Hello, On Tue, Nov 07, 2023 at 05:26:41PM -0700, Tycho Andersen wrote: > Hi all, > > At Netflix, we have a "canary" framework that will run test versions of > an application and automatically detect anomalies in various metrics. We > also have two "fleets", one full of virtual machines, and one that is a > multi-tenant container environment. > > On our full-VM fleet, one of the metrics we analyze is the number of open file > descriptors, from /proc/sys/fs/file-nr. However, no equivalent exists for the > multi-tenant container fleet, which has lead to several production issues. > > This idea is not new of course [1], but hopefully the existence of the new misc > cgroup will make it more tenable. > > I'm not really tied to any of the semantics in this series (e.g. threads could > be double counted even with a shared table), and am open to implementing this > in other ways if it makes more sense. As already raised by Christian, if the goal is monitoring, you should be able to do that easily with BPF. Thanks. -- tejun
