Hello, On Wed, Mar 09, 2022 at 12:27:15PM -0800, Yosry Ahmed wrote: ... > These problems are already addressed by the rstat aggregation > mechanism in the kernel, which is primarily used for memcg stats. We Not that it matters all that much but I don't think the above statement is true given that sched stats are an integrated part of the rstat implementation and io was converted before memcg. > - For every cgroup, we will either use flags to distinguish BPF stats > updates from normal stats updates, or flush both anyway (memcg stats > are periodically flushed anyway). I'd just keep them together. Usually most activities tend to happen together, so it's cheaper to aggregate all of them in one go in most cases. > - Provide flags to enable/disable using per-cpu arrays (for stats that > are not updated frequently), and enable/disable hierarchical > aggregation (for non-hierarchical stats, they can still make benefit > of the automatic entries creation & deletion). > - Provide different hierarchical aggregation operations : SUM, MAX, MIN, etc. > - Instead of an array as the map value, use a struct, and let the user > provide an aggregator function in the form of a BPF program. I'm more partial to the last option. It does make the usage a bit more compilcated but hopefully it shouldn't be too bad with good examples. I don't have strong opinions on the bpf side of things but it'd be great to be able to use rstat from bpf. Thanks. -- tejun
