Ok. Thanks ;-) Le mar. 20 oct. 2020 à 19:12, Matthias J. Sax <mj...@apache.org> a écrit :
> It's highly use-case dependent, but applying a filter at the end does > sound like a good solution to me. > > -Matthias > > On 10/19/20 12:40 PM, Mathieu D wrote: > > Hello there, > > > > Let's say I need to restart my streams app from a blank state (whether by > > changing app.id or using application-reset-tool). > > My app is designed on "at least once" paradigm, and outputs are upserts. > > The input topics have a few days worth of data, and the app will restart > > from there. > > > > If I restart it, states will be empty, meaning most outputs will be > wrong. > > So it should NOT upsert anything downstream for a certain period of time > > (something like my biggest time window or something) > > > > I'm thinking of a filter at the end of the topology, removing any record > > whose timestamp < topic retention + biggest window size. > > > > How do you guys manage this usually ? > > > > Thanks for your insights > > > > Mathieu > > >