Error when running OPTIMIZE FINAL

[Shlomixg]

Describe what's wrong

Our system is sensitive to duplicated data, so we use OPTIMIZE FINAL on ReplacingMergeTree tables to remove duplications.
When trying to run OPTIMIZE FINAL on 2 large tables (~2 billion tables & ~200 million) we get this error, on both tables:

Orig exception: Code: 74. DB::ErrnoException: Cannot read from file: /var/lib/clickhouse/store/c48/c487a9b6-3d95-42e8-b168-cd9c45a9758b/all_1_1134741_47/<column_name>.bin, errno: 22, strerror: Invalid argument: Cache info: Buffer path: /var/lib/clickhouse/store/c48/c487a9b6-3d95-42e8-b168-cd9c45a9758b/all_1_1134741_47/<column_name>.bin, hash key: d8713dc8fbd955852c4cac2756e46d72, file_offset_of_buffer_end: 0, internal buffer remaining read range: [0:5554177], read_type: REMOTE_FS_READ_AND_PUT_IN_CACHE, last caller: c487a9b6-3d95-42e8-b168-cd9c45a9758b::all_1_1138616_48:67, file segment info: None: (while reading column <column_name>): (while reading from part /var/lib/clickhouse/store/c48/c487a9b6-3d95-42e8-b168-cd9c45a9758b/all_1_1134741_47/ from mark 0 with max_rows_to_read = 8192): While executing MergeTreeSequentialSource. (CANNOT_READ_FROM_FILE_DESCRIPTOR) (version 23.3.1.2823 (official build))

Both tables are defined as ReplactingMergeTree, and the problematic column is the first column on both tables.
We can query the tables, insert data and delete without any problem.

We have run these commands many times in the past and never encountered this error.
We can't reproduce it on other large tables in the same environment or other environments.

How to reproduce

ClickHouse Version: 23.3.1.2823
Query: OPTIMIZE TABLE <table_name> FINAL

Expected behavior

Remove duplicated rows (take the most updated row)