Discussion: make DataFile Serializable && Deserializable

[ZENOTME]

Context

Make Datafile Serializable && Deserializable is useful, e.g. In distributed compute engine, it will create multiple writers in multiple machines and write the data in parallel and get the DataFile as the results, these DataFiles will be sent to a coordinator and append using transaction. In this case, DataFile should able to be Serializable && Deserializable.

Solution

For now, we support Serialize DataFile in _serde module and we should convert the DataFile to _serde::DataFile first, the interface looks like: pub fn try_from(value: super::DataFile, partition_type: &StructType,is_version_1: bool) -> _serde::DataFile. More detail:

iceberg-rust/crates/iceberg/src/spec/manifest.rs

Line 1361 in 98cd34d

pub fn try_into(

.

There is something we need to resolve to support Datafile Serializable && Deserializable:

1. The related interface needs to be exposed to the public
2. The interface is not friendly. If the DataFile can be self-contain, things will be easier, e.g. DataFile itself can be Serialize && Deserialize, the user doesn't need to convert it using an interface like pub fn try_from(value: super::DataFile, partition_type: &StructType,is_version_1: bool) -> _serde::DataFile

To solve the above, I think there are two solutions:

1. Make DateFile self-contain, store the partition type and version in DataFile directly so that it converts into _serde::DataFile directly and it can be Serialize && Deserialize.
2. Provide something like

struct SerializableDataFile {
  version: i32,
  partition_type: StructType
  data_file: DataFile
}

I prefer solution 1 because it looks more natural. Welcome to different opinions and solutions. cc @liurenjie1024 @Fokko @Xuanwo @c-thiel

[Fokko]

Hey @ZENOTME thanks for raising this.

Technically the Datafile is already serializable, you can encode it into Iceberg Avro :) I know how this works in Java and Python, but less known in Rust. I would expect the writer to have a copy of the Table, which contains the current partition information, would that help?

[ZENOTME]

Hey @ZENOTME thanks for raising this.

Technically the Datafile is already serializable, you can encode it into Iceberg Avro :) I know how this works in Java and Python, but less known in Rust. I would expect the writer to have a copy of the Table, which contains the current partition information, would that help?

Thanks @Fokko. I think this is the solution that works. But we should expose the _serde::DataFile so that the user can use this serializable representation. I create the PR for this #797.

[Xuanwo]

Hi, thank you @ZENOTME for starting this discussion. I prefer to make DataFile itself serializable. Maybe we can do this samething for DataFile like TableMetadata: TableMetadata itself is Serialize but we won't expose the underlying types.

[ZENOTME]

Hi, thank you @ZENOTME for starting this discussion. I prefer to make DataFile itself serializable. Maybe we can do this samething for DataFile like TableMetadata: TableMetadata itself is Serialize but we won't expose the underlying types.

To make DataFile itself serializable, I think we should contain the type info of DataFile in it.

Also I think contain type info in DataFile can also help #777.

The partition in DataFile should include types to facilitate validation. e.g. the field name and field id

I think that's a great thing to do anyway. It isn't super expensive, and will avoid folks bricking their table. Preferably by field-ID for both V1 and V2, otherwise order for V1, and field-IDs for V2.

There are two choices: contain the partition type info in DataFile or contain the whole struct type of DataFile. The reason I prefer to contain the whole struct type of DataFile is the better compatibility, e.g in the future the DataFile add more struct types.