Inconsistent results with Iter().Scanner()

[guyfedwards]

Version of driver: v1.7.0
Go version: 1.23
Scylla version: 5.2.19

I am having an issue with inconsistent results being returned in the iterator. The following queries are showing different results with count(*) being ~230k and the iter i value being ~170k. Running the count query through cqlsh gives the ~230k number so seems to be something specific to the iterator.

I have tried increasing paging, timeouts, consistency and paging manually but to no avail, any help/insight would be appreciated.

	id := "00000000-0000-0000-0000-000000000000"

	var res int
	err = cassSession.Query("SELECT count(*) FROM requests WHERE id = ?", id).Consistency(gocql.One).Scan(&res)
	if err != nil {
		log.Fatal(err)
	}

	fmt.Println("Count(*): ", res)

	i := 0
	iter := cassSession.Query("SELECT * FROM requests WHERE id = ?", id).Consistency(gocql.One).Iter()

	scanner := iter.Scanner()
	for scanner.Next() {
		i++
	}

	fmt.Println("Total i: ", i)

	if scanner.Err() != nil {
		log.Fatal(err)
	}

[guyfedwards]

I am able to reproduce this locally using a single scylla node with ~1m records.
count(*) from requests - 1200189
select count(*) from requests where id = 00000000-450e-4435-b7de-04446e58ab39; - 25576
Above iter.scanner() - 9363

For reference the indexes are:
PRIMARY KEY ((id, cassandra_unixhour), request_id)
CREATE INDEX id ON requests(id);

[joao-r-reis]

Can you test this with gocql 2.0?

[guyfedwards]

Same results with 2.0

[guyfedwards]

A bit more debugging and I'm somewhat even more confused although I wonder if it's more to do with my lack of understanding.
Running the following returns the correct result:

SELECT request_id FROM requests WHERE id = ?

Then adding more columns slowly reduces the number of results, e.g.

SELECT request_id, secondcolumn FROM requests WHERE id = ?

This returns ~600 less results than the single column query. This is however consistent with querying directly with cqlsh so that's at least something. I don't quite understand why select count(*)... would return the correct number of results but select *... would return a subset and likewise select x,y,z ... reducing results with each additional column.

For info, in the test database all columns apart from id, cassandra_unixhour, request_id, request_received are null which means it's not just excluding rows that are missing the column.