What are the Oracle Database settings for 10.5.2 and 3.5.2?

I wanted to see if anyone has set up the following settings in an Oracle Database for TRIRIGA 10.5.2 and 3.5.2:

  • 1. NLS_LENGTH_SEMANTICS: Should this be set to CHAR? In our current production environment, it’s set to BYTE, but the TRIRIGA support documentation says that this can lead to data loss, so they recommend using CHAR.
  • 2. NLS_CHARACTERSET: This is set to WE8ISO8859P1 in our current production environment, but the support document says that it must be UTF-8 or UTF-16.
  • 3. Block size: This is set to 8k, but the documentation recommends using 16k.

For (1) and (2), if you never want to store multibyte characters, then what you have is fine. But if you do, then you must use what the support documentation suggests.  Once you have your database created, it is difficult and time-consuming to change it, and it needs to be done outside of TRIRIGA. As for (3), I would encourage you to use 16k, since it will allow you better throughput and paging, unless you have a strong reason why you need to stay at 8k.

[Admin: This post is related to the 04.04.16 post about database character settings. NLS refers to National Language Support parameters in Oracle. To see other related posts, use the Multibyte tag, MBCS tag, or NLS tag.]

Continue reading

Having an issue with MBCS conversion after upgrade to 3.5.1

I’m running into a small issue with our platform upgrade to TRIRIGA 3.5.1. One of the prerequisites of our 3.5.1 upgrade was to delete any custom tables from the database, because they interfere with the MBCS conversion process. We had a custom table that we are using for integrations, so we scripted it out and deleted it before starting the installation of 3.5.1. But for some reason, TRIRIGA still thinks that the table is there, tries to perform MBCS conversion on it, and throws an Invalid Object Name exception since the table does not really exist in the database.

I am not sure why the MBCS conversion process is looking for a table that does not exist. I am thinking that there’s a list of tables maintained somewhere in TRIRIGA and that’s what’s causing this issue. I’m hoping that someone else has seen this issue before and can guide me in the right direction.

I finally figured it out. I found the script that’s used for converting varchar to nvarchar, and one of the things it does is that it creates a table called IBS_MBCS_CONVERSION_RECOVERY, which is used in case the conversion of a table fails. It turns out that we still had an entry in IBS_MBCS_CONVERSION_RECOVERY for our custom table from before, which was essentially causing the platform to attempt a conversion every time on startup. Once I deleted the entry for our custom table, the conversion error went away.

Continue reading

Installing TRIRIGA 3.4.2 on Microsoft SQL Server database

Disclaimer: TRIRIGA encourages using the latest platform release when installing new TRIRIGA databases. This ensures that you have the latest platform with fixes to known issues. However, if you need to stay at 3.4.2.x and are installing on Microsoft SQL Server, review the following as you might encounter collation issues.

New database install, single-byte

There is a known issue with a TRIRIGA 3.4.2 and SQL Server new install where the single-byte option is ignored during the installation process and the resulting TRIRIGA database will be MBCS with a Latin1_General_CS_AS collation. If a single-byte environment is absolutely necessary, we suggest using TRIRIGA 3.5 where the issue has been resolved.

New database install, multibyte, non-English collation

There is a known issue with a TRIRIGA 3.4.2 and SQL Server new install where the multibyte option is selected with a collation other than English/Latin1_General_CS_AS. The chosen collation is used when building and populating the database, but the ENVIRONMENT_PROPERTIES value for “db.mbcs.mssql.collation” is not set correctly and can cause issues when new text fields and new business objects are created and populated. You can possibly get a collation conflict exception. (Note that for English collation, there is no issue since the “db.mbcs.mssql.collation” is set correctly.) We suggest you make the following changes to ENVIRONMENT_PROPERTIES…

[Admin: This post is related to the 06.29.16 post about installing TRIRIGA 3.4.2 or 3.5.0 on a non-US English database.]

Continue reading

Having a question about the database character settings

We have received different opinions from our DBA for the international language set. Is it OK to leave it as BYTE?

  • Database Character Set: AL32UTF8
  • Character Length Semantic: BYTE
  • Message: WARNING!! The database character set is inconsistent with the character length. When using a UTF character set which supports multibyte characters the NLS_LENGTH_PARAMETER must be set to CHAR. Refer to the Administration Guide for more details

In the following Oracle 11g document and Oracle 12c document, there is a caution given by Oracle when setting the parameter to CHAR:

  • Caution: Oracle strongly recommends that you do NOT set the NLS_LENGTH_SEMANTICS parameter to CHAR in the instance or server parameter file. This may cause many existing installation scripts to unexpectedly create columns with character length semantics, resulting in runtime errors, including buffer overflows.

TRIRIGA is not one of those application that is impacted. In fact, we require it for the length of fields to be saved correctly. Oracle can be safely ignored in this case. If they do not use NLS_LENGTH_SEMANTICS = ‘CHAR’ and they put MBCS data into their database, they will run into issues when saving data. When we create tables, let’s say with the size of 100, we expect it to be 100 CHARs not 100 BYTEs, and if saving 100 MBCS characters, you will not be able to save. If you are saving a single byte date, it doesn’t matter, BYTE is fine.

Having said that, the NLS_LENGTH_SEMANTICS is not something to change midstream. If they already created their TRIRIGA database and the length semantic was at BYTE, their database fields are probably already set to BYTE. TRIRIGA is not one of those applications where this is applicable. In fact, without it set correctly, we will have save issues. We rely on the database to handle field size properly and our documentation clearly states that if you are going to save MBCS data, your NLS_LENGTH_SEMANTICS should be set to ‘CHAR’ prior to creating the TRIRIGA database.

Continue reading