print(
"PASS infrastructure: CPU/RAM, private API ports, scoped read-only file secrets, no environment passwords",
flush=True,
)
async def graceful(project, base_url):
"""Удерживает реальный POST в БД и завершает его после SIGTERM."""
replicas = project.replicas()
target, runner = replicas[:2]
client = Client(base_url)
actor, _, _, chat = await asyncio.to_thread(client.bootstrap)
path = f"/api/v1/chats/{chat['id']}/messages"
key = uuid4().hex
raw = json.dumps([actor["id"], "POST", path, key]).encode()
lock_id = int.from_bytes(hashlib.sha256(raw).digest()[:8], signed=True)
database = project.config["services"]["postgres"]
env = database["environment"]
conn = await asyncpg.connect(
host="127.0.0.1",
port=int(database["ports"][0]["published"]),
database=env["POSTGRES_DB"],
user=env["POSTGRES_USER"],
password=(ROOT / "secrets/db_password").read_text().strip(),
timeout=3,
)
network = project.config['networks']['default']['name']
address_template = '{{with index .NetworkSettings.Networks ' + json.dumps(network) + '}}{{.IPAddress}}{{end}}'
address = project.docker(
"inspect",
"--format",
address_template,
target.id,
)
check(bool(address), 'API has no address in the application network')
port = project.config["services"]["api"]["environment"]["APP_PORT"]
direct = f"http://{address}:{port}"
started = time.monotonic()
task = None
try:
async with conn.transaction():
await conn.execute("SELECT pg_advisory_xact_lock($1)", lock_id)
task = asyncio.create_task(
asyncio.to_thread(
probe,
project,
runner.id,
[
{
"url": direct + path,
"method": "POST",
"body": {"text": "graceful-" + key},
"headers": {
"Authorization": "Bearer " + actor["token"],
"Idempotency-Key": key,
"Content-Type": "application/json",
},
}
],
)
)
deadline = time.monotonic() + 3
while not await conn.fetchval(
"SELECT EXISTS(SELECT 1 FROM pg_stat_activity WHERE wait_event = 'advisory' AND datname = current_database())"
):
check(time.monotonic() < deadline, "POST did not reach database lock")
await asyncio.sleep(0.02)
started = time.monotonic()
await asyncio.to_thread(
project.docker, "kill", "--signal", "SIGTERM", target.id
)
results = await asyncio.to_thread(
probe,
project,
runner.id,
[
{"url": direct + "/health/ready"},
{"url": direct + "/health/live"},
{"url": direct + "/api/v1/users/me"},
],
)
check(
[item["status"] for item in results] == [503, 200, 503],
"incorrect drain readiness/liveness/new work",
)
await asyncio.sleep(max(0, 1.2 - (time.monotonic() - started)))
result = (await task)[0]
check(result["status"] == 201, "in-flight POST failed during SIGTERM")
await asyncio.to_thread(project.docker, "wait", target.id, timeout=10)
state = json.loads(
project.docker("inspect", "--format", "{{json .State}}", target.id)
)
check(
not state["OOMKilled"] and state["ExitCode"] != 137,
"forced shutdown instead of graceful completion",
)
check(time.monotonic() - started < 8, "shutdown exceeded Docker grace budget")
items = await asyncio.to_thread(history, client, actor, chat)
check(
result["body"]["id"] in {item["id"] for item in items},
"acknowledged in-flight message lost",
)
print(
f"PASS SIGTERM: readiness 503, live 200, new work 503, in-flight POST 201; exit in {time.monotonic() - started:.3f}s",
flush=True,
)
finally:
await conn.close()
if task:
await asyncio.gather(task, return_exceptions=True)
if any(item.id == target.id and item.running for item in project.replicas()):
await asyncio.to_thread(project.docker, "stop", target.id)
await asyncio.to_thread(project.docker, "start", target.id)
await asyncio.to_thread(project.wait_healthy, {item.id for item in replicas})
await asyncio.to_thread(project.wait_routed, target.instance_id, base_url)
class ObservedProject(ComposeProject):
"""Подсчитывает миграции и проверяет готовность перед заменой."""
def __init__(self):
self.migrations = 0
self.replacements = 0
self.replacement_started = []
super().__init__()
def run(self, *args, **kwargs):
if args[:6] == ("run", "--rm", "--no-deps", "--pull", "never", "migrate"):
self.migrations += 1
return super().run(*args, **kwargs)
def docker(self, *args, **kwargs):
if args[0] == "stop":
healthy = [
item
for item in self.replicas()
if item.id != args[1] and item.running and item.health == "healthy"
]
check(len(healthy) >= 2, "rolling update reduces healthy capacity")
self.replacements += 1
self.replacement_started.append(time.monotonic())
return super().docker(*args, **kwargs)
def rolling(base_url):
"""Заменяет образ под постоянной записью с проверкой сохранности."""
project = ObservedProject()
before = project.replicas()
client = Client(base_url)
actor, _, _, chat = client.bootstrap()
stop = Event()
results, errors = [], []
def worker():
local = Client(base_url, timeout=2)
key = uuid4().hex
while not stop.is_set():
started = time.monotonic()
try:
result = local.message(actor, chat, "rolling-" + key, key=key)
results.append(
(started, time.monotonic(), result["json"]["id"], result["elapsed"])
)
key = uuid4().hex
except Failure as error:
errors.append(failure_kind(error))
stop.wait(0.08)
with ThreadPoolExecutor(max_workers=2) as pool:
jobs = [pool.submit(worker) for _ in range(2)]
try:
time.sleep(1)
started = time.monotonic()
update(project, base_url, revision="verification-" + uuid4().hex[:12])
finished = time.monotonic()
time.sleep(1)
finally:
stop.set()
for job in jobs:
job.result(timeout=5)
after = project.replicas()
check(
project.migrations == 1 and project.replacements == len(before),
"migration/replacement count incorrect",
)
check(
{item.image for item in before}.isdisjoint({item.image for item in after}),
"new image was not deployed",
)
check(
{item.id for item in before}.isdisjoint({item.id for item in after}),
"old containers remain",
)
writes = [row for row in results if started <= row[0] <= finished]
check(bool(writes), "no writes during update")
check(
any(project.replacement_started[0] <= row[0] <= finished for row in writes),
"no writes during container replacement",
)
items = history(client, actor, chat)
check(
{row[2] for row in results} <= {item["id"] for item in items},
"rolling update lost confirmed messages",
)
check(
len(items) == len({item["text"] for item in items}),
"rolling update created duplicate messages",
)
latencies = sorted(row[3] for row in writes)
print(
json.dumps(
{
"scenario": "rolling update",
"replacements": project.replacements,
"migration_runs": project.migrations,
"successful_writes": len(writes),
"failed_attempts": len(errors),
"errors_by_kind": dict(Counter(errors)),
"elapsed_seconds": round(finished - started, 3),
"successful_rps": round(len(writes) / (finished - started), 3),
"p95_seconds": round(
latencies[min(len(latencies) - 1, int(len(latencies) * 0.95))], 3
),
}
),
flush=True,
)
def main():
"""Выполняет проверки только на отдельном тестовом проекте."""
def configure(parser):
parser.add_argument(
"--phase", choices=("all", "update-and-failover"), default="all"
)
args = arguments("Grade 4 integration checks", configure)
project = ComposeProject()
check(project.name.endswith("-test"), "use isolated test Compose project")
with deployment_lock(project):
audit(project)
if args.phase == "all":
asyncio.run(graceful(project, args.base_url))
rolling(args.base_url)
args.duration, args.concurrency = 10, 2
exercise(project, args)
audit(project)
if __name__ == "__main__":
run(main)