Piotr Not used RDF Delta myself but the Zookeeper issues sound familiar
When anything is trying to talk to Zookeeper if the ZK node they are trying to talk to stops responding then the ZK client code will fail and then re-establish a session to a diferent node. BUT this isn’t immediate for a couple of reasons: 1) The clients socket connection, if already established, will be blocked until it times out 2) The ZK nodes themselves rely on heartbeats between themselves to spot node failures, thus there is a minimum time before they consider a node to have failed and elect a new leader So 1 you can improve by tweaking client side connection timeouts [1] and 2 you can improve by tweaking Zookeeper node configurations [2] Also the node that gets killed has an impact that varies, if the leader node is killed then the ZK cluster can no longer serve write requests until a new leader is elected. So depending on what the client code is doing at the time this can lead to blocked operations until writes are permitted again. Whereas if a follower node is killed the client is only likely to be impacted if it was directly communicating with that node, otherwise the cluster continues on happily unless it has fallen below the minimum threshold for quorum. Hope this helps, Rob [1] https://zookeeper.apache.org/doc/r3.1.2/zookeeperProgrammers.html#ch_zkSessions [2] https://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html#sc_configuration On 17/08/2020, 17:11, "Piotr Nowara" <piotrnow...@gmail.com> wrote: Hi, We are testing RDF Delta with three Zookeeper instances. Sometimes when we kill one of those Zookeper instances Fuseki freezes for about 30 seconds which is bad. Is this expected? Will increasing the number of Zookeper instances help to avoid such issues? Thanks, Piotr Nowara