I'm fairly sure that moving to a pull queue would generate more overhead for each request, granted it would probably solve their current issues. However IMO sti

I'm fairly sure that moving to a pull queue would generate more overhead for each request, granted it would probably solve their current issues. However IMO sticking with a push setup and solving the brokenness would be a higher performance solution, mainly due to the simplicity. It really sounds like they need to shard their current setup as a minimum, and preferably find some way to synchronise the session table over the routes within a mesh, so any one router is aware which dyno's are blocking/available for that queued traffic.