Files
mtusi2026lab2/tools/plot_track1.py
T

346 lines
14 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import argparse
import collections
import csv
import statistics
from pathlib import Path
import matplotlib.pyplot as plt
ORDER = ("1x1", "2x1", "4x1", "8x1", "2x2", "1x4")
PALETTE = ("#2a78d6", "#eb6834", "#1baf7a", "#eda100", "#e87ba4", "#008300", "#6250d6", "#e34948")
MARKERS = ("o", "s", "^", "D", "v", "P", "X", "*")
USL_POINTS = 3
MIN_REPEATS = 3
COMPONENTS = (("api", "API"), ("postgres", "PostgreSQL"), ("lb", "HAProxy"), ("k6", "k6"))
INK, MUTED, GRID, SURFACE = "#0b0b0b", "#52514e", "#e4e3df", "#fcfcfb"
SUMMARY = (
"series variant config rates capacity_rate capacity_ok_rps capacity_p95_ms capacity_p99_ms "
"peak_rate peak_ok_rps peak_ok_min peak_ok_max peak_p95_ms peak_failed_ratio peak_dropped "
"cpu_api cpu_api_limit cpu_postgres cpu_lb cpu_k6 pg_active_avg pg_lock_avg pg_idle_tx_avg "
"pg_idle_avg pool_total pg_backends_max lb_qcur_max"
).split()
plt.rcParams.update(
{
"figure.facecolor": SURFACE,
"axes.facecolor": SURFACE,
"axes.edgecolor": MUTED,
"axes.labelcolor": MUTED,
"axes.titlecolor": INK,
"axes.titleweight": "bold",
"axes.grid": True,
"grid.color": GRID,
"grid.linewidth": 0.8,
"axes.spines.top": False,
"axes.spines.right": False,
"xtick.color": MUTED,
"ytick.color": MUTED,
"legend.frameon": False,
"lines.linewidth": 2,
"lines.markersize": 7,
"font.size": 10,
}
)
VARIANTS = {
"before": {"color": MUTED, "marker": "o"},
"after": {"color": PALETTE[1], "marker": "s"},
}
def style(name):
index = ORDER.index(name) if name in ORDER else len(ORDER)
return {"color": PALETTE[index], "marker": MARKERS[index]}
def load(paths):
groups = collections.defaultdict(lambda: collections.defaultdict(list))
for path in paths:
with Path(path).open() as handle:
for row in csv.DictReader(line for line in handle if not line.startswith("#")):
if row["disturbed"] == "True":
raise ValueError(f"disturbed run in accepted sample: {path}")
key = (Path(path).parent.name, row["variant"], row["config"])
groups[key][int(row["rate"])].append(row)
for key, rates in groups.items():
for rate, rows in rates.items():
if len(rows) < MIN_REPEATS or len({row["repeat"] for row in rows}) != len(rows):
raise ValueError(f"need at least three distinct repeats: {key}, {rate}")
return groups
def med(rows, field):
values = [float(r[field]) for r in rows]
middle = statistics.median(values)
return middle, middle - min(values), max(values) - middle
def label(key):
return key[2] if key[1] == "base" else f"{key[2]} {key[1]}"
def summarize(key, rates):
series, variant, config = key
sat = {
rate: 2 * sum(r["saturated"] == "True" for r in rows) > len(rows)
for rate, rows in rates.items()
}
capacity = max((rate for rate, bad in sat.items() if not bad), default=None)
peak = max(rates, key=lambda rate: med(rates[rate], "ok_rps")[0])
top, first = rates[peak], rates[peak][0]
replicas, workers, pool = (int(first[f]) for f in ("replicas", "workers", "pool"))
row = {
"series": series,
"variant": variant,
"config": config,
"rates": " ".join(str(r) for r in sorted(rates)),
"peak_rate": peak,
"peak_ok_rps": med(top, "ok_rps")[0],
"peak_ok_min": min(float(r["ok_rps"]) for r in top),
"peak_ok_max": max(float(r["ok_rps"]) for r in top),
"peak_p95_ms": med(top, "p95_ms")[0],
"peak_failed_ratio": med(top, "failed_ratio")[0],
"peak_dropped": med(top, "dropped_iterations")[0],
"cpu_api_limit": round(replicas * float(first["cpus"]) * 100),
"pool_total": replicas * workers * pool,
}
for field in (
"cpu_api cpu_postgres cpu_lb cpu_k6 pg_active_avg pg_lock_avg pg_idle_tx_avg "
"pg_idle_avg pg_backends_max lb_qcur_max"
).split():
row[field] = round(med(top, field)[0], 2)
if capacity is not None:
for field, name in (("ok_rps", "ok_rps"), ("p95_ms", "p95_ms"), ("p99_ms", "p99_ms")):
row[f"capacity_{name}"] = med(rates[capacity], field)[0]
row["capacity_rate"] = capacity
return row
def direct_label(axis, x, y, text):
axis.annotate(
text, (x, y), xytext=(6, 0), textcoords="offset points", va="center", color=INK, fontsize=9
)
def lines(axis, keys, groups, field):
for key in keys:
rates = sorted(groups[key])
values = [med(groups[key][rate], field) for rate in rates]
look = VARIANTS.get(key[1]) or style(key[2])
axis.errorbar(
rates,
[v[0] for v in values],
yerr=[[v[1] for v in values], [v[2] for v in values]],
capsize=3,
label=label(key),
markeredgecolor=SURFACE,
markeredgewidth=1.5,
**look,
)
direct_label(axis, rates[-1], values[-1][0], label(key))
def throughput(series, keys, groups, out):
figure, axes = plt.subplots(1, 3, figsize=(17, 4.8))
top = max(rate for key in keys for rate in groups[key])
axes[0].plot(
[0, top], [0, top], color=MUTED, linestyle="--", linewidth=1, label="RPS = нагрузка"
)
lines(axes[0], keys, groups, "ok_rps")
axes[0].set_title("Успешные RPS")
for axis, field, title in ((axes[1], "p95_ms", "p95, мс"), (axes[2], "p99_ms", "p99, мс")):
lines(axis, keys, groups, field)
axis.set_yscale("log")
axis.set_title(title)
for axis in axes:
axis.set_xlabel("Предложенная нагрузка, запросов/с")
axis.legend(loc="upper left", fontsize=9)
figure.suptitle(f"{series}: медиана 3 повторов, усы — min–max", color=INK)
figure.tight_layout()
figure.savefig(out / f"{series}-throughput.png", dpi=150)
plt.close(figure)
def cpu(series, keys, groups, out):
figure, axes = plt.subplots(1, len(keys), figsize=(5 * len(keys), 4.2), squeeze=False)
for axis, key in zip(axes[0], keys, strict=True):
rates = sorted(groups[key])
first = groups[key][rates[0]][0]
limit = int(first["replicas"]) * float(first["cpus"]) * 100
for index, (field, name) in enumerate(COMPONENTS):
values = [med(groups[key][rate], f"cpu_{field}")[0] for rate in rates]
axis.plot(rates, values, color=PALETTE[index], marker=MARKERS[index], label=name)
axis.axhline(limit, color=PALETTE[0], linestyle=":", linewidth=1.2, label="лимит API")
axis.axhline(200, color=PALETTE[1], linestyle=":", linewidth=1.2, label="лимит PostgreSQL")
axis.set_title(f"CPU, % ядра — {label(key)}")
axis.set_xlabel("Предложенная нагрузка, запросов/с")
axis.legend(fontsize=8, loc="upper left")
figure.tight_layout()
figure.savefig(out / f"{series}-cpu.png", dpi=150)
plt.close(figure)
def usl(points):
base = points[0][1] / points[0][0]
pairs = [(n - 1, n * (n - 1), n * base / x - 1) for n, x in points]
a, b, c = (sum(p[i] * p[j] for p in pairs) for i, j in ((0, 0), (0, 1), (1, 1)))
d, e = (sum(p[i] * p[2] for p in pairs) for i in (0, 1))
det = a * c - b * b
sigma, kappa = (d * c - b * e) / det, (a * e - b * d) / det
peak = ((1 - sigma) / kappa) ** 0.5 if kappa > 0 else None
return base, sigma, kappa, peak
def scaling(rows, out):
by = {(r["series"], r["config"]): r for r in rows if r["variant"] == "base"}
growth = [
by.get(("series-b", c)) or by.get(("series-a", c)) for c in ("1x1", "2x1", "4x1", "8x1")
]
growth = [r for r in growth if r]
if not growth:
return
figure, axis = plt.subplots(figsize=(7.5, 4.8))
cpus = [r["cpu_api_limit"] / 100 for r in growth]
unit = growth[0]["peak_ok_rps"] / cpus[0]
axis.plot(cpus, [unit * c for c in cpus], color=MUTED, linestyle="--", linewidth=1)
axis.annotate("линейный рост", (cpus[-1], unit * cpus[-1]), color=MUTED, fontsize=9)
fit = None
if len(growth) >= USL_POINTS:
fit = usl([(c, r["peak_ok_rps"]) for c, r in zip(cpus, growth, strict=True)])
base, sigma, kappa, _ = fit
grid = [1 + i * (cpus[-1] - 1) / 60 for i in range(61)]
model = [base * n / (1 + sigma * (n - 1) + kappa * n * (n - 1)) for n in grid]
axis.plot(
grid, model, color=PALETTE[6], linewidth=1.5, label=f"USL: σ={sigma:.3f}, κ={kappa:.4f}"
)
axis.errorbar(
cpus,
[r["peak_ok_rps"] for r in growth],
yerr=[
[r["peak_ok_rps"] - r["peak_ok_min"] for r in growth],
[r["peak_ok_max"] - r["peak_ok_rps"] for r in growth],
],
color=PALETTE[0],
marker="o",
capsize=3,
label="серия B: N × 1 CPU",
)
for r, c in zip(growth, cpus, strict=True):
direct_label(axis, c, r["peak_ok_rps"], r["config"])
for config in ("2x2", "1x4"):
r = by.get(("series-a", config))
if r:
look = style(config)
axis.plot(4, r["peak_ok_rps"], linestyle="none", label=f"серия A: {config}", **look)
after = [r for r in rows if r["variant"] == "after"]
for r in after:
width = r["cpu_api_limit"] / 100
axis.plot(
width,
r["peak_ok_rps"],
linestyle="none",
label=f"{r['config']} после улучшения",
**VARIANTS["after"],
)
axis.set_xlabel("Суммарный CPU реплик API")
axis.set_ylabel("Пик успешных RPS")
axis.set_title("Пропускная способность от ресурсов API")
axis.set_xticks(cpus)
axis.legend(loc="upper left", fontsize=9)
figure.tight_layout()
figure.savefig(out / "scaling.png", dpi=150)
plt.close(figure)
return fit
def utilisation(rows, out):
picked = [r for r in rows if r["variant"] == "base" and r["series"] != "compare"]
seen, ordered = set(), []
for config in ORDER:
for r in picked:
if r["config"] == config and config not in seen:
seen.add(config)
ordered.append(r)
if not ordered:
return
metrics = (
("CPU API, % лимита", lambda r: 100 * r["cpu_api"] / r["cpu_api_limit"]),
("CPU PostgreSQL, % лимита", lambda r: r["cpu_postgres"] / 2),
("CPU HAProxy, % лимита", lambda r: r["cpu_lb"]),
("ждут блокировку, % пула", lambda r: 100 * r["pg_lock_avg"] / r["pool_total"]),
)
figure, axis = plt.subplots(figsize=(10, 4.8))
width = 0.8 / len(metrics)
for index, (name, value) in enumerate(metrics):
xs = [i + (index - 1.5) * width for i in range(len(ordered))]
axis.bar(
xs,
[value(r) for r in ordered],
width=width - 0.02,
color=PALETTE[index],
label=name,
edgecolor=SURFACE,
)
axis.set_xticks(range(len(ordered)), [r["config"] for r in ordered])
axis.axhline(100, color=MUTED, linestyle=":", linewidth=1)
axis.set_ylabel("%")
axis.set_ylim(0, 115)
axis.set_title("Загрузка звеньев в точке пика успешных RPS", pad=34)
axis.legend(fontsize=9, ncols=4, loc="lower left", bbox_to_anchor=(0, 1.0))
figure.tight_layout()
figure.savefig(out / "utilisation.png", dpi=150)
plt.close(figure)
def write(rows, out, fit):
with (out / "capacity.csv").open("w", newline="") as handle:
writer = csv.DictWriter(handle, fieldnames=SUMMARY, extrasaction="ignore")
writer.writeheader()
writer.writerows(rows)
head = (
"| серия | вариант | конфигурация | потолок по SLO, rps | p95 там, мс | пик успешных RPS "
"[min–max] | при нагрузке | CPU API / лимит | CPU PG | CPU LB | ждут блокировку / пул |"
)
lines_ = [head, "|" + " --- |" * 11]
for r in rows:
lines_.append(
f"| {r['series']} | {r['variant']} | {r['config']} | {r.get('capacity_rate', '—')} | "
f"{r.get('capacity_p95_ms', '—')} | {r['peak_ok_rps']:.0f} "
f"[{r['peak_ok_min']:.0f}–{r['peak_ok_max']:.0f}] | {r['peak_rate']} | "
f"{r['cpu_api']:.0f} / {r['cpu_api_limit']} | {r['cpu_postgres']:.0f} | "
f"{r['cpu_lb']:.0f} | {r['pg_lock_avg']:.1f} / {r['pool_total']} |"
)
if fit:
_, sigma, kappa, peak = fit
best = f"{peak:.1f}" if peak else "—"
lines_.append(
f"\nUSL по серии B: σ = {sigma:.4f}, κ = {kappa:.5f}, пик при N* = {best} CPU"
)
(out / "capacity.md").write_text("\n".join(lines_) + "\n")
print("\n".join(lines_))
def main():
parser = argparse.ArgumentParser(description="Графики и сводка трека 1 из summary.csv")
parser.add_argument("summaries", nargs="+")
parser.add_argument("--out", required=True)
args = parser.parse_args()
out = Path(args.out)
out.mkdir(parents=True, exist_ok=True)
groups = load(args.summaries)
rank = {name: index for index, name in enumerate(ORDER)}
keys = sorted(groups, key=lambda k: (k[0], k[1] != "before", rank.get(k[2], 99)))
rows = [summarize(key, groups[key]) for key in keys]
for series in dict.fromkeys(k[0] for k in keys):
chosen = [k for k in keys if k[0] == series]
throughput(series, chosen, groups, out)
cpu(series, chosen, groups, out)
fit = scaling(rows, out)
utilisation(rows, out)
write(rows, out, fit)
if __name__ == "__main__":
main()