michaelandrepearce commented on issue #2850: ARTEMIS-2504 implement retroactive addresses URL: https://github.com/apache/activemq-artemis/pull/2850#issuecomment-538579769 @jbertram tbh im struggling to understand still the issue, if an address is paging to disk, ring queue still gets new messages delivered. on address-settings front thanks, that seems to make things work for custom wildcards. When i have a large retro-active queue, and a new consumer joins, and that consumer may take 5 minutes to process those messages, im seeing some nasty GC spikes at the initial few seconds, from debugging that seems to be when the messages are copied back to route to original address and the new queue, as the broker goes as fast as it can regardless of the consumer speed. This kind of behaviour could destabilise the broker. Ideas to solve or limit this issue: 1) is there a way to control the copy over so its not done as one big bang (e.g. as fast as broker can possible do), but feed the new queue at the rate the consumer is consuming there for avoiding the very big spike in object creation. 2) Is it possible to make it so that we dont have to copy, but it simply is a message ref? 3) If 2 is not possible because a separate address, could we have a toggle that will deploy under the original address and when deployed like such a copy isnt needed and simply message ref is added to the queue, so then its a choice for end users. this then gives a kind of flexibility depending on end user case, with trade off for either, but at least then its up for the end user.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
