Revalidating Informix foreign key data

Abstract

When testing a database migration to a new platform that is not compatible, for example with a different OS or locale, you cannot restore from a level 0 archive on the new server. Data must be copied either via flat files or directly using SQL tools. If this is from a live database that is being updated and cannot be exclusively locked (which rules out dbexport), you cannot guarantee consistency of the data being reloaded, so NOVALIDATE must be set to ensure foreign keys can be reapplied.

This article provides stored procedure language (SPL) code to identify and delete any resulting invalid data.

Content

The following SQL generator was posted in the dbexport alternative thread on the IBM Informix community forum in May 2021, the only change below being that PROCEDURE has been replaced with FUNCTION which is more correct if data is returned:

Copy to Clipboard

DROP FUNCTION IF EXISTS sp_revalidate;

CREATE FUNCTION sp_revalidate() RETURNING LVARCHAR AS sql;
/*
    Generate SQL to check FOREIGN KEYS added with NOVALIDATE
    Doug Lawry, Oninit Consulting, March 2021

Usage in "dbaccess":

UNLOAD TO 'revalidate.sql' DELIMITER ';'
    SELECT * FROM TABLE (FUNCTION sp_revalidate());
*/

DEFINE l_constrname VARCHAR(128);              -- foreign key constraint name

DEFINE l_dtabname   VARCHAR(128);              -- detail table name
    DEFINE l_ptabname   VARCHAR(128);              -- parent table name

DEFINE l_dtabid     LIKE systables.tabid;      -- detail table ID
    DEFINE l_ptabid     LIKE systables.tabid;      -- parent table ID

DEFINE l_dindexkeys LIKE sysindices.indexkeys; -- detail index column numbers
    DEFINE l_pindexkeys LIKE sysindices.indexkeys; -- parent index column numbers

DEFINE l_dcolno     LIKE syscolumns.colno;     -- detail column number
    DEFINE l_pcolno     LIKE syscolumns.colno;     -- parent column number

DEFINE l_dcolname   VARCHAR(128);              -- detail column name
    DEFINE l_pcolname   VARCHAR(128);              -- parent column name

DEFINE l_keyid      SMALLINT;                  -- index array pointer
    DEFINE l_sql        LVARCHAR;                  -- generated SQL

FOREACH

SELECT  dc.constrname,
                TRIM(dt.tabname), dt.tabid, di.indexkeys,
                TRIM(pt.tabname), pt.tabid, pi.indexkeys
        INTO    l_constrname,
                l_dtabname, l_dtabid, l_dindexkeys,
                l_ptabname, l_ptabid, l_pindexkeys
        FROM    sysconstraints AS dc
        JOIN    sysobjstate    AS do ON do.name     = dc.constrname
        JOIN    systables      AS dt ON dt.tabid    = dc.tabid
        JOIN    sysindices     AS di ON di.idxname  = dc.idxname
        JOIN    sysreferences  AS dr ON dr.constrid = dc.constrid
        JOIN    sysconstraints AS pc ON pc.constrid = dr.primary
        JOIN    systables      AS pt ON pt.tabid    = pc.tabid
        JOIN    sysindices     AS pi ON pi.idxname  = pc.idxname
        WHERE   dc.constrtype = 'R'
        AND     do.objtype = 'C'
        AND     do.state = 'E'
        ORDER   BY 2, 1

LET l_sql =
            'SELECT d.ROWID AS row_id ' ||
            'FROM ' || l_dtabname || ' AS d ' ||
            'LEFT OUTER JOIN ' || l_ptabname || ' AS p ';

FOR l_keyid = 0 TO 15

LET l_dcolno = ikeyextractcolno(l_dindexkeys, l_keyid);
            LET l_pcolno = ikeyextractcolno(l_pindexkeys, l_keyid);

IF l_dcolno = 0 THEN
                EXIT FOR;
            END IF

SELECT  TRIM(colname)
            INTO    l_dcolname
            FROM    syscolumns
            WHERE   tabid = l_dtabid
            AND     colno = l_dcolno;

SELECT  TRIM(colname)
            INTO    l_pcolname
            FROM    syscolumns
            WHERE   tabid = l_ptabid
            AND     colno = l_pcolno;

LET l_sql = l_sql ||
                DECODE(l_keyid, 0, 'ON', 'AND') ||
                ' d.' || l_dcolname || " =" ||
                ' p.' || l_pcolname || " ";

END FOR

FOR l_keyid = 0 TO 15

LET l_dcolno = ikeyextractcolno(l_dindexkeys, l_keyid);

IF l_dcolno = 0 THEN
                EXIT FOR;
            END IF

SELECT  TRIM(colname)
            INTO    l_dcolname
            FROM    syscolumns
            WHERE   tabid = l_dtabid
            AND     colno = l_dcolno;

LET l_sql = l_sql ||
                DECODE(l_keyid, 0, 'WHERE', 'AND') ||
                ' d.' || l_dcolname || " IS NOT NULL ";

END FOR

LET l_sql = l_sql ||
            'AND p.ROWID IS NULL ' ||
            'INTO TEMP temp_rowid WITH NO LOG';

RETURN l_sql WITH RESUME;

LET l_sql =
            'UNLOAD TO "'    || l_dtabname || '.' || l_constrname || '.unl" ' ||
            'SELECT * FROM ' || l_dtabname || ' ' ||
            'WHERE ROWID IN (SELECT * FROM temp_rowid)';

RETURN l_sql WITH RESUME;

LET l_sql =
            'DELETE FROM '   || l_dtabname || ' ' ||
            'WHERE ROWID IN (SELECT * FROM temp_rowid)';

RETURN l_sql WITH RESUME;

LET l_sql =
            'DROP TABLE temp_rowid';

RETURN l_sql WITH RESUME;

END FOREACH;

END FUNCTION;

The foreign key hierarchy is traversed for the entire current database. You will need to create the function in every database that needs to be checked: table names are assumed to be in the database containing the stored procedure or function unless explicitly stated otherwise, and not in the database to which the calling session is connected.

The main FOREACH SELECT statement examines system catalog tables within the database being checked as follows:

Alias	Table	Description
dc	sysconstraints	Foreign key constraint (FK)
do	sysobjstate	FK object state
dt	systables	FK table
di	sysindices	FK index
dr	sysreferences	Constraint relationship
pc	sysconstraints	Primary key constraint (PK)
pt	systables	PK table
pi	sysindices	PK index

Criterion	Description
dc.constrtype = 'R'	Referential constraint (FK)
do.objtype = 'C'	Constraint object
do.state = 'E'	Enabled (remove to include all)

As well as the necessary object names and table IDs, “sysindices.indexkeys” are selected for both the foreign and primary key. We are using table sysindices which contains the column numbers in an index as an array so is easier to use here, rather than the more commonly used sysindexes view which represents the array as 16 separate columns. We can then loop through the 16 possible elements using an undocumented internal function with this specification:

Copy to Clipboard

Overall usage is as given in the source code comments:

Copy to Clipboard

UNLOAD is implemented client-side, so will only work in dbaccess or AGS Server Studio. If using other SQL clients, you will need to save the result set in such a way that semi-colons are added to each line, or edit the SPL function to do that.

Note also the under-documented trick to treat function output as a sub-query with:

Copy to Clipboard

The file produced by executing this in the standard “stores_demo” database created by dbaccessdemo contains:

Copy to Clipboard

There are repeating groups of 4 statements per foreign key. An example group with linefeeds inserted for readability follows:

Copy to Clipboard

The first statement identifies rows in the foreign key table for which the column values are not present in the corresponding primary key, and uses the technique described here:

https://www.oninitgroup.com/faq-items/sub-queries-a-faster-alternative-to-not-exists

NULL values in foreign key columns are exempted from referential checking, so are excluded with IS NOT NULL criteria.

The ROWIDs are saved in a temporary table, which is then used to:

save row contents in standard pipe-delimited UNLOAD format to file “table-name.foreign-key-name.unl”;
delete those rows.

WITH NO LOG is appended as a precaution, though you should always have configuration parameter TEMPTAB_NOLOG set to 1, making this the default anyway. This does not cause any error if the database is not logged.

The temporary table is then dropped as the name will be reused in the next repeating group of statements.

Should you wish to report invalid data without fixing it, the RETURN of delete statements can be commented out in the code, or lines beginning with DELETE removed from what is generated afterwards.

The result can then be executed to save a flat file per foreign key. Empty files can be ignored, or removed on Unix or Linux with:

Copy to Clipboard

If you decided not include DELETE statements, file contents can be reviewed before fixing data manually.

Caveats

The code assumes ROWID is available for all tables involved. If you have Enterprise Edition and are using tables with multiple partitions, WITH ROWIDS must be applied to them. Alternatively, if every table contains a SERIAL, SERIAL8 or BIGSERIAL column, the SPL function could be recoded to use those instead. Another work-around from IDS 14.10.xC6 onwards would be to use the undocumented IFX_ROW_ID pseudo-column: queries on them no longer sequentially scan the table.

Conclusion

The tool provided in this article means that a database can be copied while fully in use, and any resulting inconsistencies in referential integrity identified and fixed afterwards as necessary.

References

Documentation links not shown in full in this article can be queried at the new IBM Informix 14.10 site:

https://www.ibm.com/docs/en/informix-servers/14.10

Disclaimer

Suggestions above are provided “as is” without warranty of any kind, either express or implied, including without limitation any implied warranties of condition, uninterrupted use, merchantability, fitness for a particular purpose, or non-infringement.

Author

Doug Lawry
Senior Informix Consultant

Revalidating Informix foreign key data

Abstract

Content

Caveats

Conclusion

References

Disclaimer

Author

Related Technical Articles:

Share this story: