Files
mtusi2026lab3/scripts/experiment.py

481 lines
18 KiB
Python

#!/usr/bin/env python3
"""Нагрузка через LB во время отказа/замены API-реплик; метрики и проверка данных.
failover — для каждой API-реплики по очереди: нагрузка → остановка этой реплики
(docker stop = SIGTERM, затем docker kill = SIGKILL) → восстановление
(scripts/restore.sh) → нагрузка после восстановления.
rolling — нагрузка во время scripts/rolling-update.sh (поочерёдная замена реплик).
Клиент ведёт себя как настоящий: неподтверждённый POST повторяет с тем же
Idempotency-Key и телом. После прогона вся история читается заново: каждое
подтверждённое сообщение на месте и не изменено, дублей нет.
Python 3.9+, стандартная библиотека; tokens не печатаются и не сохраняются.
"""
from __future__ import annotations
import argparse
import json
import math
import os
import subprocess
import sys
import threading
import time
import urllib.error
import urllib.request
import uuid
from dataclasses import dataclass
from pathlib import Path
from urllib.parse import quote
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "tests"))
from client import Client # noqa: E402
REQUEST_TIMEOUT_SEC = 5
RETRY_PAUSE_SEC = 0.05
STATS_FORMAT = "{{.Name}}\t{{.CPUPerc}}\t{{.MemUsage}}"
@dataclass
class Attempt:
"""Одна HTTP-попытка клиента; время — секунды от начала прогона"""
op: str
started: float
finished: float
status: int | None
error: str | None
instance: str | None
retry: bool
@property
def ok(self) -> bool:
return self.status in (200, 201)
class Recorder:
def __init__(self) -> None:
self.started = time.monotonic()
self.lock = threading.Lock()
self.attempts: list[Attempt] = []
self.confirmed: list[dict] = []
self.events: dict[str, float] = {}
def now(self) -> float:
return time.monotonic() - self.started
def mark(self, name: str) -> None:
self.events[name] = round(self.now(), 3)
def add(self, attempt: Attempt, message: dict | None = None) -> None:
with self.lock:
self.attempts.append(attempt)
if message is not None:
self.confirmed.append(message)
def request(
base: str, method: str, path: str, token: str, body: dict | None, key: str | None
) -> tuple[int | None, str | None, str | None, dict | None]:
headers = {"Authorization": "Bearer " + token, "Accept": "application/json"}
data = None
if body is not None:
data = json.dumps(body, ensure_ascii=False).encode()
headers["Content-Type"] = "application/json"
if key is not None:
headers["Idempotency-Key"] = key
req = urllib.request.Request(base + path, data=data, headers=headers, method=method)
try:
with urllib.request.urlopen(req, timeout=REQUEST_TIMEOUT_SEC) as response:
payload = json.loads(response.read())
return response.status, None, response.headers.get("X-Instance-Id"), payload
except urllib.error.HTTPError as error:
error.read()
return error.code, None, error.headers.get("X-Instance-Id"), None
except OSError as error:
return None, type(error).__name__, None, None
def worker(
base: str,
token: str,
chat_id: str,
worker_id: int,
recorder: Recorder,
stop: threading.Event,
) -> None:
"""Запись с повтором по тому же ключу до подтверждения, затем чтение"""
path = f"/api/v1/chats/{chat_id}/messages"
index = 0
while not stop.is_set():
key = uuid.uuid4().hex
body = {"text": f"exp-{worker_id}-{index}-{key[:8]}"}
retry = False
while not stop.is_set():
started = recorder.now()
status, error, instance, payload = request(
base, "POST", path, token, body, key
)
attempt = Attempt(
"write", started, recorder.now(), status, error, instance, retry
)
recorder.add(attempt, payload if status == 201 else None)
if status == 201:
break
retry = True
time.sleep(RETRY_PAUSE_SEC)
started = recorder.now()
status, error, instance, _ = request(
base, "GET", path + "?limit=1", token, None, None
)
recorder.add(
Attempt("read", started, recorder.now(), status, error, instance, False)
)
index += 1
def percentile(values: list[float], p: float) -> float | None:
if not values:
return None
ordered = sorted(values)
return round(1000 * ordered[max(0, math.ceil(p * len(ordered)) - 1)], 1)
def phase_stats(attempts: list[Attempt], start: float, end: float) -> dict:
inside = [a for a in attempts if start <= a.finished < end]
duration = max(end - start, 1e-9)
result: dict = {"seconds": round(duration, 3)}
for op in ("write", "read"):
ops = [a for a in inside if a.op == op]
ok = [a for a in ops if a.ok]
result[op] = {
"ok": len(ok),
"errors": len(ops) - len(ok),
"ok_rps": round(len(ok) / duration, 1),
"latency_ms": {
"p50": percentile([a.finished - a.started for a in ok], 0.50),
"p95": percentile([a.finished - a.started for a in ok], 0.95),
"p99": percentile([a.finished - a.started for a in ok], 0.99),
},
}
return result
def recovery_stats(attempts: list[Attempt], action_start: float, end: float) -> dict:
"""
Время восстановления записи: от первой ошибки после воздействия до первой
успешной записи после последней ошибки; без ошибок — 0. Отдельно —
наибольший разрыв между подтверждениями записи (задержка без ошибок).
"""
window = sorted(
(a for a in attempts if a.op == "write" and a.finished >= action_start - 1),
key=lambda a: a.finished,
)
errors = [a for a in window if not a.ok and a.finished >= action_start]
kinds: dict[str, int] = {}
for a in errors:
kind = a.error or f"HTTP {a.status}"
kinds[kind] = kinds.get(kind, 0) + 1
if errors:
last_error = errors[-1].finished
first_ok_after = next(
(a.finished for a in window if a.ok and a.finished > last_error), end
)
recovery = first_ok_after - errors[0].finished
else:
recovery = 0.0
oks = [a.finished for a in window if a.ok and a.finished <= end]
gap = max((b - a for a, b in zip(oks, oks[1:])), default=None)
return {
"write_errors": len(errors),
"error_kinds": kinds,
"write_recovery_seconds": round(recovery, 3),
"max_write_gap_seconds": None if gap is None else round(gap, 3),
}
def timeline(attempts: list[Attempt], end: float) -> list[dict]:
rows = []
for second in range(int(end) + 1):
inside = [a for a in attempts if second <= a.finished < second + 1]
instances: dict[str, int] = {}
for a in inside:
if a.ok and a.instance:
instances[a.instance] = instances.get(a.instance, 0) + 1
rows.append(
{
"t": second,
"ok": sum(a.ok for a in inside),
"errors": sum(not a.ok for a in inside),
"instances": instances,
}
)
return rows
def read_history(client: Client, actor: dict, chat_id: str) -> list[dict]:
path = f"/api/v1/chats/{chat_id}/messages?limit=100"
items: list[dict] = []
while True:
page = client.call("GET", path, actor=actor, schema="MessagePage")["json"]
items.extend(page["items"])
if page["next_cursor"] is None:
return items
path = f"/api/v1/chats/{chat_id}/messages?limit=100&cursor=" + quote(
page["next_cursor"], safe=""
)
def integrity(client: Client, actor: dict, chat_id: str, recorder: Recorder) -> dict:
history = read_history(client, actor, chat_id)
by_id = {m["id"]: m for m in history}
texts = [m["text"] for m in history]
confirmed = {m["id"]: m for m in recorder.confirmed}
return {
"messages_in_history": len(history),
"confirmed_unique": len(confirmed),
"confirmed_missing_or_changed": sum(
by_id.get(i) != m for i, m in confirmed.items()
),
"duplicate_ids": len(history) - len(by_id),
"duplicate_texts": len(texts) - len(set(texts)),
"retried_writes_confirmed_as_replay": len(recorder.confirmed) - len(confirmed),
}
def run(command: list[str], env: dict | None = None) -> str:
result = subprocess.run(
command, cwd=ROOT, env=env, check=True, capture_output=True, text=True
)
return result.stdout.strip()
def api_containers() -> list[str]:
ids = run(["docker", "compose", "--env-file", ".env", "ps", "-q", "api"]).split()
names = [run(["docker", "inspect", "-f", "{{.Name}}", i]).lstrip("/") for i in ids]
return sorted(names)
def instance_of(container: str) -> str:
return run(["docker", "exec", container, "cat", "/tmp/instance-id"])
def container_stats() -> list[str]:
names = ["messenger-lab3-nginx-1", "messenger-lab3-postgres-1", *api_containers()]
output = run(["docker", "stats", "--no-stream", "--format", STATS_FORMAT, *names])
return output.splitlines()
def scenario(
args: argparse.Namespace,
label: str,
action: list[str],
restore: list[str] | None,
target: str | None = None,
env: dict | None = None,
) -> dict:
client = Client(args.base_url, timeout=REQUEST_TIMEOUT_SEC)
alice, _, _, chat = client.bootstrap()
base = client.base
recorder = Recorder()
stop = threading.Event()
threads = [
threading.Thread(
target=worker,
args=(base, alice["token"], chat["id"], n, recorder, stop),
daemon=True,
)
for n in range(args.workers)
]
for thread in threads:
thread.start()
time.sleep(args.before / 2)
stats = container_stats()
time.sleep(args.before / 2)
instances_before = sorted(
{a.instance for a in recorder.attempts if a.ok and a.instance}
)
recorder.mark("action_start")
action_output = run(action, env)
recorder.mark("action_end")
restore_output = ""
if restore is not None:
time.sleep(args.outage)
recorder.mark("restore_start")
restore_output = run(restore)
recorder.mark("restore_end")
time.sleep(args.after)
stop.set()
for thread in threads:
thread.join(timeout=REQUEST_TIMEOUT_SEC + 1)
recorder.mark("end")
events = recorder.events
end = events["end"]
after_start = events.get("restore_end", events["action_end"])
attempts = recorder.attempts
seen_after = [
a for a in attempts if a.ok and a.finished >= after_start and a.instance
]
back_in_rotation = next(
(
round(a.finished - after_start, 3)
for a in sorted(seen_after, key=lambda a: a.finished)
if a.instance not in instances_before
),
None,
)
served = sorted(
(a for a in attempts if a.ok and a.instance), key=lambda a: a.finished
)
instances_seen: dict[str, list[float]] = {}
for a in served:
offset = round(a.finished - events["action_start"], 3)
instances_seen.setdefault(a.instance, [offset, offset])[1] = offset
phases = {
"before": phase_stats(attempts, 0, events["action_start"]),
"during": phase_stats(attempts, events["action_start"], after_start),
"after": phase_stats(attempts, after_start, end),
}
target_after_stop = None
if target is not None:
target_after_stop = sum(
1
for a in attempts
if a.ok
and a.instance == target
and a.finished > events["action_end"]
and a.finished < events.get("restore_start", end)
)
return {
"scenario": label,
"target_instance": target,
"workers": args.workers,
"events_seconds": events,
"instances_before": instances_before,
"instances_after": sorted({a.instance for a in seen_after}),
"target_responses_while_stopped": target_after_stop,
"restored_replica_in_rotation_seconds": back_in_rotation
if restore is not None
else None,
# alias → [первый, последний] успешный ответ, секунды от начала воздействия
"instances_seen_from_action": instances_seen,
"phases": phases,
"recovery": recovery_stats(attempts, events["action_start"], end),
"integrity": integrity(client, alice, chat["id"], recorder),
"container_stats_before_action": stats,
"action_output": action_output.splitlines(),
"restore_output_tail": restore_output.splitlines()[-2:],
"timeline": timeline(attempts, end),
}
def failover(args: argparse.Namespace) -> list[dict]:
results = []
for signal in args.signals:
for container in api_containers():
target = instance_of(container)
if signal == "SIGTERM":
action = ["docker", "stop", "--timeout", "10", container]
else:
action = ["docker", "kill", "--signal", "SIGKILL", container]
label = f"{signal} {container}"
print(f"== {label} (instance {target})", flush=True)
result = scenario(
args, label, action, ["./scripts/restore.sh"], target=target
)
print(json.dumps(brief(result), ensure_ascii=False), flush=True)
results.append(result)
return results
def rolling(args: argparse.Namespace) -> list[dict]:
env = dict(os.environ)
env.setdefault("APP_REVISION", f"exp-{int(time.time())}")
print(f"== rolling update to revision {env['APP_REVISION']}", flush=True)
result = scenario(
args, "rolling update", ["./scripts/rolling-update.sh"], None, env=env
)
result["revisions_after"] = run(["./scripts/instances.sh"]).splitlines()
print(json.dumps(brief(result), ensure_ascii=False), flush=True)
return [result]
def brief(result: dict) -> dict:
"""Краткая строка прогона для терминала; полная версия — в --output"""
return {
"scenario": result["scenario"],
"rps_ok_write": {k: v["write"]["ok_rps"] for k, v in result["phases"].items()},
"rps_ok_read": {k: v["read"]["ok_rps"] for k, v in result["phases"].items()},
"errors": {
k: v["write"]["errors"] + v["read"]["errors"]
for k, v in result["phases"].items()
},
"recovery": result["recovery"],
"target_responses_while_stopped": result["target_responses_while_stopped"],
"restored_replica_in_rotation_seconds": result[
"restored_replica_in_rotation_seconds"
],
"instances_seen_from_action": result["instances_seen_from_action"],
"integrity": result["integrity"],
}
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("mode", choices=["failover", "rolling"])
parser.add_argument("--base-url", default=None)
parser.add_argument("--workers", type=int, default=4)
parser.add_argument("--before", type=float, default=8, help="нагрузка до отказа")
parser.add_argument("--outage", type=float, default=10, help="реплика остановлена")
parser.add_argument("--after", type=float, default=8, help="после восстановления")
parser.add_argument(
"--signals",
nargs="+",
default=["SIGTERM", "SIGKILL"],
choices=["SIGTERM", "SIGKILL"],
)
parser.add_argument("--output", default=None, help="JSON с полными результатами")
args = parser.parse_args()
started = time.strftime("%Y-%m-%dT%H:%M:%S%z")
results = failover(args) if args.mode == "failover" else rolling(args)
report = {
"mode": args.mode,
"started": started,
"profile": {
"workers": args.workers,
"loop": "closed: POST message (retry same key until 201) + GET limit=1",
"client_timeout_seconds": REQUEST_TIMEOUT_SEC,
"phases_seconds": {
"before": args.before,
"outage": args.outage,
"after": args.after,
},
},
"runs": results,
}
if args.output:
destination = ROOT / args.output
destination.parent.mkdir(parents=True, exist_ok=True)
destination.write_text(json.dumps(report, ensure_ascii=False, indent=2) + "\n")
print(f"saved: {args.output}")
failed = [
r["scenario"]
for r in results
if r["integrity"]["confirmed_missing_or_changed"]
or r["integrity"]["duplicate_ids"]
or r["integrity"]["duplicate_texts"]
or r["target_responses_while_stopped"]
]
if failed:
raise SystemExit(f"FAIL: потеря/дубль или отказ не изолирован: {failed}")
print(f"PASS: прогонов {len(results)}, подтверждённые сообщения целы, дублей нет")
if __name__ == "__main__":
main()