Uploaded image for project: 'Derby'
  1. Derby
  2. DERBY-2204

Foreign key constraint backing index creation can be smarter when foreign key is a subset of the table's primary key

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Major
    • Resolution: Unresolved
    • Affects Version/s: None
    • Fix Version/s: None
    • Component/s: SQL
    • Urgency:
      Normal
    • Bug behavior facts:
      Performance

      Description

      When a foreign key's columns is a subset (or matches) the table's primary key then there is no need to create an additional backing index. Such an index will occupy disk space and decrease insert/update/delete performance.

      Two typical situations where this is a problem are shown in the schema for the oe system test.

      1) A multiple column primary key has a sub-set of columns which are a primary key in another table.

      ALTER TABLE CUSTOMER ADD CONSTRAINT
      CUSTOMER_PK PRIMARY KEY(C_W_ID, C_D_ID, C_ID);

      ALTER TABLE CUSTOMER ADD CONSTRAINT
      C_D_FK_DISTRICT FOREIGN KEY (C_W_ID,C_D_ID) REFERENCES DISTRICT;

      Derby will create two backing indexes here, one for the primary key and one for the foreign key. Derby could be improved
      so that no index is created for the foreign key, since Derby can perform an index scan on a sub-set of the first N columns in an index.
      Not sure how easy fixing the runtime portion of constraint manipulation is, if the constraint enforcement is via compiled SQL queries then it should be easier than if they are direct scans against the access api.

      2) A primary key in one table matches a primary key in another table, typically when the "sub" table contains additional optional information related to the main table.

      ALTER TABLE NEWORDERS ADD CONSTRAINT
      NEWORDERS_PK PRIMARY KEY(NO_W_ID, NO_D_ID, NO_O_ID);

      ALTER TABLE NEWORDERS ADD CONSTRAINT
      NO_O_FK FOREIGN KEY (NO_W_ID, NO_D_ID, NO_O_ID) REFERENCES ORDERS;

      Derby will create two identical backing indexes here, one for the primary key and one for the foreign key.
      Fixing this case might be easier than 1) since Derby already supports having mutliple logical indexes map to a single physical index.

      In both cases work is needed to handle dropping of the primary key while the foreign key constraint is kept:

      • in case 1) a new index will need to be built using the sub-set of the columns, thus dropping a primary key constraint could take time.
      • in case 2) possibly the backing index can be re-used but it needs to be converted from a unique index to a non-unique one.

      Also the case where the primary key is created after the foreign key needs to be covered, though that could be a follow on.

        Attachments

          Activity

            People

            • Assignee:
              Unassigned
              Reporter:
              djd Daniel John Debrunner
            • Votes:
              0 Vote for this issue
              Watchers:
              0 Start watching this issue

              Dates

              • Created:
                Updated: