Spark SQL CTE在查询中忽略命名空间

Question

在本地 spark-sql 或通过 pyspark spark.sql(...) 运行 Spark，如果我在查询中使用 CTE，然后使用不正确的命名空间/数据库引用 CTE，则查询工作正常（意外）。当我在生产环境中（在 Databricks 上）运行查询时，我得到一个 Table or view not found 错误（预期）。

意外通过行为可以通过spark-sql在本地重现：

WITH myview AS (
  SELECT 1 AS column
)

SELECT
  *
FROM
  invalid_namespace.myview;

当我预计它失败时返回“1”。

有人可以帮我在本地解决这个问题，以便我们在部署之前进行适当的测试吗？

从终端重现的具体步骤：

$ spark-sql
...
spark-sql> WITH some_new_cte AS (SELECT 1 AS column)
         > SELECT * FROM namespace_does_not_exist.some_new_cte;
...
1
Time taken: 2.294 seconds, Fetched 1 row(s)
spark-sql>

Answer 1

如果看查询计划，其实解析失败

== Parsed Logical Plan ==
CTE [myview]
:  +- SubqueryAlias `myview`
:     +- Project [1 AS column#0]
:        +- OneRowRelation
+- 'Project [*]
   +- 'UnresolvedRelation `invalid_namespace`.`myview`

== Analyzed Logical Plan ==
column: int
Project [column#0]
+- SubqueryAlias `myview`
   +- Project [1 AS column#0]
      +- OneRowRelation

== Optimized Logical Plan ==
Project [1 AS column#0]
+- OneRowRelation

== Physical Plan ==
*(1) Project [1 AS column#0]
+- Scan OneRowRelation[]

您的查询返回“1”的原因是因为 spark 看到您的视图在同一个查询中，所以它只是忽略了您的命名空间。如果命名空间真的不存在，就会失败。

Answer 2

这似乎是一个影响 Spark 2.4.0 到 2.4.5 版本的错误（我没有检查

2.4.0 意外成功：

$ spark-submit --version
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.4.0
      /_/

Using Scala version 2.11.12, OpenJDK 64-Bit Server VM, 1.8.0_262
Branch
Compiled by user  on 2018-10-29T06:22:05Z
Revision
Url
Type --help for more information.

$ echo "WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;" | SPARK_CONF_DIR=spark_conf spark-sql
20/07/17 10:59:05 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Spark master: local[*], Application Id: local-1595008748636
spark-sql> WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;
1
Time taken: 1.822 seconds, Fetched 1 row(s)
spark-sql>

已测试但未显示：2.4.1、2.4.2、2.4.3、2.4.4（均意外成功）

2.4.5 意外成功：

$ spark-submit --version
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.4.5
      /_/

Using Scala version 2.11.12, OpenJDK 64-Bit Server VM, 1.8.0_262
Branch HEAD
Compiled by user centos on 2020-02-02T19:38:06Z
Revision cee4ecbb16917fa85f02c635925e2687400aa56b
Url https://gitbox.apache.org/repos/asf/spark.git
Type --help for more information.

$ echo "WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;" | SPARK_CONF_DIR=spark_conf spark-sql
20/07/17 10:59:55 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Spark master: local[*], Application Id: local-1595008798737
spark-sql> WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;
1
Time taken: 2.155 seconds, Fetched 1 row(s)
spark-sql>

2.4.6 中的预期失败：

$ spark-submit --version
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.4.6
      /_/

Using Scala version 2.11.12, OpenJDK 64-Bit Server VM, 1.8.0_262
Branch HEAD
Compiled by user holden on 2020-05-29T23:47:51Z
Revision 807e0a484d1de767d1f02bd8a622da6450bdf940
Url https://gitbox.apache.org/repos/asf/spark.git
Type --help for more information.

$ echo "WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;" | SPARK_CONF_DIR=spark_conf spark-sql
20/07/17 11:00:40 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Spark master: local[*], Application Id: local-1595008843321
spark-sql> WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;
20/07/17 11:00:44 WARN ObjectStore: Failed to get database global_temp, returning NoSuchObjectException
20/07/17 11:00:44 WARN ObjectStore: Failed to get database bad_namespace, returning NoSuchObjectException
Error in query: Table or view not found: `bad_namespace`.`mycte`; line 1 pos 49;
'Project [*]
+- 'UnresolvedRelation `bad_namespace`.`mycte`

spark-sql>

3.0.0 中的预期失败：

$ spark-submit --version
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 3.0.0
      /_/

Using Scala version 2.12.10, OpenJDK 64-Bit Server VM, 1.8.0_262
Branch HEAD
Compiled by user ubuntu on 2020-06-06T11:32:25Z
Revision 3fdfce3120f307147244e5eaf46d61419a723d50
Url https://gitbox.apache.org/repos/asf/spark.git
Type --help for more information.

$ echo "WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;" | SPARK_CONF_DIR=spark_conf spark-sql
20/07/17 11:01:40 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
20/07/17 11:01:42 WARN HiveConf: HiveConf of name hive.stats.jdbc.timeout does not exist
20/07/17 11:01:42 WARN HiveConf: HiveConf of name hive.stats.retries.wait does not exist
20/07/17 11:01:44 ERROR ObjectStore: Version information found in metastore differs 1.2.0 from expected schema version 2.3.0. Schema verififcation is disabled hive.metastore.schema.verification
20/07/17 11:01:44 WARN ObjectStore: setMetaStoreSchemaVersion called but recording version is disabled: version = 2.3.0, comment = Set by MetaStore georgeleslie-waksman@10.0.1.178
Spark master: local[*], Application Id: local-1595008901594
spark-sql> WITH mycte AS (SELECT 1 AS column) SELECT * FROM bad_namespace.mycte;
20/07/17 11:01:45 WARN ObjectStore: Failed to get database global_temp, returning NoSuchObjectException
20/07/17 11:01:45 WARN ObjectStore: Failed to get database bad_namespace, returning NoSuchObjectException
Error in query: Table or view not found: bad_namespace.mycte; line 1 pos 49;
'Project [*]
+- 'UnresolvedRelation [bad_namespace, mycte]

spark-sql>