michaelandrepearce commented on issue #2850: ARTEMIS-2504 implement retroactive 
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 
   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 
   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:

With regards,
Apache Git Services

Reply via email to