Question

我的数据如下：

select * from test;

 department | employee | batch_number | hash
------------+----------+--------------+-------
 dep1       | Bart     |            1 | hash1
 dep1       | Bart     |            1 | hash2
 dep1       | Lisa     |            3 | hash3
 dep1       | Lisa     |            4 | hash4
 dep1       | John     |            5 | hash5
 dep1       | Lucy     |            6 | hash6
 dep1       | Bart     |            7 | hash7
 dep1       | Bart     |            7 | hash8

我想用where上的batch_number子句，ordering上的batch_number和{{1}上的in谓词来查询数据}}。

在关系数据库中，这看起来像

employee

因为我想在hash上订购，所以我尝试将其作为群集密钥：

batch_number

但是，这不允许在我的索引上带有CREATE TABLE keyspace.test( department TEXT, batch_number INT, hash TEXT, employee TEXT, PRIMARY KEY ((department), batch_number, hash) ) WITH CLUSTERING ORDER BY (batch_number DESC); CREATE INDEX tst_emp ON keyspace.test (employee);谓词的查询：

in

所以我也尝试将select * from keyspace.test where department='dep1' and employee in ('Bart','Lucy','John'); InvalidRequest: Error from server: code=2200 [Invalid query] message="IN predicates on non-primary-key columns (employee) is not yet supported"列添加为集群键：

employee

但这失败了，因为我不能在CREATE TABLE keyspace.test( department TEXT, batch_number INT, hash TEXT, employee TEXT, PRIMARY KEY ((department), batch_number, hash, employee) ) WITH CLUSTERING ORDER BY (batch_number DESC);上放置非EQ关系：

batch_number

但是，每当我将select * from keyspace.test where department='dep1' and batch_number > 1 and employee in ('Bart','Lucy','John'); InvalidRequest: Error from server: code=2200 [Invalid query] message="Clustering column "employee" cannot be restricted (preceding column "batch_number" is restricted by a non-EQ relation)"放在employee之前，我都会失去订购batch_number的能力：

batch_number

那么哪种表设计可以进行这样的查询？可以在Cassandra中完成吗？

编辑：

我希望能够在此表上运行的其他查询是：

CREATE TABLE keyspace.test(
    department      TEXT,
    employee        TEXT,
    batch_number    INT,
    hash            TEXT,
    PRIMARY KEY ((department), employee, batch_number, hash)
);

select * from keyspace.test where department='dep1' and employee in ('Bart','Lucy','John') ORDER BY batch_number DESC;
InvalidRequest: Error from server: code=2200 [Invalid query] message="Order by currently only support the ordering of columns following their declared order in the PRIMARY KEY"

和

select * from keyspace.test where department='X' and batch_number=Y

Answer 1

使用实例化视图，您可以重新排列数据：

CREATE MATERIALIZED VIEW mv_test AS 
SELECT
   department,
   batch_number,
   employee,
   hash 
FROM
   test 
WHERE
   department IS NOT NULL 
   AND batch_number IS NOT NULL 
   AND employee IS NOT NULL 
   AND hash IS NOT NULL 
PRIMARY KEY (department, employee, batch_number, hash) 
WITH clustering 
ORDER BY
(batch_number DESC);

我可以执行以下查询：

SELECT * FROM mv_test 
WHERE
   department = 'dep1' 
   AND employee IN 
   (
      'Bart',
      'Lisa'
   )
   AND batch_number > 3;

根据聚类顺序对结果进行排序：

 department | employee | batch_number | hash
------------+----------+--------------+-------
       dep1 |     Bart |            7 | hash7
       dep1 |     Bart |            7 | hash8
       dep1 |     Lisa |            4 | hash4

尽管>子句是非等号子句，但是IN虽然具有多个值，但仍具有确定性，这就是为什么我相信您可以毫无问题地过滤键的原因。由于batch_number是您要过滤的最后一件事，因此允许使用任何类型的where子句。我假设您一直有department。

请注意，实例化视图impact performance。更具体地说，写性能。但是，与ALLOW FILTERING相比，读取性能是有益的。

更新：

在物化视图末尾指定的顺序为batch_number，但是，它将首先在department上排序，然后依次在employee和batch_number上排序，因此不能保证batch_number的顺序。据我所知，这是没有办法的。另一种数据库解决方案可能更可取。

更新2：

如Apache邮件链中所述（请参阅下面的评论），实例化视图不被视为可用于生产。但是，Datastax认为它们是可用的，但要注意使用上述最佳做法。就个人而言，我对物化视图没有任何麻烦。当然，这是一个简单的单个数据中心集群，并且考虑到最佳实践提到了更复杂的设置，因此在这种情况下它们可能会崩溃。

Answer 2

如果需要，您可以在employee上使用索引，甚至可以将其从主键中删除。您可能需要放弃使用IN，但可以拆分查询并将结果加入客户端。

CREATE TABLE tk.test_good(
    department      TEXT,
    batch_number    INT,
    employee        TEXT,
    hash            TEXT,
    PRIMARY KEY ((department), batch_number, hash)
)WITH CLUSTERING ORDER BY (batch_number DESC);

CREATE INDEX IF NOT EXISTS employee_idx ON tk.test_good ( employee );

select * from tk.test_good where department='dep1' and employee='Bart' and batch_number >= 2 limit 3;
select * from tk.test_good where department='dep1' and employee='Lucy' and batch_number >= 2 limit 3;
select * from tk.test_good where department='dep1' and employee='John' and batch_number >= 2 limit 3;

此方法的缺点是索引可能太大。但是我不知道数据池的大小，因此请您自行考虑。

Cassandra表设计用于使用ORDER，LIMIT和IN谓词进行查询

2 个答案: