[rabbitmq-discuss] LIFO queue

[Simon MacMullen]

On 29/04/2014 10:27, Nagy, Attila wrote:
> Sadly, to be somewhat useful, it will be much uglier and more complex
> than a LIFO queue.

Oh, of course.

> Is this much harder to implement on the server side, or you don't see
> enough interest?

Bit of both. We've had a lot more requests for priority queues than LIFO 
over the years, and we're only just starting to work on them. LIFO seems 
like a pretty obscure request with only two people asking in 4 years I'm 
afraid.

And it's not completely obvious how to implement it; if you wanted to do 
it in terms of the existing queue implementation you'd have the problem 
that it's designed to move queues through to the head; that one way 
movement is a fairly big assumption (and it's more code than you would 
think - due to persistence and paging large parts of the queue may not 
be in memory at any given time).

Or you could do a new implementation; the in-memory version would be 
trivial but then you'd have to do persistence from scratch if you wanted 
it; also not very appealing. Or you could back to SQL or whatever; 
performance would be poor but you'd be able to do all sorts of 
random-access things and LIFO would fall out very easily.

But these are all quite big tasks I'm afraid.

Cheers, Simon