Sounds interesting. Can you give some concrete example(s)? Are you proposing hedged reads across servers? If so how will you handle the fact that you could read old data? (lagging follower)
Patrick On Wed, May 4, 2016 at 4:31 PM, rahul gidwani <[email protected]> wrote: > Sometimes with downed hosts, there is a few second delay from the client > (which is okay for most situations). But some requests can not tolerate > this delay and need to detect failure very quickly. > > I was wondering if the community would be interested in allowing for the > client to perform hedged reads (configurable) if needed. > > This could satisfy those situations where you can tolerate some latency for > most of your requests, but also accommodate those requests that have more > strict latency guarantees. > > If interested we could discuss the design and proposed implementation > details but right now am just wondering if the community would be > interested in such a feature. > > Thanks > rahul
