ROWID changed after CALL ducklake_rewrite_data_files

[korolmi]

ROWID is not described in docu (at least I could not find docu or blog about rowid),

so behaviour described below might be OK while it looks a bit strange.

What I did: I was curious how ROWIDs are calculated by ducklake internals, so I made some tests. I will write my thoughts about ROWIDs somewhere else more broadly, here is brief result, if I:

• insert 2 rows into table
• delete 1 row
• insert 2 more rows into table
• delete 1 of last rows
• select rowid,*
• CALL ducklake_rewrite_data_files (delete_threshold => 0.5 to make rewrite happen)
• select rowid,*

results of selects differ:

(first select)
┌───────┬───────┐
│ rowid │ id │
│ int64 │ int32 │
├───────┼───────┤
│ 1 │ 1 │
│ 3 │ 3 │
└───────┴───────┘
(call ...)
┌─────────┐
│ Success │
│ boolean │
├─────────┤
│ 0 rows │
└─────────┘
(second select)
┌───────┬───────┐
│ rowid │ id │
│ int64 │ int32 │
├───────┼───────┤
│ 0 │ 1 │
│ 1 │ 3 │
└───────┴───────┘

ROWIDs of BOTH rows of the table are different after CALL ducklake_rewrite_data_files...

Question: is it OK?

If yes then I doubt about ducklake lineage functionality.

If not - there is a bug.

Initially I did a bit more complicated test in which only one ROWID was different after CALL, which make overall picture dizzy.

Full SQL script to reproduce:

ATTACH 'ducklake:dlrowids.ducklake' AS dl;
create table dl.rows1 ( id int );
insert into dl.rows1 values (0), (1);
delete from dl.rows1 where id=0;
insert into dl.rows1 values (2), (3);
delete from dl.rows1 where id=2;
select rowid,* from dl.rows1;
CALL ducklake_rewrite_data_files('dl', delete_threshold => 0.5);
select rowid,* from dl.rows1;