Any empty plan with just the header containing the option does seem odd... A separate (non-plan) message seems more reasonable to me..
Aman On Mon, Mar 31, 2014 at 6:36 PM, Jason Altekruse <[email protected]>wrote: > Hey Drillers, > > I submitted a pull request to Julian today for the option parsing in optiq > and he already pulled it in, so I was going to finish up connecting drill > to it. Previously I was converting the rel node for setting options into a > new Drill operator. This has been replaced with specifying the options in > the plan header, with the options being read and set in the foreman when > the query is submitted. > > For options provided though SQL, I believe that single query strings can > only contain a single option setting statement and nothing else. This would > leave us with an empty plan, but meaningful headers. Do we want to define a > different type of message for these and other possible non-query messages > passed to Drill? I haven't checked to see it the current plan reader will > accept an empty plan graph, but I wanted to confirm if this would be a > reasonable way to do it. > > -Jason >
