database-optimization（数据库优化）

Database Optimization

EXPLAIN Analysis

Always run EXPLAIN ANALYZE before optimizing. Read the output bottom-up.

-- PostgreSQL
EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT) SELECT ...;

-- MySQL
EXPLAIN ANALYZE SELECT ...;

Key metrics to watch:

• Seq Scan on large tables = missing index
• Nested Loop with high row count = consider hash/merge join
• Sort without index = add index on sort column
• Rows estimated vs actual divergence = stale statistics, run ANALYZE

Index Strategies

B-tree (default, most cases)

CREATE INDEX idx_users_email ON users (email);
CREATE INDEX idx_orders_user_date ON orders (user_id, created_at DESC);

Use for: equality, range queries, sorting. Column order matters in composite indexes: put equality columns first, then range/sort columns.

Partial Index (PostgreSQL)

CREATE INDEX idx_orders_pending ON orders (created_at)
  WHERE status = 'pending';

Use when queries always filter on a specific condition. Dramatically smaller than full indexes.

GIN (PostgreSQL - arrays, JSONB, full-text)

CREATE INDEX idx_products_tags ON products USING GIN (tags);
CREATE INDEX idx_docs_search ON documents USING GIN (to_tsvector('english', content));

GiST (PostgreSQL - spatial, range types)

CREATE INDEX idx_locations_point ON locations USING GiST (coordinates);
CREATE INDEX idx_events_period ON events USING GiST (tsrange(start_at, end_at));

Covering Index (index-only scans)

-- PostgreSQL
CREATE INDEX idx_users_email_name ON users (email) INCLUDE (name);

-- MySQL
CREATE INDEX idx_users_email_name ON users (email, name);

N+1 Query Detection

Symptom: 1 query to fetch parent + N queries for each child.

# BAD: N+1
users = db.query(User).all()
for user in users:
    print(user.orders)  # triggers query per user

# GOOD: eager load
users = db.query(User).options(joinedload(User.orders)).all()

// BAD: N+1
const users = await User.findAll();
for (const user of users) {
  const orders = await Order.findAll({ where: { userId: user.id } });
}

// GOOD: batch load
const users = await User.findAll({ include: [Order] });

Detection: enable query logging, count queries per request. More than 10 queries for a single endpoint is a red flag.

Connection Pooling

Rule of thumb: pool_size = (core_count * 2) + disk_count
Typical web app: 10-20 connections per app instance

PostgreSQL:

• Use PgBouncer in transaction mode for serverless/high-connection scenarios
• Set idle_in_transaction_session_timeout = '30s'
• Monitor with pg_stat_activity

MySQL:

• Set max_connections based on available RAM (each connection uses ~10MB)
• Use ProxySQL for connection multiplexing
• Monitor with SHOW PROCESSLIST

Read Replicas

• Route all SELECT queries to replicas
• Route all writes to primary
• Account for replication lag (typically 10-100ms)
• Never read-after-write from a replica; use primary for consistency-critical reads
• Use connection-level routing, not query-level

# SQLAlchemy read replica routing
class RoutingSession(Session):
    def get_bind(self, mapper=None, clause=None):
        if self._flushing or self.is_modified():
            return engines["primary"]
        return engines["replica"]

Partition Strategies

Range Partitioning (time-series data)

-- PostgreSQL
CREATE TABLE events (
    id bigint GENERATED ALWAYS AS IDENTITY,
    created_at timestamptz NOT NULL,
    data jsonb
) PARTITION BY RANGE (created_at);

CREATE TABLE events_2025_q1 PARTITION OF events
    FOR VALUES FROM ('2025-01-01') TO ('2025-04-01');
CREATE TABLE events_2025_q2 PARTITION OF events
    FOR VALUES FROM ('2025-04-01') TO ('2025-07-01');

Hash Partitioning (even distribution)

CREATE TABLE sessions (
    id uuid PRIMARY KEY,
    user_id bigint NOT NULL
) PARTITION BY HASH (user_id);

CREATE TABLE sessions_0 PARTITION OF sessions FOR VALUES WITH (MODULUS 4, REMAINDER 0);
CREATE TABLE sessions_1 PARTITION OF sessions FOR VALUES WITH (MODULUS 4, REMAINDER 1);

Partition when tables exceed 50-100GB or when you need to drop old data quickly.

Query Optimization Checklist

1. Run EXPLAIN ANALYZE and read the plan
2. Check for sequential scans on tables with >10K rows
3. Verify index usage (check idx_scan in pg_stat_user_indexes)
4. Look for implicit type casts that prevent index use
5. Replace SELECT * with specific columns
6. Add LIMIT to queries that only need a subset
7. Use EXISTS instead of COUNT(*) > 0
8. Batch INSERT/UPDATE operations (500-1000 rows per batch)
9. Avoid functions on indexed columns in WHERE clauses
10. Monitor slow query log (pg: log_min_duration_statement = 100)

Dangerous Patterns

• LIKE '%term%' on unindexed columns (use full-text search instead)
• ORDER BY RANDOM() (use TABLESAMPLE or application-level randomization)
• SELECT DISTINCT masking a join problem
• Missing WHERE on UPDATE/DELETE (always verify with SELECT first)
• Long-running transactions holding locks
• Using OFFSET for deep pagination (use keyset/cursor pagination instead)