Databricks SQL ExecuteStatement Failing During Chunked Updates to a Single Cell (Large JSON Field)

从报告中描述的 SQLAlchemy 和 Databricks SQL ExecuteStatement 路径开始，然后使用一个大型 JSON 字段重现分块 UPDATE 序列。比较失败时的请求大小，并检查 connector 对重试代码 429 和 503 的处理。完成的标准是记录请求限制和一种获批准的增量更新实践。

We’re encountering issues with Databricks SQL when attempting to update a row that contains a large JSON array field. Originally, we tried inserting the entire JSON directly into the column, but this failed due to request size limitations.
To address that, we redesigned the approach to:

• Split the full JSON into smaller chunks (~100 items)
• Append each chunk incrementally to the same row/column using separate UPDATE statements
• Commit each chunk using a new thread and session to ensure SQLAlchemy thread safety

Despite chunking, the request eventually fails when the dmt_data field grows large enough (presumably ~1–2MB compressed). The SQL API returns:

pgsqlCopyEdit(databricks.sql.exc.RequestError) Error during request to server. 
ExecuteStatement command can only be retried for codes 429 and 503 

This confirms that each UPDATE's request body is still exceeding Databricks SQL's internal limits, even though we’re only appending small pieces.
What We’re Looking For:

• Confirmation of the exact request body size limit for INSERT/UPDATE operations over Databricks SQL
• Recommended practice for incrementally updating a single JSON column that grows over time