I think in that case overcommit will happily say the allocation worked. Unless you also zero the entire chunk of memory and then get OOM killed on the write. I

I think in that case overcommit will happily say the allocation worked. Unless you also zero the entire chunk of memory and then get OOM killed on the write. I suppose you can try to reliable target "seriously wild allocation fails" without leaving too much memory on the table. 0: Heuristic overcommit handling. Obvious overcommits of address space are refused. Used for a typical system. It ensures a seriously wild allocation fails while allowing overcommit to reduce swap usage. root is allowed to allocate slightly more memory in this mode. This is the default. https://www.kernel.org/doc/Documentation/vm/overcommit-accounting https://www.kernel.org/doc/Documentation/vm/overcommit-accou... Running in an environent without overcommit would allow you to handle it gracefully though, although bringing its own zoo of nasty footguns. See this recent discussion on what can happen when turning off overcommit: https://news.ycombinator.com/item?id=46300411 https://news.ycombinator.com/item?id=46300411