On Wed, September 27, 2006 2:31 pm, Joe Orton wrote: > The new approach is exactly the same for other bucket types, FILE should > not be treated as special just to avoid that. Other bucket types will > cause the same memory consumption issue (notably CGI/PIPE).
I looked at this issue, but I could not see a bucket type that would typically hold more data in one bucket than available RAM (unless I am missing the behaviour of one of the bucket types, which is also likely). The ideal solution so far seems to be to follow Niklas' strategy of supporting read-while-cache, and then to write the cache file in parallel using the copy_body method. Bodies are then copied in parallel regardless of bucket type. Regards, Graham --
