Thanks for your reply Lei. Yes you answered my questions. I assume Helix will remember the preference list (assignment) in a resource's ideal state.
On Mon, Jun 6, 2016 at 11:04 AM, Lei Xia <[email protected]> wrote: > Hi, Neutron > > Could you be more specific on your question? In semi-auto mode, you (the > client) will specify a preference list for each partition (or Helix can > generate the list for you too by calling admin.rebalance()). The list is > fixed, i,e Helix will not automatically recalculate the list. > > Given a preference list for a partitions, for example: > > { p0: [node-1, node-2, node-3]. ...} > > Helix will try to bring p0 to online state for node 1,2,3. If node-1 is > disconnected from zookeeper (crashed, for example), the state of p0 on > node-1 will be offline. Once node-1 comes back, Helix will bring p0 on > node-1 back from offline to online. > > Not sure if this answers your question. > > > Thanks > Lei > > > On Fri, Jun 3, 2016 at 2:55 PM, Neutron sharc <[email protected]> > wrote: > >> Hi the team, >> >> semi-auto mode supports a feature that, after a failed participant >> comes back online, its owned replicas will be reused again (transit >> from offline to slave etc). How can Helix recognize the replicas that >> are owned by a participant after it reconnects after a failure? We >> are trying to build such a feature in a user-defined rebalancer. You >> input is highly appreciated. >> >> >> -neutron >> > > > > -- > > *Lei Xia *Senior Software Engineer > Distributed Data Systems/Nuage & Helix > LinkedIn > > [email protected] > www.linkedin.com/in/lxia1
