|
| 1 | +Thse directories take the Query structure returned by the parser, and |
| 2 | +generate a plan used by the executor. The /plan directory generates the |
| 3 | +plan, the /path generates all possible ways to join the tables, and |
| 4 | +/prep handles special cases like inheritance. /utils is utility stuff. |
| 5 | + |
| 6 | +planner() |
| 7 | + handle inheritance by processing separately |
| 8 | +-init_query_planner() |
| 9 | + preprocess target list |
| 10 | + preprocess qualifications(WHERE) |
| 11 | +--query_planner() |
| 12 | + pull out constants from target list |
| 13 | + get a target list that only contains column names, no expressions |
| 14 | + if none, then return |
| 15 | +---subplanner() |
| 16 | + make list of relations in target |
| 17 | + make list of relations in where clause |
| 18 | + find which relations can do merge sort and hash joins |
| 19 | +----find_paths() |
| 20 | + find scan and all index paths for each relation not yet joined |
| 21 | + one relation, return |
| 22 | + find selectivity of columns used in joins |
| 23 | +-----find_join_paths() |
| 24 | + Summary: With OPTIMIZER_DEBUG defined, you see: |
| 25 | + |
| 26 | + Tables 1, 2, 3, and 4 are joined as: |
| 27 | + {1 2},{1 3},{1 4},{2 3},{2 4} |
| 28 | + {1 2 3},{1 2 4},{2 3 4} |
| 29 | + {1 2 3 4} |
| 30 | + |
| 31 | + Actual output tests show combinations: |
| 32 | + {4 2},{3 2},{1 4},{1 3},{1 2} |
| 33 | + {4 2 3},{1 4 2},{1 3 2} |
| 34 | + {4 2 3 1} |
| 35 | + |
| 36 | + Cheapest join order shows: |
| 37 | + {4 2},{3 2},{1 4},{1 3},{1 2} |
| 38 | + {3 2 4},{1 4 2},{1 3 2} |
| 39 | + {1 4 2 3} |
| 40 | + |
| 41 | + It first finds the best way to join each table to every other |
| 42 | + table. It then takes those joined table combinations, and joins |
| 43 | + them to the other joined table combinations, until all tables are |
| 44 | + joined. |
| 45 | + |
| 46 | + jump to geqo if needed |
| 47 | + again: |
| 48 | + find_join_rels(): |
| 49 | + for each joinrel: |
| 50 | + find_clause_joins() |
| 51 | + for each join on joinrel: |
| 52 | + if a join from the join clause adds only one relation, do the join |
| 53 | + or find_clauseless_joins() |
| 54 | + find_all_join_paths() |
| 55 | + generate paths(nested,mergesort) for joins found in find_join_rels() |
| 56 | + prune_joinrels() |
| 57 | + remove from the join list the relation we just added to each join |
| 58 | + prune_rel_paths() |
| 59 | + set cheapest and perhaps remove unordered path, recompute table sizes |
| 60 | + if we have not done all the tables, go to "again" |
| 61 | + do group(GROUP) |
| 62 | + do aggregate |
| 63 | + put back constants |
| 64 | + re-flatten target list |
| 65 | + make unique(DISTINCT) |
| 66 | + make sort(ORDER BY) |