It looks that when whole dataset to sort is small enough to fit into allowed memory for external sort implementation then sorting works fine (sorry for overcomplicated example, I was testing multiple theories):

// putenv('FLOW_EXTERNAL_SORT_MAX_MEMORY=10M');
$arr = [];
for ($j = 100; $j > 0; $j--) {
    for ($i = 200; $i > 0; $i--) {
        $arr[] = ['id' => str_pad((string) $j, 5, '0', STR_PAD_LEFT).'-'.str_pad((string) $i, 3, '0', STR_PAD_LEFT)];
    }
}

shuffle($arr);

data_frame()
    ->read(from_array($arr))
    ->sortBy(ref('id'))
    ->limit(10)
    ->write(to_output(truncate: true))
    ->run();

Output:

+-----------+
|        id |
+-----------+
| 00001-001 |
| 00001-002 |
| 00001-003 |
| 00001-004 |
| 00001-005 |
| 00001-006 |
| 00001-007 |
| 00001-008 |
| 00001-009 |
| 00001-010 |
+-----------+

But when we set memory limit to anything below amount required to run sorting (example above reaches ~60MB for my tests), by e.g. uncommenting putenv('FLOW_EXTERNAL_SORT_MAX_MEMORY=10M'); the output for the same code as above is:

+-----------+
|        id |
+-----------+
| 00100-200 |
| 00100-199 |
| 00100-198 |
| 00100-197 |
| 00100-196 |
| 00100-195 |
| 00100-194 |
| 00100-193 |
| 00100-192 |
| 00100-191 |
+-----------+

I believe it may be a bug and should work the same, no matter if we use (filesystem) cache to support sorting or not, right? 🙂

Thanks in advance for taking a look into this!