Measure scaling cost and confirm two-replica operating point

This commit is contained in:
Самохвалов Андрей Ремович
2026-10-02 13:59:18 +03:00
parent 23ed63f593
commit 750e56d0c6
13 changed files with 739 additions and 5 deletions
+86
View File
@@ -0,0 +1,86 @@
#!/usr/bin/env python3
"""Render the three-repeat scale matrix as small, dependency-free SVG plots."""
import argparse
import csv
import json
from pathlib import Path
import statistics
from benchmark import validate_matrix
COLORS = {1: "#2563eb", 2: "#16a34a", 4: "#dc2626"}
def svg(mode, metric, rows, output):
width, height = 720, 420
left, right, top, bottom = 75, 25, 35, 65
plot_w, plot_h = width - left - right, height - top - bottom
rates = sorted({row["rate"] for row in rows})
max_y = max((row[metric] or 0 for row in rows), default=0) * 1.1 or 1
x = lambda rate: left + (rate - rates[0]) / (rates[-1] - rates[0]) * plot_w
y = lambda value: top + plot_h * (1 - value / max_y)
label = "Успешные запросы/с" if metric == "successful_rps" else "p95, мс"
parts = [f'<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {width} {height}" role="img" aria-label="{mode}: {label}">',
'<rect width="100%" height="100%" fill="white"/>',
f'<text x="{left}" y="23" font-family="sans-serif" font-size="16">{mode}: {label}</text>']
for step in range(5):
value = max_y * step / 4
py = y(value)
parts += [f'<line x1="{left}" x2="{width-right}" y1="{py:.1f}" y2="{py:.1f}" stroke="#ddd"/>',
f'<text x="{left-8}" y="{py+4:.1f}" text-anchor="end" font-family="sans-serif" font-size="11">{value:.0f}</text>']
for rate in rates:
px = x(rate)
parts += [f'<text x="{px:.1f}" y="{height-35}" text-anchor="middle" font-family="sans-serif" font-size="11">{rate}</text>']
for replicas in (1, 2, 4):
points = [(row["rate"], row[metric]) for row in rows if row["replicas"] == replicas and row[metric] is not None]
path = " ".join(f'{"M" if index == 0 else "L"}{x(rate):.1f},{y(value):.1f}' for index, (rate, value) in enumerate(points))
parts.append(f'<path d="{path}" fill="none" stroke="{COLORS[replicas]}" stroke-width="2.5"/>')
for rate, value in points:
parts.append(f'<circle cx="{x(rate):.1f}" cy="{y(value):.1f}" r="4" fill="{COLORS[replicas]}"/>')
legend_x = left + 180 * (replicas == 2) + 360 * (replicas == 4)
parts.append(f'<text x="{legend_x}" y="{height-10}" fill="{COLORS[replicas]}" font-family="sans-serif" font-size="12">{replicas} API</text>')
parts += [f'<text x="{left+plot_w/2}" y="{height-48}" text-anchor="middle" font-family="sans-serif" font-size="11">Заданная интенсивность, запросов/с</text>', '</svg>']
output.write_text("\n".join(parts) + "\n", encoding="utf-8")
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("input", type=Path)
args = parser.parse_args()
rows = [json.loads(line) for line in args.input.read_text().splitlines() if line.strip()]
validate_matrix(rows)
summary = []
for mode in ("fixed", "growth"):
for replicas in (1, 2, 4):
for rate in sorted({row["rate"] for row in rows}):
group = [row for row in rows if (row["mode"], row["replicas"], row["rate"]) == (mode, replicas, rate)]
def latency(p):
values = [row["successful_latency_ms"][p] for row in group if row["successful_latency_ms"][p] is not None]
return round(statistics.median(values), 3) if values else None
summary.append({"mode": mode, "replicas": replicas, "rate": rate,
"successful_rps": round(statistics.median(row["successful_rps"] for row in group), 3),
"error_ratio": round(statistics.median(row["error_ratio"] for row in group), 6),
"p50_ms": latency("p50"), "p95_ms": latency("p95"), "p99_ms": latency("p99"),
"api_cpu_percent": round(statistics.median(row["docker"]["api"]["avg_cpu_percent"] for row in group), 3),
"postgres_cpu_percent": round(statistics.median(row["docker"]["postgres"]["avg_cpu_percent"] for row in group), 3),
"lb_cpu_percent": round(statistics.median(row["docker"]["lb"]["avg_cpu_percent"] for row in group), 3),
"api_memory_mib": round(statistics.median(row["docker"]["api"]["max_memory_mib"] for row in group), 3),
"postgres_memory_mib": round(statistics.median(row["docker"]["postgres"]["max_memory_mib"] for row in group), 3),
"lb_memory_mib": round(statistics.median(row["docker"]["lb"]["max_memory_mib"] for row in group), 3),
"pool_wait_ms": round(statistics.median(row["pool"]["wait_ms"] for row in group), 3),
"queue_dropped": statistics.median(row["queue_dropped"] for row in group)})
with (args.input.parent / "summary.csv").open("w", newline="") as file:
writer = csv.DictWriter(file, fieldnames=summary[0].keys(), lineterminator="\n")
writer.writeheader()
writer.writerows(summary)
for mode in ("fixed", "growth"):
group = [row for row in summary if row["mode"] == mode]
svg(mode, "successful_rps", group, args.input.parent / f"{mode}-rps.svg")
svg(mode, "p95_ms", group, args.input.parent / f"{mode}-p95.svg")
if __name__ == "__main__":
main()