[
https://issues.apache.org/jira/browse/SPARK-60089?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-60089:
-----------------------------------
Labels: pull-request-available (was: )
> Surface remote JDBC insufficient-privilege errors as a dedicated error class
> ----------------------------------------------------------------------------
>
> Key: SPARK-60089
> URL: https://issues.apache.org/jira/browse/SPARK-60089
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 4.2.0
> Reporter: Milos Stojanovic
> Priority: Major
> Labels: pull-request-available
>
> When \{{JDBCRDD.resolveTable}} resolves the output schema of a JDBC read, it
> classifies driver errors through dialect hooks:
> \{{isObjectNotFoundException}}, \{{isNotSelectableObjectException}} and
> \{{isSyntaxErrorBestEffort}}. The last one raises
> \{{JDBC_EXTERNAL_ENGINE_SYNTAX_ERROR}}.
> A permission denial from the remote engine has no hook of its own, so it ends
> up as a different error depending on the dialect:
> * Dialects that treat class-42 or \{{42000}} SQLSTATEs as syntax errors
> (MySQL, Oracle, Snowflake, Teradata, DB2, Databricks) report it as
> \{{JDBC_EXTERNAL_ENGINE_SYNTAX_ERROR}}. MySQL, for example, returns SQLSTATE
> \{{42000}} for both syntax errors and access-denied errors.
> * PostgreSQL has propagated \{{42501}} as an unclassified driver error since
> SPARK-59336. Through the JDBC catalog it surfaces as
> \{{FAILED_JDBC.LOAD_TABLE}}, and the driver message is only in the cause
> chain.
> In both cases callers can't tell "the user lacks privileges" apart from "the
> SQL is invalid".
> h3. Proposal
> Follow the existing syntax-error path:
> * Add \{{JdbcDialect.isInsufficientPrivilegeException(e: SQLException):
> Boolean}}, defaulting to \{{false}}.
> * Check it first in \{{JDBCRDD.resolveTable}}, before the not-found and
> syntax-error checks, and raise a new
> \{{JDBC_EXTERNAL_ENGINE_INSUFFICIENT_PRIVILEGE}} error (SQLSTATE \{{42501}}).
> The error carries the pushed-down query, the remote SQLSTATE and the remote
> message.
> * \{{AggregatedDialect}} returns true if any of its delegate dialects
> recognizes the error, so a user-registered dialect that doesn't override the
> hook can't turn off built-in detection.
> * Per-dialect detection:
> ||Dialect||Codes||
> |PostgreSQL|SQLSTATE 42501|
> |MySQL / MariaDB|1044, 1142, 1143, 1227, 1370|
> |SQL Server|229, 230, 262, 297, 300|
> |Oracle|ORA-01031|
> |Snowflake|3001|
> |Teradata|3523, 3524, 5315|
> |DB2|-551, -552|
> |Databricks|SQLSTATE 42501, or INSUFFICIENT_PERMISSIONS in the message|
> Oracle ORA-00942 and Snowflake 2003 ("does not exist or not authorized") are
> ambiguous, so they stay classified as object-not-found.
> This can't be a new subclass of \{{JDBC_EXTERNAL_ENGINE_SYNTAX_ERROR}},
> because SQLSTATE is defined per error class and that class uses \{{42000}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
