Support delete data files in fast append action

[mnpw]

Is your feature request related to a problem or challenge?

The transaction API exposes FastAppendAction for making commits to catalog.

Equality delete writer was added with #703 to support writing equality delete data files. However, FastAppendAction does not support committing equality delete data files.

See DataContentType check – https://github.com/apache/iceberg-rust/blob/main/crates/iceberg/src/transaction.rs#L443-L447

Describe the solution you'd like

It would be great if we can enhance FastAppendAction to support committing equality delete data files as well. I am willing to work on this.

Willingness to contribute

I would be willing to contribute to this feature with guidance from the Iceberg Rust community

[jonathanc-n]

@mnpw take a look at #1017 before proceeding.

[liurenjie1024]

Thanks @mnpw for raising this. But this should not be included in fast append action since deletion typically requires conflict detection.

[liurenjie1024]

We should close this issue as it's not following iceberg's design.

[mnpw]

But this should not be included in fast append action since deletion typically requires conflict detection.

@liurenjie1024 What do you think about a transaction action for only delete files, perhaps RowDeleteAction just like FastAppendAction? RowDeleteAction can take DataContentType::EqualityDeletes files and commit them into a new snapshot. On conflict during snapshot creation this action can behave similarly to FastAppendAction.

My use-case is being able to write delete files and commit them into a new snapshot. Please let me know if there is any other better way for the same.

[liurenjie1024]

On conflict during snapshot creation this action can behave similarly to FastAppendAction.

Sorry, I don't get this point. FastAppendAction will never conflict with other transactions. By conflict detection I mean ensuring snapshot isolation during concurrent write: https://iceberg.apache.org/docs/nightly/reliability/#concurrent-write-operations

My use-case is being able to write delete files and commit them into a new snapshot. Please let me know if there is any other better way for the same.

Do you mean to write delete files only? I can understand such case, but we still need to do conflict detection for concurrent writes.

[ZENOTME]

I think the intent of #798 is similar to this issue. We end up needing to implement RowDeltaAction for this intent.

@liurenjie1024 What do you think about a transaction action for only delete files, perhaps RowDeleteAction just like FastAppendAction? RowDeleteAction can take DataContentType::EqualityDeletes files and commit them into a new snapshot. On conflict during snapshot creation this action can behave similarly to FastAppendAction.

For the action of only deleting files, there may be concurrent new data append(or overwrite behavior) between them. These deleted files will affect the new append data and cause undefined behavior. So looks like we can't avoid the conflict detection. We can open the issue to track this.

My use-case is being able to write delete files and commit them into a new snapshot. Please let me know if there is any other better way for the same.

Before we complete RowDeleteAction, personally I think maybe you can try to hack the fast append to append the deleted data file as #798.(If you just want to try some simple case) It only works in simple cases (e.g. no concurrency write) but doesn't mean it's right.

[ZENOTME]

Create an issue to track RowDeleteAction: #1104