Locking multiple rows in repeatable read mode

[Marko Vukovic]

If I understand correctly how repeatable read works, on the first select in a trasaction the database creates a snapshot of its current state and with few exceptions like "insert into ... select ..." the transaction does not see any changes hapenning outside of it.

In my situation I have to lock two or more arbitrary rows identified discretly by their primary key values. If I lock them independently like this:

select * from t where pk=3 for update;
select * from t where pk=8 for update;

another transaction may squezee between these two statements and do something the first transaction would not be aware of. I cannot process the records independently with a commit between them because this has to be dealt with in "all or nothing" manner. Is the only solution to lock both records at the same time like this:

select * from t where pk in (3,8) for update;

I am not particularly fond of that idea because I do not like building the statements of arbitrary lentght like this one dynamically. Also, would it honour the order in which it locks them so I can avoid a deadlock by sorting them?

Thank you for your response in advance.
Marko