Uploaded image for project: 'OpenJPA'
  1. OpenJPA
  2. OPENJPA-1689

The mapping tool does not remove user created sequences on PostgreSQL

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Closed
    • Minor
    • Resolution: Fixed
    • 2.0.0
    • 2.1.0
    • jdbc
    • None
    • PostgreSQL 8.4

    Description

      As reported on the users forum:

      <Christopher Schmidt>
      Hi all, using OpenJPA 2.0.0 with Postgresql 8.4 JDBC4

      I want to create the schema with the following property:
      <property name="openjpa.jdbc.SynchronizeMappings"
      value="buildSchema(SchemaAction='drop,add')"/>

      The entity is defined as follows:

      @Entity
      @Table(name = "obj_item")
      @Inheritance(strategy = InheritanceType.JOINED)
      @SequenceGenerator(name = "obj_item_id_seq", sequenceName =
      "obj_item_id_seq", allocationSize = 1)
      class ObjectItem ...

      it seems that the sequence will not be dropped - so I get the
      following exception:
      org.apache.openjpa.persistence.PersistenceException: ERROR: relation
      "obj_item_id_seq" already exists

      {stmnt 1834517285 CREATE SEQUENCE obj_item_id_seq START WITH 1}

      [code=0, state=42P07]
      at org.apache.openjpa.jdbc.meta.MappingTool.record(MappingTool.java:556)...
      </Christopher Schmidt>

      <Jeremy Bauer>
      As it turns out, I don't think this problem is the same as OPENJPA-1259. There is special code in the PostgresDictionary that treats sequences suffixed with "_SEQ" as system managed sequences. The comment in PostgresDictionary.isSystemSequence reads:

      // filter out generated sequences used for bigserial cols, which are
      // of the form <table>_<col>_seq

      This explains why the same code worked for me on DB2. Are you able to modify the sequence name? (ex. sequenceName = "seq_obj_item_id") If not, please file a new JIRA. It may/should be possible to make the code a bit smarter by actually verifying _seq suffixed sequences are for a bigserial column[1] instead of just making the assumption.

      [1] http://www.postgresql.org/docs/8.4/static/datatype-numeric.html#DATATYPE-SERIAL
      </Jeremy Bauer>

      I have a simple testcase that demonstrates the problem and only spent a very brief amount of time looking into a fix. Ideally, OpenJPA should be able to identify user created vs. system created sequences with suffix "_seq" and behave appropriately.

      Attachments

        1. OPENJPA-1689_test.patch
          6 kB
          Jeremy Bauer

        Issue Links

          Activity

            People

              techhusky Jeremy Bauer
              techhusky Jeremy Bauer
              Votes:
              0 Vote for this issue
              Watchers:
              0 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: