docs: REPORT lab2, доказательства со стенда из чистого clone, размеры индексов

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Alexey Bodanyuk
2026-10-12 20:20:00 +03:00
co-authored by Claude Opus 5.5
parent eb92e8e91e
commit d65ba991d7
15 changed files with 613 additions and 25 deletions
+42
View File
@@ -0,0 +1,42 @@
#!/usr/bin/env python3
"""Index sizes of every variant on the SAME 121 120-row dataset (no writes in between).
python bench/track2/index_sizes.py # ~2 min, writes evidence/track2/index-sizes.csv
"""
from __future__ import annotations
import csv
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent))
import plans # noqa: E402 (same image, limits, schema and deterministic seed)
def main() -> int:
plans.start_container("2", "2g")
try:
plans.psql("".join(p.read_text(encoding="utf-8") for p in sorted((plans.ROOT / "migrations").glob("*.sql"))))
plans.psql(plans.SEED)
rows = int(plans.psql("SELECT count(*) FROM messages;"))
out = []
for variant in plans.VARIANTS:
info = plans.use_variant(variant)
secondary = sum(int(b) for name, b in info["index_bytes"].items() if name != "messages_pkey")
out.append({"variant": variant, "rows": rows, "table_bytes": info["table_bytes"],
"pkey_bytes": int(info["index_bytes"]["messages_pkey"]), "secondary_index_bytes": secondary,
"secondary_bytes_per_row": round(secondary / rows, 1), "index_build_seconds": info["index_build_s"]})
print(out[-1], flush=True)
finally:
plans.sh(["docker", "rm", "-f", "-v", plans.CLIENT, plans.NAME], check=False)
path = plans.ROOT / "evidence/track2/index-sizes.csv"
with path.open("w", newline="", encoding="utf-8") as f:
writer = csv.DictWriter(f, fieldnames=list(out[0]))
writer.writeheader()
writer.writerows(out)
print("done:", path)
return 0
if __name__ == "__main__":
sys.exit(main())
+6 -3
View File
@@ -42,6 +42,7 @@ def fmt(x: float) -> str:
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"))
@@ -60,14 +61,16 @@ def main() -> int:
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) | Размер индексов, МиБ |",
"| 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 = meta["sizes"][variant]["index_bytes"]
idx_mib = sum(int(b) for b in idx.values()) / 2**20
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))