I’ve been creating a lot of new tables (“objects”, in Django parlance) in Buttondown lately. These are fairly boring tables. Here are a few of them:

  • ArchiveImport (to track folks importing their archives)
  • PeacefulEmailAddress (to track email addresses that are confirmed as users to be valid, even if parts of my validation infrastructure think otherwise)
  • ImageUploadAttempt (to track images that folks have uploaded to S3)

These are…very simple models! To wit:

import uuid
from enum import Enum

from django.db import models

from emails.models import Newsletter
from miscellany.fields import EnumField


class SubscriberImportSource(Enum):
    BUTTONDOWN = "buttondown"  # This means imports/exports.
    CSV = "csv"
    MAILCHIMP = "mailchimp"
    SUBSTACK = "substack"
    TINYLETTER = "tinyletter"
    TYPEFORM = "typeform"
    DRIP = "drip"
    GENERIC = "generic"

    @classmethod
    def choices(cls):
        return [(x.value, x.name) for x in cls]


class SubscriberImportStatus(Enum):
    FAILED = "failed"
    IN_PROGRESS = "in_progress"
    NOT_STARTED = "not_started"
    SUCCEEDED = "succeeded"

    @classmethod
    def choices(cls):
        return [(x.value, x.name) for x in cls]


class SubscriberImport(models.Model):
    """
    Mostly used for auditing.
    """

    id = models.UUIDField(primary_key=True, default=uuid.uuid4, editable=False)
    newsletter = models.ForeignKey(
        Newsletter, on_delete=models.CASCADE, related_name="subscriber_imports"
    )
    creation_date = models.DateTimeField(auto_now_add=True)
    text = models.TextField(blank=True, null=True)
    source = EnumField(SubscriberImportSource)
    status = EnumField(
        SubscriberImportStatus, default=SubscriberImportStatus.NOT_STARTED.value
    )
    row_count = models.IntegerField(blank=True, null=True)
    subscriber_count = models.IntegerField(blank=True, null=True)

Historically, I’ve been weirdly reticent to enshrine things in the database if I could do without it. It’s not clear where this impulse came from, but I’m willing to blame a not-so-distant past where coupling things to the database meant incurring a tax: you have to worry about migrations, you have to worry about state, you have to juggle another ball in the air.

And what’s the point of doing that if you can just, say, jam a bunch of strings in a flag (to take the PeacefulEmailAddress case) or paginate over S3 (to take ImageUploadAttempt)?

Well, there are a lot of points:

  1. You get a nice admin interface for free
  2. You get metrics for free through the ORM
  3. You get lifecycle interactions like post-save and pre-delete signals for free
  4. You get organization for free (everything is in a reasonable place! everything has a creation_date and is_deleted attribute!)

I feel like… all of this is obvious in retrospect. Of course the database is a good place to put state; it’s inexpensive in every sense of the word, and I have a lot of code (both that I’ve written and that library authors have written) to make dealing with the database very very easy.

It’s honestly unclear why I wasn’t doing this from day one except, as aforementioned, inaccurate muscle memory. Rewriting a lot of bits of business logic (imports, billing, etc.) to rely on the database has meant more reasonable and more testable code. It makes me curious about what things for which my mental cache is outdated. (The answer is probably front-end unit testing.)

Liked this post? You should subscribe to my newsletter and follow me on Twitter.

(I've got an RSS feed, too, if you'd prefer.)