Hello, Gattaka, you wrote: G> Colleagues, G> At us in a command dispute concerning that one this and the same or different things. G> one state that it is strongly different entities Here they [basically] are right. G> also it is necessary to do three classes and three calls in a method. These are insignificant details of specific implementation. G> others say that it is an overhead projector and enough one essence - a broad gull. Caused one method and all on it, he already in itself(himself) decides what to write down in prometeus for metrics, and by default all in a broad gull. These are insignificant details of specific implementation. G> while at such approach it is not visible , i.e. we presume not to aggregate to ourselves on the client of the metrics, and all on udp in a local network. G> what sharing at you? Why it such? Whether there are powerful arguments that it was such as at you, instead of differently. Yes, we write to a broad gull, and already any analysts are under construction of a broad gull. It is made so because from client side there can be any device (even the most feeble), and from a server farm mass of idle computing powers and improbable volume of storage. Thus it is possible to build any arbitrary trejsy/metrics, for any current and historical period that, obviously, not to do in a case if from the client merged restricted, already processed, trejsy/metrics and [interesting now] the crude data . By experience even if the strong-willed optimistical decision was accepted that "is enough to store 640 kilobyte broad gulls for six months", upon, rather regularly there are questions "and as something / developed / in dynamics since 10 months ago", on it a period of storage of crude dens approximately 18 months. But also after that broad gulls are not deleted, and carefully press close the maximum compression, are written on a ribbon and postponed in archive.