Implemented custom SQL DB registration #917
Conversation
This commit introduces an optional `dbType` parameter to various JDBC read functions. It allows for specifying the database type directly, enhancing flexibility and control over query execution. This change ensures backward compatibility by defaulting to type inference when `dbType` is not provided.
This update introduces the ability to specify custom database types in the data schema generation process. The changes include the addition of a `dbTypeClassName` field and modifications to pass `DbType` to relevant functions. This enhances flexibility in handling various database types beyond the default configurations.
This update refactors the DataSchemaGenerator to use ServiceLoader for loading DbTypeProvider and removes hard-coded class loading. It also changes the annotation parameter from dbTypeClassName to dbTypeKClass for better type safety.
|
Generated sources will be updated after merging this PR. |
| @@ -113,16 +113,18 @@ public data class DbConnectionConfig(val url: String, val user: String = "", val | |||
| * @param [tableName] the name of the table to read data from. | |||
| * @param [limit] the maximum number of rows to retrieve from the table. | |||
| * @param [inferNullability] indicates how the column nullability should be inferred. | |||
| * @param [dbType] the type of database, could be a custom object, provided by user, optional, default is `null`. | |||
There was a problem hiding this comment.
maybe specify what happens when it's null, aka, it will be inferred from dbConfig
| @@ -142,12 +145,13 @@ public fun DataFrame.Companion.readSqlTable( | |||
| tableName: String, | |||
| limit: Int = DEFAULT_LIMIT, | |||
| inferNullability: Boolean = true, | |||
| dbType: DbType? = null, | |||
There was a problem hiding this comment.
While having a dbType argument here is great for building and debugging support for new DbTypes, I do feel like it's not "complete custom support" of new datatypes.
To be fully extensible, like with DataFrame.read(), I think we should support having service-registered DbTypes. Then a user should be able to do:
implementation("org.jetbrains.kotlinx.dataframe...")
implementation("com.custom.dbtype-for-df")and DataFrame.readSqlTable(MyCustomDbConnection) should work, recognizing the custom DBType that's on the classpath.
Still, it's just an idea and having this dbType argument here is still a must for when the automatic dbtype detection fails.
| @@ -967,4 +972,127 @@ class JdbcTest { | |||
| } | |||
| exception.message shouldBe "H2 database could not be specified with H2 dialect!" | |||
| } | |||
|
|
|||
| // helper object created for API testing purposes | |||
| object CustomDB: H2(MySql) | |||
There was a problem hiding this comment.
If it's only for testing, i'd recommend to keep H2 final and use object CustomDB : DbType by H2(MySql)
|
|
||
| @Test | ||
| fun `read from table from custom database`() { | ||
| val tableName = "Customer" |
There was a problem hiding this comment.
Most of assertions test correct reading of data, but actually functionality under test is correct propagation of DbType, as i understand. So most of it is duplicated code? In any case, changes that will trigger H2 tests assertions to fail, will fail assertions here too. For this reason please consider a different approach here
This pull request introduces several enhancements to the
dataframe-jdbcmodule, primarily focusing on adding support for custom database types (DbType) in various methods. Additionally, it includes a minor change to make theH2class extensible.Also, this PR in the client application is based on 0.15-dev built on this branch
Enhancements to database type support:
dataframe-jdbc/src/main/kotlin/org/jetbrains/kotlinx/dataframe/io/readJdbc.kt:dbTypeparameter to multiple methods, allowing custom database type objects to be passed. This parameter defaults tonulland, if not provided, the database type is determined from the connection.Minor changes:
dataframe-jdbc/src/main/kotlin/org/jetbrains/kotlinx/dataframe/io/db/H2.kt:H2class to beopen, allowing it to be extended.For finalization, it requires