Bug: Seg fault in copy from statement

[acquamarin]

Kùzu version

master

What operating system are you using?

No response

What happened?

The COPY FROM statement seg faults in kuzu:

kuzu ~/tmp/kz2
Opening the database at path: /Users/bgoosman/tmp/kz2 in read-write mode.
Enter ":help" for usage hints.
kuzu> CREATE NODE TABLE Entity(category string, label string, PRIMARY KEY (label));
┌────────────────────────────────┐
│ result                         │
│ STRING                         │
├────────────────────────────────┤
│ Table Entity has been created. │
└────────────────────────────────┘
(1 tuple)
(1 column)
Time: 0.35ms (compiling), 2.24ms (executing)
kuzu> CREATE REL TABLE RELATED_TO(FROM Entity TO Entity, source_id int64);
┌────────────────────────────────────┐
│ result                             │
│ STRING                             │
├────────────────────────────────────┤
│ Table RELATED_TO has been created. │
└────────────────────────────────────┘
(1 tuple)
(1 column)
Time: 0.05ms (compiling), 4.42ms (executing)
kuzu> COPY Entity FROM "/Users/bgoosman/kineviz/sightxr_api_ex/_build/dev/lib/sightxr_api/priv/static/parquet/2024-12-09T21:20:51.913417Z/Entity.parquet";
┌───────────────────────────────────────────────────┐
│ result                                            │
│ STRING                                            │
├───────────────────────────────────────────────────┤
│ 1059 tuples have been copied to the Entity table. │
└───────────────────────────────────────────────────┘
(1 tuple)
(1 column)
Time: 5.48ms (compiling), 106.60ms (executing)
kuzu> COPY RELATED_TO FROM "/Users/bgoosman/kineviz/sightxr_api_ex/_build/dev/lib/sightxr_api/priv/static/parquet/2024-12-09T21:20:51.913417Z/RELATED_TO.parquet";
Pipelines Finished: 0/4
Current Pipeline Progress: 100%
[1]    70205 segmentation fault  kuzu ~/tmp/kz2

However, the load from works:

LOAD FROM "/Users/bgoosman/kineviz/sightxr_api_ex/_build/dev/lib/sightxr_api/priv/static/parquet/2024-12-09T21:20:51.913417Z/RELATED_TO.parquet" return *

Are there known steps to reproduce?

No response

[ray6080]

The seg fault is within index lookup during copy rel. It is due to mismatched data types between the input parquet file and table schema. More specifically, the table schema expects input file columns to be (FROM STRING, TO STRING, source_id INT64], while the input parquet file is (FROM STRING, source_id INT64, TO STRING). This leads to incorrect type parsing in index lookup, which leads to seg fault.

There are several things we should do to properly address issues similar to this one:

• Binder should check data types and cast if necessary on primary key columns. We should expect binder to throw error or add implicit casting in this case.
• A better mechanism for users to provide FROM and TO columns during COPY. e.g., COPY rel(_from, source_id, _to) FROM 'test.paruqet'
• Optionally, index lookup or hash index should protect against this kind of behaviour.