Why Building Small Projects is the Best Way to Learn Programming

       

The Problem: Schema Changes on 2TB Tables

       

-- The migration that would lock our users table for 47 minutes
  ALTER TABLE auth_user ADD COLUMN last_active_at TIMESTAMP;

         

The Solution: Triple-Phase Migration

       

Phase 1: Add Nullable Column

       

# migration_0001.py
  class Migration(migrations.Migration):
      operations = [
          migrations.AddField(
              model_name='user',
              name='last_active_at',
              field=models.DateTimeField(null=True),
          )
      ]

         

Phase 2: Backfill with Django Batch Updates

       

# Backfill script
  from django.db import transaction
  
  with transaction.atomic():
      for user in User.objects.iterator(chunk_size=5000):
          User.objects.filter(pk=user.pk).update(
              last_active_at=user.session_set.latest().created_at
          )

         

Phase 3: Set NOT NULL

       

-- Final migration (run during low traffic)
  ALTER TABLE auth_user ALTER COLUMN last_active_at SET NOT NULL;

         

Key Metrics

                                                                                                                                                                                                                                                             

Approach	Downtime	CPU Spike
Direct ALTER TABLE	47 min	82%
Triple-Phase	0	23%