Files
mtusi2026lab3/research/analyze_compare.py
T

104 lines
3.6 KiB
Python

import argparse
import csv
import json
import statistics
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt # noqa: E402
def main():
parser = argparse.ArgumentParser()
parser.add_argument("directory")
directory = Path(parser.parse_args().directory)
meta = json.loads((directory / "run.json").read_text())
assert meta["complete"]
rows = [
json.loads((directory / name).read_text()) for name in meta["cells"]
]
keys = [
"version",
"pg_cpu_limit",
"repeat",
"successful_rps",
"errors",
"p50_ms",
"p95_ms",
"p99_ms",
]
with (directory / "summary.csv").open("w", newline="") as stream:
writer = csv.DictWriter(stream, fieldnames=keys)
writer.writeheader()
writer.writerows({k: r[k] for k in keys} for r in rows)
lines = [
"# Насыщение PostgreSQL и устранение повторного ACL-запроса",
"",
meta["profile"],
"",
"Медианы трёх чередующихся повторов.",
"",
"| PG CPU | Версия | RPS | p95 ms | Ошибки |",
"|---|---|---:|---:|---:|",
]
fig, axes = plt.subplots(1, 2, figsize=(9, 4), constrained_layout=True)
for ax, cpu in zip(axes, [0.1, 1.0], strict=True):
values = []
for version in ["before", "after"]:
group = [
r
for r in rows
if r["version"] == version and r["pg_cpu_limit"] == cpu
]
median = statistics.median(r["successful_rps"] for r in group)
values.append(median)
lines.append(
f"| {cpu} | {version} | {median:.1f} | {statistics.median(r['p95_ms'] for r in group):.2f} | {sum(r['errors'] for r in group)} |"
)
ax.bar(["before", "after"], values, color=["#d48748", "#3c88b5"])
ax.set(
title=f"PostgreSQL {cpu} CPU", ylabel="Successful HTTP requests/s"
)
ax.grid(axis="y", alpha=0.2)
fig.savefig(directory / "comparison.png", dpi=160)
plt.close(fig)
lines += [
"",
"Before revision: `" + meta["images"]["before"]["revision"] + "`.",
"After revision: `" + meta["images"]["after"]["revision"] + "`.",
"",
"При PG1CPU выигрыш RPS наблюдается во всех трёх парах; при PG0.1CPU один after хуже before. Короткие окна и большой разброс не задают production SLA.",
"",
f"Всего успешных HTTP: {sum(r['success'] for r in rows)}; ошибок: {sum(r['errors'] for r in rows)}.",
"",
"Доли CPU-периодов PG с throttling (по каждому прогону):",
]
for cpu in [0.1, 1.0]:
ratios = []
for r in rows:
if r["pg_cpu_limit"] != cpu:
continue
def parse(text):
return {
s.split()[0]: int(s.split()[1]) for s in text.splitlines()
}
before, after = parse(r["pg_cpu_before"]), parse(r["pg_cpu_after"])
periods = after["nr_periods"] - before["nr_periods"]
ratios.append(
round(
(after["nr_throttled"] - before["nr_throttled"])
/ max(1, periods),
3,
)
)
lines.append(f"- PG{cpu}CPU: {ratios}")
(directory / "analysis.md").write_text("\n".join(lines) + "\n")
if __name__ == "__main__":
main()