性能指南

# pandas: Loads ALL data, then filters in memory
df = pd.read_csv("huge.csv")       # Load 10GB
df = df[df['year'] == 2024]        # Filter in Python

# DataStore: Filter at source
ds = pd.read_csv("huge.csv")       # Just metadata
ds = ds[ds['year'] == 2024]        # Filter in SQL
df = ds.to_df()                    # Only load filtered data

# DataStore: Only reads name, age columns
ds = pd.read_parquet("wide_table.parquet")
result = ds.select('name', 'age').to_df()

# vs pandas: Reads all 100 columns, then selects

# DataStore: One optimized SQL query
result = (ds
    .filter(ds['amount'] > 100)
    .groupby('region')
    .agg({'amount': 'sum'})
    .sort('sum', ascending=False)
    .head(10)
    .to_df()
)

# Becomes:
# SELECT region, SUM(amount) FROM data
# WHERE amount > 100
# GROUP BY region ORDER BY sum DESC LIMIT 10

# DataStore excels: 19.93x faster
result = ds.groupby('category')['amount'].sum()

# DataStore excels: 5-6x faster
result = (ds
    .filter(ds['date'] >= '2024-01-01')
    .filter(ds['amount'] > 100)
    .groupby('region')
    .agg({'amount': ['sum', 'mean', 'count']})
    .sort('sum', ascending=False)
    .head(20)
)

# DataStore: Only loads what you need
ds = pd.read_parquet("huge_file.parquet")
result = ds.filter(ds['id'] == 12345).to_df()  # Fast!

# DataStore: Combines into single SQL
ds['total'] = ds['price'] * ds['quantity']
ds['is_large'] = ds['total'] > 1000
ds = ds.filter(ds['is_large'])

# For very small datasets, overhead is minimal for both
# Performance difference is negligible
small_df = pd.DataFrame({'x': range(100)})

# Single slice operations without aggregation
df = df[df['x'] > 10]  # pandas slightly faster
ds = ds[ds['x'] > 10]  # DataStore comparable

# pandas required for custom Python code
def complex_function(row):
    return custom_logic(row)

df['result'] = df.apply(complex_function, axis=1)

# to_df() does NOT copy data - it's a zero-copy operation
result = ds.filter(ds['x'] > 10).to_df()  # No data conversion overhead

# Same for creating DataStore from DataFrame
ds = DataStore(existing_df)  # No data copy

from chdb.datastore.config import config

config.use_performance_mode()

# Now all operations use SQL-first execution with no pandas overhead:
# - Parallel Parquet reading (no preserve_order)
# - Single-SQL aggregation (filter+groupby in one query)
# - No row-order preservation overhead
# - No MultiIndex, no dtype corrections
result = (ds
    .filter(ds['amount'] > 100)
    .groupby('region')
    .agg({'amount': ['sum', 'mean', 'count']})
)

# CSV: Slower, reads entire file
ds = pd.read_csv("data.csv")

# Parquet: Faster, columnar, compressed
ds = pd.read_parquet("data.parquet")

# Convert once, benefit forever
df = pd.read_csv("data.csv")
df.to_parquet("data.parquet")

# Good: Filter first, then aggregate
result = (ds
    .filter(ds['date'] >= '2024-01-01')  # Reduce data early
    .groupby('category')['amount'].sum()
)

# Less optimal: Process all data
result = (ds
    .groupby('category')['amount'].sum()
    .filter(ds['sum'] > 1000)  # Filter too late
)

# Good: Column pruning
result = ds.select('name', 'amount').filter(ds['amount'] > 100)

# Less optimal: All columns loaded
result = ds.filter(ds['amount'] > 100)  # Loads all columns

# GroupBy is where DataStore shines
# Up to 20x speedup!
result = ds.groupby('category').agg({
    'amount': ['sum', 'mean', 'count', 'max'],
    'quantity': 'sum'
})

# Don't load entire result if you only need a sample
result = ds.filter(ds['type'] == 'A').head(100)  # LIMIT 100

# Avoid this for large results
# result = ds.filter(ds['type'] == 'A').to_df()  # Loads everything

# Good: Single execution
result = ds.filter(ds['x'] > 10).filter(ds['y'] < 100).to_df()

# Bad: Multiple executions
result1 = ds.filter(ds['x'] > 10).to_df()  # Execute
result2 = result1[result1['y'] < 100]       # Execute again

# View the query plan before executing
query = ds.filter(...).groupby(...).agg(...)
query.explain()  # Check if operations are pushed down

# Then execute
result = query.to_df()

from chdb.datastore.config import config, get_profiler

config.enable_profiling()

# Run your workload
result = your_pipeline()

# View report
profiler = get_profiler()
profiler.report()

Performance Report
==================
Step                    Duration    % Total
----                    --------    -------
SQL execution           2.5s        62.5%     <- Bottleneck!
read_csv                1.2s        30.0%
Other                   0.3s        7.5%

# Test approach 1
profiler.reset()
result1 = approach1()
time1 = profiler.get_steps()[-1]['duration_ms']

# Test approach 2
profiler.reset()
result2 = approach2()
time2 = profiler.get_steps()[-1]['duration_ms']

print(f"Approach 1: {time1:.0f}ms")
print(f"Approach 2: {time2:.0f}ms")