Hitachi Vantara Pentaho Community Forums
Results 1 to 14 of 14

Thread: Different result on 2 systems

  1. #1
    Join Date
    Jul 2006
    Posts
    17

    Default Different result on 2 systems

    Hello everybody,

    I am facing a strange problem: I run a job in Spoon on my local Windows 2000 PC that writes a total of 902 records into a destination table on a Postgres database. 852 records of these 902 contain a value in a specific column which is correct (these values come out of a dimension lookup step in the transformation).

    When I call exactly the same job out of an xaction on our Debian server it also writes 902 records into the destination table but only 612 records contain a value in the above mentioned column.

    Just a first guess: Is it possible that my server does not wait for the 240 missing values to come in from the dimension lookup step into the stream and finishes the job/transformation too early ?

    Thanks for any input and thoughts.
    Jens

  2. #2
    Join Date
    May 2006
    Posts
    4,882

    Default

    A hunch... your target table does not contain the same rows?

    Regards,
    Sven

  3. #3
    Join Date
    Jul 2006
    Posts
    17

    Default

    Hi Sven,

    no, the target table contains in both cases exactly the same number and values of rows. Just the column in question does not contain the same values in both tries.

    Meanwhile I played around a bit and have the assumption that it may be a problem with NULL values.

    Does it sound like a reasonable guess that my local Spoon (v. 2.5.0) on my Windows client handles the incoming NULL values in a different way when performing the dimension lookup than the Kettle Component in the PCI (v. 1_5_4) on the Debian server handles them ?

    Anybody comments or thoughts ?

    Regards
    Jens

  4. #4
    Join Date
    May 2006
    Posts
    4,882

    Default

    Quit possibly, if the pentaho version is a GA version I think it's Kettle 2.3.1 which is in there and between 2.3.1 and 2.5 I remember some bugs in dimension lookup have been fixed (regarding null valueS)

    Regards,
    Sven

  5. #5
    Join Date
    Jul 2006
    Posts
    17

    Default

    Just exchanged the kettle.jar from my PC to the server and from the server to my PC. No new result.

    On my PC it runs correctly with the older server jar and within the PCI it does not run correctly with the newer jar.

  6. #6
    Join Date
    May 2006
    Posts
    4,882

    Default

    next test... export your table.... run the old version... truncate and import the earlier export and run the new version.

    Regards,
    Sven

  7. #7
    Join Date
    Jul 2006
    Posts
    17

    Default

    just for your info: my transformation always truncates the destination table but in addition - to be 100% sure - I also did it manually everytime before I ran the transformation.

    just made another test: i copied my complete local Kettle folder to the debian linux machine and started carte.sh there.

    then i ran the transformation from my pc remotely on that server as slave server.

    result is still unchanged: the column in the destination table receives more NULL values as if i ran it locally.

    further i tested it on different java jre's (1.4 and 1.5) and both worked fine on windows machines but not on that linux server.

    what makes me wonder is that the column in question receives MORE NULL values than from my pc but not ONLY NULL values.

    so the lookup of the values seem to work only sometimes (not never and not always but something in between if you know what i want to say).

    is it possible that the lookup of a timestamp value using a date value works fine on my windows machine but not or different under linux ?

    i am not giving up ;-)

  8. #8
    Join Date
    May 2006
    Posts
    4,882

    Default

    which database are you using?

    Regards,
    Sven

  9. #9
    Join Date
    Jul 2006
    Posts
    17

    Default

    the source is informix, the destination is postgres

    i receive the date value from informix and use it to lookup on the postgres

    p.s.: thanks for your continous help and tests

  10. #10
    Join Date
    Jul 2006
    Posts
    17

    Default

    p.s.s.: both databases are neither running on my pc nor on the linux server in question

  11. #11
    Join Date
    May 2006
    Posts
    4,882

    Default

    But I'm almost at a loss... sanitize your transformation and attach it or send it over.

    Regards,
    Sven

  12. #12
    Join Date
    Jul 2006
    Posts
    17

    Default

    Here we go with the transformation that runs differently on Windows and Debian Etch.
    Attached Files Attached Files

  13. #13
    Join Date
    Jul 2006
    Posts
    17

    Default

    I made some further investigations and found out the following:

    I decreased the number of input rows to only 3 which look like this

    Row ID / Date
    1 / 2007-04-03
    2 / NULL
    3 / 2007-04-04
    4 / 2007-04-02

    The NULL value in Row ID 2 is correct, there exists no value for that row (it's something like 'date last invoice' and here we don't have any).

    As soon as the Linux transformation reaches the first NULL value (Row ID 2), all later values result in a NULL value (e.g. Row IDs 3 + 4 in this case).

    This happens on our Debian system only, on the W2K system it returns the correctz dates without problem.

    Any further hint where to search ?

    Thanks for any input in this matter.
    Jens

  14. #14
    Join Date
    May 2006
    Posts
    4,882

    Default

    If you're sure the Kettle installation is the same, then it's something else:
    - different version of JDK, or even if it's the same some JDK, some system specific problem.
    - ... out of ideas for now.

    Regards,
    Sven

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
Privacy Policy | Legal Notices | Safe Harbor Privacy Policy

Copyright © 2005 - 2019 Hitachi Vantara Corporation. All Rights Reserved.