Hardware and performance

Yes, we did it on mid-July. I don't have my notes at hand
now (they are at work) but this is roughly what we did:

1.- We setup a second moodle installation using the exact
    same version of the code (in fact, we copied the files
    we were using in production). We did a standard install
    from scratch. This makes sure the postgres tables are
    correctly created.

2.- We stopped both the mysql moodle and the postgres
    moodle, to make data static.

3.- We dumped mysql database with the '-c' option (this is
    very important, otherwise postgres wasn't very happy
    with the inserts), using something like:

    mysqldump -c -t moodle > moodle.dump

4.- We stripped all the mdl_log table data, as it was huge
    and made the import into postgres very long, and
    stripped the '`' from the mysql table names (postgres
    doesn't like them), with

    grep '^INSERT INTO ' moodle.dump | grep -v '`mdl_log`' \
      | sed -e 's/`//g' > postgres.dump

5.- We emptied all the postgres tables (so the import
    doesn't duplicate content or stops due to duplicated
    values), but left the tables themselves. We ended up
    using a sql script, as we needed to do this several
    times (see why below).

6.- We connected to postgres as the moodle user an issued
    the following commands:

      \set ON_ERROR_STOP on
      \o /tmp/postgres-logs.txt
      BEGIN;
      \i /path/to/postgres.dump
      COMMIT;

    We set ON_ERROR_STOP to stop processing the dump file as
    soon as it encounters an error. Having a look at
    /tmp/postgres-logs.txt we could see why it was
    stopping.

    We also used transactions (the BEGIN/COMMIT commands) to
    speedup the insertion process quite a bit.

7.- As the table definitions are no 100% equal between mysql
    and postgres (at least they weren't in our 1.5.3+
    version), we had to loop through 5 and 6 several times,
    correcting the offending data in postgres.dump where
    possible, or 'fixing' the table definitions in MySQL to
    make them compatible with the postgres ones, and going
    to step 3 again.

8.- Once we had a successful import (after only a few
    changes), we needed to reset the postgres sequences to
    keep them in sync with real data. Patrick Li wrote the
    following one-liner
    (http://moodle.org/mod/forum/discuss.php?d=40836#193749):

    PGPASSWORD={password} psql -qAt -U {username} -c "\ds" {database} \
      | cut -d '|' -sf 2 \
      | sed "s/^\(\(.*\)_id_seq\)$/SELECT setval\('\1', \(SELECT MAX\(id\) FROM \2\), true\);/" \
      | PGPASSWORD={password} psql -U {username} {database}

    Replace {username}, {password} and {database} with your
    own ones.

As far as I remember, that was more or less what we did.

Saludos. Iñaki.