メインコンテンツへスキップ
メインコンテンツへスキップ

explain() メソッド

explain() メソッドは DataStore のクエリに対する実行計画を表示し、どのような処理が行われ、どのような SQL が生成されるかを把握するのに役立ちます。

基本的な使い方​

from chdb import datastore as pd

ds = pd.read_csv("sales.csv")

query = (ds
    .filter(ds['amount'] > 1000)
    .groupby('region')
    .agg({'amount': ['sum', 'mean']})
    .sort('sum', ascending=False)
)

# View execution plan
query.explain()

構文​

explain(verbose=False) -> None

パラメーター:

パラメーター型デフォルト説明
verboseboolFalse追加のメタデータを表示します

出力フォーマット​

標準出力​

================================================================================
Execution Plan (in execution order)
================================================================================

 [1] 📊 Data Source: file('sales.csv', 'csv')

Operations:
────────────────────────────────────────────────────────────────────────────────
    ️  Segment 1 [chDB] (from source): Operations 2-5
    ️  Note: SQL operations after Pandas ops use Python() table function

 [2] 🚀 [chDB] WHERE: "amount" > 1000
 [3] 🚀 [chDB] GROUP BY: region
 [4] 🚀 [chDB] AGGREGATE: sum(amount), avg(amount)
 [5] 🚀 [chDB] ORDER BY: sum DESC

────────────────────────────────────────────────────────────────────────────────
Final State: 📊 Pending (lazy, not yet executed)
             └─> Will execute when print(), .to_df(), .execute() is called

────────────────────────────────────────────────────────────────────────────────
Generated SQL Query:
────────────────────────────────────────────────────────────────────────────────

SELECT region, SUM(amount) AS sum, AVG(amount) AS mean
FROM file('sales.csv', 'csv')
WHERE "amount" > 1000
GROUP BY region
ORDER BY sum DESC

================================================================================

アイコンの凡例​

アイコン意味
📊データソース
🚀chDB (SQL) 操作
🐼pandas 操作

詳細な出力​

query.explain(verbose=True)

Verbose モードでは、各オペレーションごとに、内部の行順序トラッキング機構を含む完全な SQL クエリなど、追加の詳細が表示されます。

3 つの実行フェーズ​

explain の出力では、処理が次の 3 つのフェーズに分けて示されます。

フェーズ1: SQL クエリ構築（遅延評価）​

SQL にコンパイルされる操作:

  1. Source: file('sales.csv', 'CSVWithNames')
  2. Filter: amount > 1000      
  3. GroupBy: region
  4. Aggregate: sum(amount)

フェーズ 2: 実行ポイント​

トリガーが発生したとき:

  5. Execute SQL -> DataFrame
     Trigger: to_df() called

フェーズ 3: DataFrame の操作​

実行後に行う操作:

  6. [pandas] pivot_table(...)
  7. [pandas] apply(custom_func)

プランを理解する​

ソース情報​

Source: file('sales.csv', 'CSVWithNames')
  • file() - ClickHouse の file() テーブル関数
  • 'CSVWithNames' - ヘッダー行付きのファイル形式

その他のソースタイプ:

Source: s3('bucket/data.parquet', ...)
Source: mysql('host', 'db', 'table', ...)
Source: __dataframe__  (pandas DataFrame input)

フィルタ処理​

Filter: amount > 1000 AND status = 'active'

適用される WHERE 句を表示します。

GROUP BY と集約​

GroupBy: region, category
Aggregate: sum(amount), avg(amount), count(id)

GROUP BY のカラムと集約関数を表示します。

ソート操作​

Sort: sum DESC, region ASC

ORDER BY 句を示します。

LIMIT 操作​

Limit: 10
Offset: 100

LIMIT および OFFSET を示します。

エンジン情報​

verbose モードを使用すると、どのエンジンが使われるかを確認できます。

Filter: amount > 1000
  - Engine: chdb
  - Pushdown: Yes

Apply: custom_function
  - Engine: pandas
  - Pushdown: No

プッシュダウン​

  • はい: 処理はデータソース（SQL）側で実行されます
  • いいえ: 処理は pandas 側での実行が必要です

例​

単純なクエリ​

ds = pd.read_csv("data.csv")
ds.filter(ds['age'] > 25).explain()

================================================================================
Execution Plan (in execution order)
================================================================================

 [1] 📊 Data Source: file('data.csv', 'csv')

Operations:
────────────────────────────────────────────────────────────────────────────────
    ️  Segment 1 [chDB] (from source): Operations 2-2

 [2] 🚀 [chDB] WHERE: "age" > 25

────────────────────────────────────────────────────────────────────────────────
Generated SQL Query:
────────────────────────────────────────────────────────────────────────────────

SELECT * FROM file('data.csv', 'csv') WHERE "age" > 25

================================================================================

高度な集約​

query = (ds
    .filter(ds['date'] >= '2024-01-01')
    .filter(ds['amount'] > 100)
    .select('region', 'category', 'amount')
    .groupby('region', 'category')
    .agg({
        'amount': ['sum', 'mean', 'count']
    })
    .sort('sum', ascending=False)
    .limit(20)
)
query.explain()

================================================================================
Execution Plan (in execution order)
================================================================================

 [1] 📊 Data Source: file('sales.csv', 'csv')

Operations:
────────────────────────────────────────────────────────────────────────────────
    ️  Segment 1 [chDB] (from source): Operations 2-8

 [2] 🚀 [chDB] WHERE: "date" >= '2024-01-01'
 [3] 🚀 [chDB] WHERE: "amount" > 100
 [4] 🚀 [chDB] SELECT: region, category, amount
 [5] 🚀 [chDB] GROUP BY: region, category
 [6] 🚀 [chDB] AGGREGATE: sum(amount), avg(amount), count(amount)
 [7] 🚀 [chDB] ORDER BY: sum DESC
 [8] 🚀 [chDB] LIMIT: 20

────────────────────────────────────────────────────────────────────────────────
Generated SQL Query:
────────────────────────────────────────────────────────────────────────────────

SELECT region, category, 
       SUM(amount) AS sum, 
       AVG(amount) AS mean, 
       COUNT(amount) AS count
FROM file('sales.csv', 'csv')
WHERE "date" >= '2024-01-01' AND "amount" > 100
GROUP BY region, category
ORDER BY sum DESC
LIMIT 20

================================================================================

SQL と pandas の混在​

処理をすべて SQL 側に委譲できない場合、実行プランには複数のセグメントが表示されます。

query = (ds
    .filter(ds['age'] > 25)           # SQL
    .groupby('city')                   # SQL
    .agg({'salary': 'mean'})           # SQL
    .apply(lambda x: x * 1.1)          # pandas (triggers segment split)
    .filter(ds['mean'] > 50000)        # SQL (new segment)
)
query.explain()

================================================================================
Execution Plan (in execution order)
================================================================================

 [1] 📊 Data Source: file('data.csv', 'csv')

Operations:
────────────────────────────────────────────────────────────────────────────────
    ️  Segment 1 [chDB] (from source): Operations 2-4
    ️  Segment 2 [Pandas] (on DataFrame): Operation 5
    ️  Segment 3 [chDB] (on DataFrame): Operation 6
    ️  Note: SQL operations after Pandas ops use Python() table function

 [2] 🚀 [chDB] WHERE: "age" > 25
 [3] 🚀 [chDB] GROUP BY: city
 [4] 🚀 [chDB] AGGREGATE: avg(salary)
 [5] 🐼 [Pandas] APPLY: lambda
 [6] 🚀 [chDB] WHERE: "mean" > 50000

================================================================================

explain() を使ったデバッグ​

フィルター条件のロジックを確認する​

# Verify your filter is correct
query = ds.filter((ds['age'] > 25) & (ds['city'] == 'NYC'))
query.explain()
# Output shows: Filter: age > 25 AND city = 'NYC'

カラム選択を検証する​

# Check column pruning
query = ds.select('name', 'age').filter(ds['age'] > 25)
query.explain()
# Output shows: SELECT name, age FROM ... WHERE age > 25

集約を理解する​

# Check aggregation functions
query = ds.groupby('dept').agg({'salary': ['sum', 'mean', 'std']})
query.explain()
# Output shows: SELECT dept, SUM(salary), AVG(salary), stddevPop(salary)

ベストプラクティス​

1. 大規模クエリ実行前の確認​

# Always explain first for large data
query = ds.complex_pipeline()
query.explain()  # Check plan

# If plan looks correct
result = query.to_df()  # Execute

2. デバッグ時は VERBOSE を使用する​

# When something seems wrong
query.explain(verbose=True)
# Shows engine selection and pushdown info

3. to_sql() との比較​

# explain() shows the plan
query.explain()

# to_sql() shows just the SQL
print(query.to_sql())

# Both useful for different purposes

4. プッシュダウンの状態を確認する​

# Verbose mode shows if operations are pushed down
query.explain(verbose=True)

# If Pushdown: No, operation runs in pandas
# Consider restructuring query for better performance