Hello, _NN _, you wrote: _NN> Memory leaks - application restarting: how in Instagram disconnected garbage collector Python and started to live... _NN> That think of it? Garbage collector shut-down gives a double scoring: We released almost 8 operative storage on each server and could use them for creation amounts of worker processes on servers with restricted throughput of storage, or to reduce an amount of restartings of processes on servers with restriction on capacity of the CPU; Throughput of the CPU also increased, as the amount of the instructions executable for one clock period (IPC) increases almost on 10 %. If they so are torn to control manually storage could use a C ++. But clear business that could not because enough to esteem history Instagram. Here after all business in scaling if the server one or the several people do as can. Iron to buy more cheaply, than to employ programmers for rigid optimization. But if the project shot, and such units for thousand and tens thousand servers it is "suddenly clarified that optimization gives not sickly saving. And as a matter of fact that they need to do, here they and are engaged in administration of servers. Disconnecting garbage collector Python (GC) which releases storage, tracing and deleting not used data, Instagram began to work on 10 % faster. Yes-yes, you did not mishear! Yes-yes, using a C ++ they could achieve still productivity, but it is necessary to be realists all the same. The decision already works let and through a bum, and probably all perfectly know as eulogized the same garbage collectors which appeared in reply to attempt to control operative storage automatically, that is finally not to care of optimization. Well and they too could control manually. More precisely programmers could control, and in our case it is a question of administration, and is fine. In what as a matter of fact a question? You do not think that they begin to rewrite all system from zero. And over reliability there it is hardly necessary.