[rabbitmq-discuss] memory overflow behavior

[Matthias Radestock]

Dmitriy,

On 05/06/12 17:12, Dmitriy Stolyarov wrote:
> Did a test with posting messages to a queue that no one is reading from,
> far beyond the high memory watermark. The behavior at that point for
> rabbit was to start moving queue contents to disk, throttling the
> producer and keeping the memory usage in a good shape. However when a
> client finally tried to connect to read from the queue, it looked like
> the rabbit node tried to load everything that was offloaded to disk back
> to memory at the same time, before starting to serve the client. Could
> this be done as an incremental process?

Messages are only paged back into memory as needed, i.e. rabbit is 
already doing what you suggest. So there must be some other explanation 
for the behaviour you saw.

Matthias.