I'll explain how I broke completely the indexing of 10 shared clusters in production and how we worked during hours to put the service back on track.