Hi Aitozi, The example you posted does not correspond to how the Spool operator is implemented in Calcite.
The Calcite plan has a tree structure not a DAG structure. Best, Stamatis On Mon, Mar 13, 2023 at 4:34 PM Aitozi <[email protected]> wrote: > Hi Stamatis, > Thanks for your reply. As I understand, for the Spool operator it can > has multiple > consumer as its parents. For HepPlanner and CommonRelSubExprRule it will > inject > the parents into the HepRuleCall, So during the match/onMatch, we can see > the whole structure > of the Spool and its parents. > > for example, in this structure: > > Sink Sink > | | > Project Project > \ / > Spool > | > Scan > > Then In this way, I can see all the parents of the Spool and push down the > unified projects down to the Spool operator. > I think it's a useful optimize rule for the common subExprs. > > Best, > Aitozi. > > Stamatis Zampetakis <[email protected]> 于2023年3月13日周一 20:27写道： > > > Hi Aitozi, > > > > All Calcite's built-in operators (including Spool) have a single parent > so > > I don't fully understand what you mean by "parent components". > > > > What exactly do you expect to see in the rule match? Can you provide some > > concrete examples? > > > > In order to get the parent of the spool operator I guess you would have > to > > create a rule with the following supplier: > > > > withOperandSupplier(b0 -> b0.operand(RelNode.class).oneInput(b1 -> > > b1.operand(Spool.class).anyInputs())) > > > > Best, > > Stamatis > > > > On Sun, Mar 12, 2023 at 5:07 PM Aitozi <[email protected]> wrote: > > > > > It seems for a specific rule: CommonRelSubExprRule, the HepPlanner will > > > inject the parent nodes. I will try on this. > > > > > > Best, > > > Aitozi > > > > > > Aitozi <[email protected]> 于2023年3月12日周日 09:26写道： > > > > > > > Hi > > > > Recently, I've been conducting research on how to maximize the > > > > potential of Spool for CTE reuse optimization. My goal is to > streamline > > > the > > > > push down process and filters across all consumer nodes. However, I > > have > > > > encountered a roadblock as the current rule pattern fails to > accurately > > > > describe the structure of Spool and its parent components. So, in > the > > > rule > > > > match we can not see the structure as a whole. Would greatly > appreciate > > > any > > > > guidance or insights on this topic. > > > > > > > > Thanks, > > > > Aitozi > > > > > > > > > >
