Our API server was hardcoded to talk directly to the node that was taken offline yesterday for a particular backend operation. This process ran away and caused a memory leak. The process has been updated to use DNS and should no longer occur. It has been resolved.