Files
mtusi2026lab2/bench/track2/summarize.py
T

114 lines
6.3 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.
#!/usr/bin/env python3
"""Aggregate track 2 raw data into tables (summary.md) and charts (plans.png).
python bench/track2/summarize.py # matplotlib from requirements-bench.txt for the chart
"""
from __future__ import annotations
import csv
import json
import statistics
import sys
from collections import defaultdict
from pathlib import Path
ROOT = Path(__file__).resolve().parents[2]
OUT = ROOT / "evidence/track2"
SIZE_ORDER = {"big": 0, "medium": 1, "small": 2}
def load(name: str) -> list[dict]:
with (OUT / name).open(encoding="utf-8") as f:
return list(csv.DictReader(f))
def agg(rows: list[dict], keys: tuple[str, ...], fields: tuple[str, ...]) -> dict:
groups: dict = defaultdict(list)
for r in rows:
groups[tuple(r[k] for k in keys)].append(r)
out = {}
for k, g in groups.items():
out[k] = {f: statistics.median(float(r[f]) for r in g) for f in fields}
out[k]["tps_min"] = min(float(r["tps"]) for r in g)
out[k]["tps_max"] = max(float(r["tps"]) for r in g)
out[k]["n"] = len(g)
return out
def fmt(x: float) -> str:
return f"{x:,.0f}".replace(",", " ") if x >= 100 else (f"{x:.1f}" if x >= 10 else f"{x:.2f}")
def main() -> int:
sys.stdout.reconfigure(encoding="utf-8")
reads, writes, wal = load("reads.csv"), load("writes.csv"), load("wal.csv")
sizes_rows = {r["variant"]: r for r in load("index-sizes.csv")}
meta = json.loads((OUT / "plans-meta.json").read_text(encoding="utf-8"))
r = agg(reads, ("variant", "query", "chat"), ("tps", "p50_ms", "p95_ms", "p99_ms"))
w = agg(writes, ("variant",), ("tps", "p50_ms", "p95_ms", "p99_ms"))
wal_by = {v: [int(r["wal_bytes_per_insert"]) for r in wal if r["variant"] == v] for v in {r["variant"] for r in wal}}
for v, values in wal_by.items():
w[(v,)]["wal_bytes_per_insert"] = statistics.median(values)
w[(v,)]["wal_min"], w[(v,)]["wal_max"] = min(values), max(values)
sizes = {"big": 60000, "medium": 15000, "small": 80}
md = ["## Чтение: одна страница (LIMIT 51) со случайной позиции чата", "",
"Медиана 3 повторов по 10 с, 4 клиента pgbench, prepared statements. TPS min–max — разброс повторов.", "",
"| Индекс | Запрос | Чат (сообщений) | TPS (min–max) | p50, мс | p95, мс | p99, мс |",
"| --- | --- | ---: | ---: | ---: | ---: | ---: |"]
for (variant, query, chat), v in sorted(r.items(), key=lambda kv: (kv[0][0] != "keyset_index", kv[0][0], kv[0][1], SIZE_ORDER[kv[0][2]])):
md.append(f"| {variant} | {query} | {chat} ({sizes[chat]:,}) | {fmt(v['tps'])} ({fmt(v['tps_min'])}–{fmt(v['tps_max'])}) "
f"| {fmt(v['p50_ms'])} | {fmt(v['p95_ms'])} | {fmt(v['p99_ms'])} |".replace(",", " ", 1))
md += ["", "## Запись: транзакция сообщения приложения (UPDATE chats … RETURNING + INSERT)", "",
"TPS и задержка — медиана 5 повторов по 10 с, 8 клиентов. WAL — медиана 3 детерминированных прогонов: "
"5 000 транзакций, 1 клиент, autovacuum выключен, CHECKPOINT перед прогоном.", "",
"| Индексы messages | TPS (min–max) | p50, мс | p95, мс | p99, мс | WAL, байт/вставку (min–max) | Доп. индексы на 121 120 строк, МиБ |",
"| --- | ---: | ---: | ---: | ---: | ---: | ---: |"]
for variant in ("no_index", "keyset_index", "keyset_plus_sender"):
v = w[(variant,)]
idx_mib = int(sizes_rows[variant]["secondary_index_bytes"]) / 2**20 # same dataset for every variant
md.append(f"| {variant} | {fmt(v['tps'])} ({fmt(v['tps_min'])}–{fmt(v['tps_max'])}) | {fmt(v['p50_ms'])} | {fmt(v['p95_ms'])} "
f"| {fmt(v['p99_ms'])} | {v['wal_bytes_per_insert']:.0f} ({v['wal_min']}–{v['wal_max']}) | {idx_mib:.1f} |")
md += ["", "Размеры индексов измерены отдельно на одинаковых 121 120 строках (bench/track2/index_sizes.py); "
"первичный ключ во всех вариантах — " + f"{int(sizes_rows['no_index']['pkey_bytes']) / 2**20:.1f} МиБ, таблица — "
f"{int(sizes_rows['no_index']['table_bytes']) / 2**20:.1f} МиБ."]
(OUT / "summary.md").write_text("\n".join(md) + "\n", encoding="utf-8", newline="\n")
print("\n".join(md))
try:
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
except ImportError:
print("matplotlib is not installed: skipping plans.png")
return 0
fig, axes = plt.subplots(1, 2, figsize=(15, 5))
chats = ["small", "medium", "big"]
x = [sizes[c] for c in chats]
styles = {("keyset_index", "keyset"): ("#12a150", "-o"), ("keyset_index", "offset"): ("#12a150", "--s"),
("chat_id_index", "keyset"): ("#e07b00", "-o"), ("chat_id_index", "offset"): ("#e07b00", "--s"),
("no_index", "keyset"): ("#b03060", "-o"), ("no_index", "offset"): ("#b03060", "--s")}
for (variant, query), (color, style) in styles.items():
axes[0].plot(x, [r[(variant, query, c)]["p95_ms"] for c in chats], style, color=color, label=f"{variant}, {query}")
axes[0].set_xscale("log")
axes[0].set_yscale("log")
axes[0].set_xlabel("messages in the chat")
axes[0].set_ylabel("p95 latency of one page, ms (log)")
axes[0].set_title("Read: random page of a chat")
axes[0].grid(True, which="both", alpha=0.3)
axes[0].legend(fontsize=8)
names = ["no_index", "keyset_index", "keyset_plus_sender"]
axes[1].bar([n.replace("_", "\n") for n in names], [w[(n,)]["wal_bytes_per_insert"] for n in names], color="#2a6fdb")
for i, n in enumerate(names):
axes[1].text(i, w[(n,)]["wal_bytes_per_insert"], f"{w[(n,)]['tps']:.0f} TPS", ha="center", va="bottom")
axes[1].set_ylabel("WAL bytes per inserted message")
axes[1].set_title("Write cost of indexes (WAL: 5000 inserts; TPS: 8 clients)")
fig.suptitle("Track 2: 121 120 messages, chats of 60 000 / 15 000 / 2 000 / 80; PostgreSQL 18.6, 2 CPU / 2 GiB")
fig.tight_layout()
fig.savefig(OUT / "plans.png", dpi=120)
print("chart:", OUT / "plans.png")
return 0
if __name__ == "__main__":
sys.exit(main())