We're seeing errors about aborted JSON happening right about when indexes get changed over each time (and only on production), which seems consistent with the fact we just svc -d and then svc -u our search servers to load new indexes. Presumably we should be taking each server out of the loadbalancer when it's restarting.