[Java][FlightRPC] Flight SQL JDBC driver does not expose per-batch app_metadata from FlightStream
- 主要言語
- Java
- スター
- 94
- フォーク
- 152
- 平均マージ
- 3日 16時間
- マージ済み PR(30日)
- 11
説明
### Summary
The Flight SQL JDBC driver (flight-sql-jdbc-driver) currently exposes FlightInfo.app_metadata via ArrowFlightJdbcFlightStreamResultSet.getAppMetadata() (added in ~15.0), but per-batch app_metadata from FlightStream.getLatestMetadata() is received internally and silently discarded — it is never surfaced to JDBC callers.
### Motivation
Servers use per-batch app_metadata on FlightData messages to communicate execution statistics (rows scanned, elapsed time, query state, etc.) — similar to what Trino exposes via QueryStats. This is particularly valuable for the final batch, which servers commonly use as a "trailing metadata" pattern to report post-execution stats.
Currently there is no way to access this information through the JDBC layer without dropping down to the raw FlightClient, which defeats the purpose of the JDBC driver abstraction.
### Current behavior
FlightInfo.app_metadata → accessible via rs.unwrap(ArrowFlightJdbcFlightStreamResultSet.class).getAppMetadata() ✅
Per-batch FlightData.app_metadata → received by FlightStream.getLatestMetadata() internally, but never stored or exposed ❌
**Also worth noting**: getAppMetadata() returns a snapshot from query submission time only. For long-running queries using PollFlightInfo, updated app_metadata from subsequent poll cycles is never surfaced. This limits its usefulness for engines that progressively update stats during execution.
### Requested change
Add a method to ArrowFlightJdbcFlightStreamResultSet (accessible via unwrap()) such as:
```
public ArrowBuf getLatestBatchMetadata() {
if (currentEndpointData == null || currentEndpointData.getStream() == null) {
return null;
}
return currentEndpointData.getStream().getLatestMetadata();
}
```
This would follow the same unwrap() pattern already established for getAppMetadata() and give callers access to per-batch metadata, including trailing stats on the final batch.
### Component(s)
FlightRPC, Java
コントリビューションガイド
調査の方向性
ArrowFlightJdbcFlightStreamResultSet と既存の getAppMetadata() メソッドから始め、currentEndpointData と FlightStream.getLatestMetadata() を追跡します。完了の条件は、JDBC の呼び出し元が unwrap() を使用してバッチごとの最新メタデータ(最後のバッチの末尾メタデータを含む)を取得でき、raw の FlightClient にアクセスする必要がないことです。
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- java, sql
- 領域
- api, database
- issue の種類
- 機能追加
- 難易度
- 3/5
- 見積もり時間
- 1〜2日
- 活発さ
- 静か
- 明瞭さ
- 明確に書かれている
- 初心者へのやさしさ
- 75/100