Files
mtusi2026lab4-fork/student-tests/memory_experiment.py
T
2026-11-01 01:18:40 +03:00

70 lines
3.6 KiB
Python

#!/usr/bin/env python3
"""Peak-RAM experiment: N concurrent uploads, reads VmHWM (peak RSS) of every API replica process.
Docker mode (default): `docker compose exec -T api` is not enough to pick each replica, so we iterate
`docker compose ps -q api` and `docker exec <id> cat /proc/1/status`. Local mode: --pids 123,456.
Run on a fresh stack (restart api first) so VmHWM starts from the idle baseline.
"""
import argparse, json, re, subprocess, sys, time
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
sys.path[:0] = [str(Path(__file__).resolve().parents[1] / "tests"), str(Path(__file__).parent)]
from client import Client, check, run
import imagegen
def reset_peak(args):
"""Writing 5 to /proc/PID/clear_refs resets VmHWM to the current RSS (own process, no privileges)."""
if args.pids:
for pid in args.pids.split(","):
Path(f"/proc/{pid}/clear_refs").write_text("5")
else:
for cid in subprocess.check_output(["docker", "compose", "ps", "-q", "api"], text=True).split():
subprocess.check_call(["docker", "exec", cid, "sh", "-c", "echo 5 > /proc/1/clear_refs"])
def read_status(args):
out = {}
if args.pids:
for pid in args.pids.split(","):
out[pid] = Path(f"/proc/{pid}/status").read_text()
else:
for cid in subprocess.check_output(["docker", "compose", "ps", "-q", "api"], text=True).split():
out[cid[:12]] = subprocess.check_output(["docker", "exec", cid, "cat", "/proc/1/status"], text=True)
res = {}
for k, t in out.items():
g = lambda n: int(re.search(n + r":\s+(\d+) kB", t).group(1)) // 1024
res[k] = {"VmRSS_MiB": g("VmRSS"), "VmHWM_MiB": g("VmHWM")}
return res
def main():
p = argparse.ArgumentParser()
p.add_argument("--base-url", default=None); p.add_argument("--ca-file", default=None)
p.add_argument("--concurrency", type=int, default=10); p.add_argument("--pids", default="")
p.add_argument("--profile", choices=["big-bytes", "max-pixels"], default="big-bytes")
p.add_argument("--output", default=None)
args = p.parse_args()
check(args.concurrency >= 10, "the lab requires >= 10 concurrent uploads")
c = Client(args.base_url, args.ca_file, timeout=120)
alice, bob, eve, chat = c.bootstrap()
if args.profile == "big-bytes": # ~9.6 MiB incompressible RGB PNG: stresses body streaming
data = imagegen.png(1830, 1830, color_type=2, noise=True, level=0)
else: # exactly 40 MP, tiny file: stresses the decoder
data = imagegen.png(8000, 5000, bit_depth=8, color_type=0)
check(len(data) <= 10 * 1024 * 1024, "payload above the limit")
reset_peak(args)
before = read_status(args)
def one(_):
t = time.monotonic(); r = c.call("POST", f"/api/v1/chats/{chat['id']}/attachments?filename=m.png", 201, actor=alice, raw=data, headers={"Content-Type": "image/png"}, schema="Attachment")
return time.monotonic() - t
t0 = time.monotonic()
with ThreadPoolExecutor(args.concurrency) as pool:
lat = sorted(pool.map(one, range(args.concurrency)))
after = read_status(args)
rep = {"profile": args.profile, "concurrency": args.concurrency, "payload_bytes": len(data), "wall_seconds": round(time.monotonic() - t0, 2),
"latency_s": {"min": round(lat[0], 2), "max": round(lat[-1], 2)}, "before": before, "after_peak": after,
"naive_estimate_MiB": round(args.concurrency * len(data) / 2**20 * 2)}
print(json.dumps(rep, indent=2))
if args.output: Path(args.output).write_text(json.dumps(rep, indent=2) + "\n")
if __name__ == "__main__":
run(main)