As long as the nodes are "up" during the planning phase they will be included in the planning. If they go "down" after planning, (i.e. during execution) and fragments are requested, they will not report, and will fail the query.
So if you start off with 5 nodes, but node 4 is down for patches, and then you submit a query, the remaining 4 nodes will execute the query with no issues. If you start out with 5 nodes, all up and running, you submit a query, and during execution, you take down node 4, that query will fail. There is work being done to help with "taking down" nodes sanely to help with patching. Resending work expected from a certain node to another node is "hard" if I understand past email chains correctly. You can't just retry fragments on other nodes once Node 4 goes down. It will take a Drill expert to explain that better than I can though. John On Mon, Sep 11, 2017 at 12:30 PM, Padma Penumarthy <ppenumar...@mapr.com> wrote: > Did you mean to say “we could not execute any queries” ? > > Need more details about configuration you have. > When you say data is available on other nodes, is it because you > have replication configured (assuming it is DFS) ? > > What exactly are you trying and what error you see when you try to > execute the query ? > > Thanks, > Padma > > > On Sep 11, 2017, at 9:40 AM, Kshitija Shinde <kshs...@gmail.com<mailto: > kshs...@gmail.com>> wrote: > > Hi, > > We have installed drill in distributed mode. While testing drillbit we have > observed that if one of node is done then we could execute any queries > against the drill even if data is available on other nodes. > > > > Is there any workaround for this? > > > > Thanks, > > Kshitija > >
