[Bug] RemoteTableQuery keySerializer should not include partition keys

org.apache.paimon.flink.query.RemoteTableQuery.lookup から開始し、InternalRowSerializer.toBinaryRow の使用箇所を追跡します。パーティションの型とトリミングされたキーの型が異なるパーティション分割された主キー・テーブルに対するリモートルックアップ結合を再現し、その後、シリアライゼーションがパーティションフィールドを読み取らなくなり、ClassCastException が解消されたことを確認します。

Search before asking

• I searched in the issues and found nothing similar.

Paimon version

When doing a lookup join to the remote query service (for primary key tables), we pass the partition keys, a bucket number, and a row to join to the existing table. Right now, serialization of the row fails because the serializer expected by the remote query service client assumes that the row includes partition fields, but we do not pass them (this is the same in LocalTableQuery, not a bug).

Let's change the key serializer here to avoid looking for partition keys.

Compute Engine

Flink

Minimal reproduce step

Use the remote query service to perform a lookup join from a flink row to a paimon table. The table must be partitioned, and the partitioned key and trimmed key must have different types (e.g. int vs. string). I'll include a test case in the PR that trips it up.

What doesn't meet your expectations?

java.lang.ClassCastException: class org.apache.paimon.data.BinaryString cannot be cast to class java.lang.Integer                                                                                   
        at org.apache.paimon.data.GenericRow.getInt(GenericRow.java:152)
        at org.apache.paimon.data.serializer.InternalRowSerializer.toBinaryRow(InternalRowSerializer.java:123)
        at org.apache.paimon.flink.query.RemoteTableQuery.lookup(RemoteTableQuery.java:73)

Anything else?

No response

Are you willing to submit a PR?

• I'm willing to submit a PR!