获取行数 / 文档数（快速）

Oracle

-- Statistics-based (fast, estimated):
SELECT num_rows FROM dba_tables WHERE table_name = 'CUSTOMER';
-- Exact: SELECT COUNT(*) FROM customer;

num_rows 反映最近一次 ANALYZE/DBMS_STATS 的结果；需要精确值请使用 COUNT(*)。

PostgreSQL

-- Estimated (instant):
SELECT reltuples::bigint FROM pg_class WHERE relname = 'customer';
-- Exact: SELECT count(*) FROM customer;

reltuples 是截至最近一次 ANALYZE/VACUUM 的估算值；在大表上执行完整 COUNT 代价很高。

MySQL

-- Estimated (InnoDB):
SELECT table_rows FROM information_schema.tables
WHERE table_name = 'customer';
-- Exact: SELECT COUNT(*) FROM customer;

在 InnoDB 中，table_rows 是估算值，误差可达 50%。

SQL Server

SELECT SUM(row_count) AS rows
FROM sys.dm_db_partition_stats
WHERE object_id = OBJECT_ID('dbo.customer')
  AND index_id IN (0, 1);

index_id 0（堆）或 1（聚集）给出行数；比 COUNT(*) 扫描快得多。

MongoDB

// Fast (from metadata):
db.customer.estimatedDocumentCount()
// Exact (can count with a filter): db.customer.countDocuments({})

estimatedDocumentCount 读取元数据（即时）；countDocuments 执行实际扫描。

Redis

DBSIZE
# Specific pattern: count roughly with SCAN (do not use KEYS * in production)

DBSIZE 以 O(1) 返回所选数据库中的键总数。