Paging does concern disk usage, but the page-full-policy won't necessarily correspond to what needs to happen when the disk is "full" according to either the max-disk-usage or the min-disk-free settings. Therefore, rather than overloading the page-full-policy I think it would be better to add a new setting (e.g. disk-full-policy) that could be set to either DROP or FAIL.
Would you mind opening a Jira [1] for this? If you don't have a Jira account I'll expedite your account approval when you request one. Justin [1] https://issues.apache.org/jira/browse/ARTEMIS On Thu, Nov 6, 2025 at 5:51 AM Iliya Grushevskiy <[email protected]> wrote: > Hi, > > I understand that blocking the producer when address paging is enabled and > the disk is full is the intended behavior. I would like to propose > considering not only the addressFullMessagePolicy parameter but also the > pageFullMessagePolicy when the disk is full and acting accordingly. It > seems that addressFullMessagePolicy relates to memory, while > pageFullMessagePolicy concerns disk usage. > > In my case, the problem looks like this: > - Clients should fail fast if there is any problem sending messages and > switch to another message broker. > - It is impossible to predict message distribution between queues and set > page byte limits accurately. > - Once the disk usage reaches its limits, clients get blocked for an > unknown period of time, without any possibility to redirect messages to > another broker. > - Changing the addressFullMessagePolicy to FAIL would stop message paging. > > Regards > Iliya Grushevskiy > > > > >
