This sucks:

SELECT X.* FROM no_chain_samplemodel as X
       JOIN (SELECT user_id, MAX(timestamp) AS timestamp
       FROM no_chain_samplemodel
       GROUP BY user_id) AS Y
       ON (X.user_id = Y.user_id and X.timestamp = Y.timestamp)i
       WHERE X.staff_id = %s

This is great:

SELECT DISTINCT ON (user_id) FROM no_chain_samplemodel 
      WHERE timestamp <= '2017-01-01' ORDER BY user_id ASC, timestamp DESC;

With the Django ORM:

SampleModel.objects.distinct('user_id').\
    filter(timestamp__gt=mydate).order_by('user_id', '-timestamp')

From this

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.mysql',
        ...
     }
}

To this

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.postgresql_psycopg2',
        ...
     }
}

Massive fintech Django project:

• 190k lines of Python code
• > 100 Apps
• 383 tables to migrate
• 3000 tests running in ~3 minutes

• adapt code
• migrate data
• profit!

Pgloader to the rescue:

No live replication == Downtime!!!

• drop foreign keys (ForeignKey → IntegerField)
• adapt queries
• write a database router

• get Postgres on CI (stable tests)
• search for untested raw queries
• manual testing on real data

• use migrations for everything
• test everything
• NEVER rely on implicit ordering
• make Django apps really independent
• split that monolith ASAP

Hofstadter’s Database Migration Law:

Migrating from MySQL to Postgres always takes longer than you expect, even when you take into account Hofstadter’s Law.

@andreacrotti https://www.iwoca.co.uk/