apache / apache/arrow-java

flight-sql-jdbc-driver: add option to rely on Arrow Flight SQL Bulk Ingestion for batched inserts

未关闭
#931 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
Type: enhancement
主要语言
Java
星标
94
派生
152
平均合并
3 天 16 小时
30 天内合并 PR
11

描述

### Describe the enhancement requested

Arrow Flight SQL has a feature for ingesting massive datasets, Bulk Ingestion: https://github.com/apache/arrow/issues/38255
It would be beneficial to use those special RPC methods for batched prepared statement calls when the prepared statement is strictly for inserting data.
E.g. when Spark is used for writing data, it generates a simple SQL query like `"INSERT INTO table(field1, field2, ...) VALUES (?, ?, ...)"`, creates a prepared statement, and then uses the prepared statement update RPC method to insert the rows of the dataset. If this feature is implemented, it would be possible for the driver to instead use the `DoPut(CommandStatementIngest)` command associated with the Bulk Ingestion feature instead.

There are Arrow Flight SQL server implementations that work like this: when a `DoAction(ActionCreatePreparedStatementRequest)` is executed, the server creates up to two versions of the data structure underlying the instance of the PreparedStatement. One is a handle to a full-scale query engine execution procedure (e.g. DataFusion's logical plan), and another is a handle to a very simple procedure that just stores the received record batches in the storage - of course, the second procedure is only possible to be generated when the query is of a certain form; like the one used by Spark. The point is that this simple procedure also works without overheads associated with the more general interface of prepared statement API - for example, it does not need to do a transposition of PreparedStatement parameters into record batches.
I think that it should be possible to move this logic for deciding to use Bulk Ingestion into the jdbc driver.

Usecase for this integration is this: developers of Arrow Flight SQL servers could implement bulk ingestion command handlers and avoid implementing special logic for handling batched inserts in a special manner. Then the client would use this newly introduced driver option to allow the driver to decide to use the bulk ingestion RPC methods for inserting data.

贡献指南

打开贡献指南

调研方向

首先检查 JDBC 驱动程序的 prepared statement 更新路径和 Flight SQL Bulk Ingestion RPC,尤其是 DoPut(CommandStatementIngest) 和 ActionCreatePreparedStatementRequest。定义一个选项如何识别符合条件的仅插入批处理语句,并通过 bulk ingestion 对其进行路由,同时兼容其他 prepared statement。

由索引模型根据 Issue 内容生成。

评估

技术栈
java
领域
database
Issue 类型
功能
难度
5/5
预计耗时
一周以上
活跃度
停滞
描述清晰度
需要澄清
新手友好度
35/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。