postgresql/contrib/pg_upgrade/TESTING

contrib/pg_upgrade/TESTING

The most effective way to test pg_upgrade, aside from testing on user
data, is by upgrading the PostgreSQL regression database.

This testing process first requires the creation of a valid regression
database dump.  Such files contain most database features and are
specific to each major version of Postgres.

Here are the steps needed to create a regression database dump file:

1)  Create and populate the regression database in the old cluster
    This database can be created by running 'gmake installcheck' from
    src/test/regression.

2)  Use pg_dump to dump out the regression database.  Use the new 
    cluster's pg_dump on the old database to minimize whitespace 
    differences in the diff.

3)  Adjust the regression database dump file

    a)  Perform the load/dump twice
        This fixes problems with the ordering of COPY columns for
        inherited tables.

    b)  Change CREATE FUNCTION shared object paths to use '$libdir'
        The old and new cluster will have different shared object paths.

    c)  Fix any wrapping format differences
        Commands like CREATE TRIGGER and ALTER TABLE sometimes have
        differences.

    d)  For pre-9.0, change CREATE OR REPLACE LANGUAGE to CREATE LANGUAGE

    e)  For pre-9.0, remove 'regex_flavor'

    f)  For pre-9.0, adjust extra_float_digits
        Postgres 9.0 pg_dump uses extra_float_digits=-2 for pre-9.0
        databases, and extra_float_digits=-3 for >= 9.0 databases.
        It is necessary to modify 9.0 pg_dump to always use -3, and
        modify the pre-9.0 old server to accept extra_float_digits=-3.

Once the dump is created, it can be repeatedly loaded into the old
database, upgraded, and dumped out of the new database, and then
compared to the original version. To test the dump file, perform these
steps:

1)  Create the old and new clusters in different directories.

2)  Copy the regression shared object files into the appropriate /lib
    directory for old and new clusters.

3)  Create the regression database in the old server.

4)  Load the dump file created above into the regression database;
    check for errors while loading.

5)  Upgrade the old database to the new major version, as outlined in
    the pg_upgrade manual section.

6)  Use pg_dump to dump out the regression database in the new cluster.

7)  Diff the regression database dump file with the regression dump
    file loaded into the old server.

The shell script test.sh in this directory performs more or less this
procedure.  You can invoke it by running

    gmake check

or by running

    gmake installcheck

if "gmake install" (or "gmake install-world") were done beforehand.
When invoked without arguments, it will run an upgrade from the
version in this source tree to a new instance of the same version.  To
test an upgrade from a different version, invoke it like this:

    gmake installcheck oldbindir=...otherversion/bin oldsrc=...somewhere/postgresql

In this case, you will have to manually eyeball the resulting dump
diff for version-specific differences, as explained above.