Remove support for legacy driver and upgrade to 0.4.1 in ClickHouse connector

[ebyhr]

Description

• Remove support for legacy driver in ClickHouse since it took more than 1 year after becoming the deprecation
• Upgrade ClickHouse JDBC driver to 0.4.1. Even the latest driver doesn't provide ResultSetMetaData for query pass-through table function

Fixes #16188

Release notes

(x) Release notes are required, with the following suggested text:

# ClickHouse
* Remove support for legacy JDBC driver. ({issue}`16193`)

[zhicwu]

Hi @ebyhr, can we bump to 0.4.1? Although it's just out for 2 days, but it fixed a few critical bugs found in 0.4.0 like 1) incorrect nested array value; 2) broken serde for Nested data type which may lead to failures when deserializing Nested and JSON values; and 3) bitmap64 support.

Apart from that, I think there's one breaking change worthy of mention here. Starting from 0.4.0, to support binary string(mapping String and FixedString to byte array), you have to explicitly set use_binary_string to true. This is because 1) ~10% performance improvement for deserializing each String column and less memory footprint when use_binary_string=false, which is default; and 2) binary data can be also represented as Array(*Int8) in ClickHouse.

[ebyhr]

@zhicwu Thanks for sharing the information! Changed to 0.4.1 and enabled use_binary_string property.

[ebyhr]

web-ui-checks hit #16179

[zhicwu]

@zhicwu Thanks for sharing the information! Changed to 0.4.1 and enabled use_binary_string property.

Thank you and apologize for additional changes at your end. Does it make sense to you to map VARBINARY to Array(UInt8) in ClickHouse?

[ebyhr]

@zhicwu What about read logic? Reading Array(UInt8) as VARBINARY looks weird to me.

[zhicwu]

@zhicwu What about read logic? Reading Array(UInt8) as VARBINARY looks weird to me.

I'm not joking, but is Array(BYTE) better? BYTE is an alias of Int8.

Looks like the mapped type in ClickHouse doesn't matter when reading/writing varbinary in trino, according to code below:

trino/plugin/trino-base-jdbc/src/main/java/io/trino/plugin/jdbc/StandardColumnMappings.java

Lines 348 to 356 in fba39d6

	public static SliceReadFunction varbinaryReadFunction()
	{
	return (resultSet, columnIndex) -> wrappedBuffer(resultSet.getBytes(columnIndex));
	}
	public static SliceWriteFunction varbinaryWriteFunction()
	{
	return SliceWriteFunction.of(Types.VARBINARY, (statement, index, value) -> statement.setBytes(index, value.getBytes()));
	}

[ebyhr]

@zhicwu Mapping to Array(UInt8) or Array(BYTE) changes the behavior of NULL handling as far as I confirmed. I will keep as-is in this PR. Please feel free to send a new one after this PR.

[ebyhr]

Addressed comments.

[Praveen2112]

Do we have this API in the legacy version ?

[Praveen2112]

In this case can we squash with the next commit - I'm happy with both the cases.

[ebyhr]

It existed, but the package was changed in the new version.