On Tue, Sep 21, 2010 at 01:20:08PM +0100, David Cantrell wrote: >My first choice would be to just use the filesystem. Have the >producer(s) write its files somewhere and then mv them to a directory >that the consumer knows about once it's completely written.
OK, a bit more detail. This is a pretty lightweight thing; the idea is for the producer to generate and cache sound files from a text-to-speech system, while the consumer plays them over VoIP. (This is all happening inside an AGI script.) That way there should be a delay while it's generating the first file, but by the time that has been played the second file ought to be ready. So the consumer should not delete the files once it's played them, because they might be useful later. Of course there would be no difficulty in creating a second set of files which contained pointers to the first, and _those_ could be deleted once used. R