EnableBatchedInserts strips backtick quoting from column names containing dots

[o-shevchenko]

Description

When EnableBatchedInserts=1 is set as a JDBC connection property, the Databricks JDBC driver strips backtick quoting from column names during its internal SQL reconstruction in PreparedStatementBatchExecutor.executeBatchedInsert(). This causes PARSE_SYNTAX_ERROR for column names containing dots (e.g., col.name), as the unquoted dot is interpreted as a schema/table separator.

Steps to Reproduce

1. Connect to Databricks using JDBC with EnableBatchedInserts=1
2. Create a table with column names containing dots (e.g., col.name)
3. Use PreparedStatement.addBatch() + executeBatch() to insert data into the table

Expected Behavior

The driver should preserve backtick quoting around column names when reconstructing the batched INSERT SQL:

INSERT INTO `table` (`col.name`, `col2`) VALUES (?, ?), (?, ?), ...

Actual Behavior

The driver strips backtick quoting from column names during SQL reconstruction:

INSERT INTO `table` (col.name, col2) VALUES (?, ?), (?, ?), ...

This results in:

[PARSE_SYNTAX_ERROR] Syntax error at or near 'name'. SQLSTATE: 42601

Environment

• Databricks JDBC Driver v3.x
• Connection properties: EnableBatchedInserts=1, BatchInsertSize=10000
• jOOQ with SQLDialect.DATABRICKS (generates correct backtick-quoted SQL)

Workaround

Currently, the only workaround is to disable EnableBatchedInserts, which results in ~10x slower write performance (e.g., 141s vs 12s for 100 records).

Additional Context

The issue is in the driver's PreparedStatementBatchExecutor.executeBatchedInsert() method, which parses the original INSERT statement, extracts column names (stripping backticks), and reconstructs a multi-row INSERT without re-quoting identifiers that require it.

This affects any column name containing special characters that require quoting (dots, spaces, reserved words, etc.).

[o-shevchenko]

Root Cause Analysis

Found the exact source of the bug in InsertStatementParser.java:

1. parseColumns() strips backticks:

private static List<String> parseColumns(String columnsStr) {
    return Arrays.stream(columnsStr.split(","))
        .map(String::trim)
        .map(col -> col.replaceAll("^`|`$", "")) // Remove backticks if present
        .filter(col -> !col.isEmpty())
        .collect(Collectors.toList());
}

This strips backticks from `col.name` → col.name

2. generateMultiRowInsert() doesn't re-quote:

sql.append("INSERT INTO ")
    .append(insertInfo.getTableName())
    .append(" (")
    .append(String.join(", ", insertInfo.getColumns()))  // bare names, no backticks!
    .append(") VALUES ");

Suggested Fix

Re-quote column names with backticks in generateMultiRowInsert():

String columnList = insertInfo.getColumns().stream()
    .map(col -> "`" + col + "`")
    .collect(Collectors.joining(", "));
sql.append("INSERT INTO ")
    .append(insertInfo.getTableName())
    .append(" (")
    .append(columnList)
    .append(") VALUES ");

Or alternatively, preserve the original quoting in parseColumns() instead of stripping it.

[o-shevchenko]

I've opened a PR #1285
@samikshya-db could you please take a look?
It will be great to include it to upcoming 3.3.1 release, since it unblocks BatchedInserts feature that we can't use because of this bug today.
BatchedInserts improves performance dramatically for us, but we can't use it for our tables as is.

[vikrantpuppala]

@o-shevchenko thanks for filing the issue and PR, will get this in. However, we've completed internal testing for the 3.3.1 release and will need to instead prioritize a follow up release soon to include this bug fix.

[github-actions]

This issue has been marked as Stale because it has been open for 30 days with no activity. If you would like the issue to remain open, please remove the stale label or comment on the issue.