Thanks you Steph, thank you Enrico. I'll share your feedback with our architects
Regards, Rémi -----Message d'origine----- De : Steph van Schalkwyk <[email protected]> Envoyé : Monday, September 14, 2020 18:53 À : [email protected] Cc : Stephane Galles <[email protected]> Objet : [EXTERNAL] Re: Zookeeper on Kubernetes and Presistent Volumes Remi We are using ZK in k8s with SOLR and Fusion. We are using PVs. I cannot see how one could not use PVs. Nonetheless, we are using 5 zk instead of the usual 3 we use outside of k8s, and we are working on an elaborate restore schema in case the quorum is broken. Hope this helps. Steph *steph van schalkwyk+1.314.452.2896 (Tel/SMS)* On Mon, Sep 14, 2020 at 10:21 AM Enrico Olivelli <[email protected]> wrote: > Remi, > sorry for the late reply. > > Your cluster would be able to work if and only if at least X/2 + 1 > servers are up and running and properly connected to the other peers. > any server that is restarted will rejoin from scratch. > It is very dangerous ! you won't be able to recover and probably it > will be very hard to understand what happened (you are going to lose > all of your tx > logs) > > I don't have experience with Kubernetes and non persistent volumes but > as far as I know it is not supposed to work > > Enrico > > > Il giorno ven 11 set 2020 alle ore 14:21 Remi Serrano > <[email protected]> ha scritto: > > > Hello mailing list, > > > > We are assessing running Zookeeper in Kubernetes. There are a bunch > > of examples around and they all use Kubernetes Persistent Volumes. > > For some underlying technical reasons, we would like to avoid the > > use of Kubernetes Persistent Volumes. > > What is the risk to setup a ZK cluster on Kubernetes without > > persistent volume ? > > Sub-question, what happen if a ZK node without persistent get killed > > and rebooted by Kubernetes (ie: without any data) ? > > > > Thanks for your help > > > > Rémi > > > > >
