Question

我正在编写一个脚本来SELECT查询数据库并解析~33,000条记录。不幸的是，我在cursor.fetchone() / cursor.fetchall()阶段遇到了问题。

我首先尝试在光标上迭代一条记录，如下所示：

# Run through every record, extract the kanji, then query for FK and weight
printStatus("Starting weight calculations")
while True:
    # Get the next row in the cursor
    row = cursor.fetchone()
    if row == None:
        break

    # TODO: Determine if there's any kanji in row[2]

    weight = float((row[3] + row[4]))/2
    printStatus("Weight: " + str(weight))

基于printStatus的输出（它打印出时间戳加上传递给它的任何字符串），脚本花了大约1秒来处理每一行。这让我相信每次循环迭代时都会重新运行查询（使用LIMIT 1或其他东西），因为在SQLiteStudio [i]和[/]中运行一次相同的查询需要大约1秒钟我]返回所有33,000行。我计算出，按照这个速度，通过所有33,000条记录大约需要7个小时。

我尝试使用cursor.fetchall（）来代替：

results = cursor.fetchall()

# Run through every record, extract the kanji, then query for FK and weight
printStatus("Starting weight calculations")
for row in results:
    # TODO: Determine if there's any kanji in row[2]

    weight = float((row[3] + row[4]))/2
    printStatus("Weight: " + str(weight))

不幸的是，Python可执行文件在到达cursor.fetchall()行时锁定了25％的CPU和大约6MB的RAM。我让脚本运行了大约10分钟，但什么都没发生。

是否有大约33,000个返回的行（大约5MB的数据）太多，Python无法一次抓取？我一次又一次地迭代了吗？或者我能做些什么来加快速度？

编辑：这是一些控制台输出

12:56:26.019: Adding new column 'weight' and related index to r_ele
12:56:26.019: Querying database
12:56:28.079: Starting weight calculations
12:56:28.079: Weight: 1.0
12:56:28.079: Weight: 0.5
12:56:28.080: Weight: 0.5
12:56:28.338: Weight: 1.0
12:56:28.339: Weight: 3.0
12:56:28.843: Weight: 1.5
12:56:28.844: Weight: 1.0
12:56:28.844: Weight: 0.5
12:56:28.844: Weight: 0.5
12:56:28.845: Weight: 0.5
12:56:29.351: Weight: 0.5
12:56:29.855: Weight: 0.5
12:56:29.856: Weight: 1.0
12:56:30.371: Weight: 0.5
12:56:30.885: Weight: 0.5
12:56:31.146: Weight: 0.5
12:56:31.650: Weight: 1.0
12:56:32.432: Weight: 0.5
12:56:32.951: Weight: 0.5
12:56:32.951: Weight: 0.5
12:56:32.952: Weight: 1.0
12:56:33.454: Weight: 0.5
12:56:33.455: Weight: 0.5
12:56:33.455: Weight: 1.0
12:56:33.716: Weight: 0.5
12:56:33.716: Weight: 1.0

这是SQL查询：

//...snip (it wasn't the culprit)...

SQLiteStudio的EXPLAIN QUERY PLAN输出：

0   0   0   SCAN TABLE r_ele AS re USING COVERING INDEX r_ele_fk (~500000 rows)
0   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 1
1   0   0   SEARCH TABLE re_pri USING INDEX re_pri_fk (fk=?) (~10 rows)
0   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 2
2   0   0   SEARCH TABLE ke_pri USING INDEX ke_pri_fk (fk=?) (~10 rows)
2   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 3
3   0   0   SEARCH TABLE k_ele USING AUTOMATIC COVERING INDEX (value=?) (~7 rows)
3   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 4
4   0   0   SEARCH TABLE k_ele USING COVERING INDEX idx_k_ele (fk=?) (~10 rows)
0   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 5
5   0   0   SEARCH TABLE k_ele USING COVERING INDEX idx_k_ele (fk=?) (~10 rows)
0   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 6
6   0   0   SEARCH TABLE re_pri USING INDEX re_pri_fk (fk=?) (~10 rows)
0   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 7
7   0   0   SEARCH TABLE ke_pri USING INDEX ke_pri_fk (fk=?) (~10 rows)
7   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 8
8   0   0   SEARCH TABLE k_ele USING AUTOMATIC COVERING INDEX (value=?) (~7 rows)
8   0   0   EXECUTE CORRELATED SCALAR SUBQUERY 9
9   0   0   SEARCH TABLE k_ele USING COVERING INDEX idx_k_ele (fk=?) (~10 rows)

Answer 1

SQLite即时计算结果记录。 fetchone很慢，因为它必须为r_ele中的每条记录执行所有子查询。 fetchall甚至更慢，因为它只需要为所有记录执行fetchone。

SQLite 3.7.13估计value列上的所有查找都会非常慢，因此会为此查询创建一个临时索引。您应该创建一个永久索引，以便SQLite 3.6.21：

使用它

CREATE INDEX idx_k_ele_value ON k_ele(value);

如果这样做没有帮助，请使用较新的SQLite版本更新到Python，或使用内置较新的SQLite版本的其他数据库库，例如APSW。

Python在fetchone上运行缓慢，挂起在fetchall上

1 个答案: