Implementing queue semantics (novice)

Hi,

since I am new to writing stored procedures I'd like to ask first bevore I
do a mistake.

I want to implement some kind of queue (fifo). There are n users/processes
that add new records to a table and there are m consumers that take out
these records and process them.
It's however possible for a consumer to die or loose connection while
records must not be unprocessed. They may rather be processed twice.

This seems to me as a rather common problem. But also with atomicy-holes to
fall into.
How is this commonly implemented?

I can imagine an 'add' and a 'get' function together with one aditional
'processed' timestamp-column?

Thanks for helping me do the right.