From a9ce2c0511150af56302b42fb379a1fe144aa483 Mon Sep 17 00:00:00 2001
From: Nick Krichevsky <nick@ollien.com>
Date: Mon, 10 Jul 2023 19:52:39 -0400
Subject: [PATCH] Change posts to use random-ish ids

---
 printpub/post/apps.py                         |  2 +-
 printpub/post/id.py                           | 34 +++++++++++++++++++
 .../post/migrations/0002_alter_post_id.py     | 25 ++++++++++++++
 tests/post/id_test.py                         | 16 +++++++++
 4 files changed, 76 insertions(+), 1 deletion(-)
 create mode 100644 printpub/post/id.py
 create mode 100644 printpub/post/migrations/0002_alter_post_id.py
 create mode 100644 tests/post/id_test.py

diff --git a/printpub/post/apps.py b/printpub/post/apps.py
index 939c84d..2d86189 100644
--- a/printpub/post/apps.py
+++ b/printpub/post/apps.py
@@ -2,5 +2,5 @@ from django.apps import AppConfig
 
 
 class PostConfig(AppConfig):
-    default_auto_field = "django.db.models.BigAutoField"
+    default_auto_field = "printpub.post.id.IDField"
     name = "printpub.post"
diff --git a/printpub/post/id.py b/printpub/post/id.py
new file mode 100644
index 0000000..f896f7e
--- /dev/null
+++ b/printpub/post/id.py
@@ -0,0 +1,34 @@
+import math
+import secrets
+import time
+
+import django.db.models
+
+
+def generate():
+    """
+    Generate an identifier that is "unique" (has some amount of random bits, but is not totally random)
+
+    Inspired heavily by https://instagram-engineering.com/sharding-ids-at-instagram-1cf5a71e5a5c
+    but without the sharding
+    """
+    EPOCH = 1600000000000
+    # Using 42 bits for the time means that we won't see a rollover here for over 200 years
+    # (2**43 minus the time this was written (07/10/23), minus our epoch, gives 276 years in ms).
+    # We're not trying to recover the bits here anyway, though, so perhaps in 300 years this won't matter anyway.
+    TIME_SIZE = 42
+    # The remaining 22 bits can be used for randomness (this means we will have 2**22 ids possible per ms, which is plenty.)
+    random_part = secrets.randbits(64 - TIME_SIZE)
+    now_millis = math.floor(time.time() * 1000)
+
+    return ((now_millis - EPOCH) << (64 - TIME_SIZE)) | random_part
+
+
+class IDField(django.db.models.BigAutoField):
+    """
+    An identifier for a post which is partially random, and partially based on the timestamp.
+    """
+
+    def __init__(self, *args, **kwargs):
+        upd_kwargs = {**kwargs, "default": generate}
+        super().__init__(*args, **upd_kwargs)
diff --git a/printpub/post/migrations/0002_alter_post_id.py b/printpub/post/migrations/0002_alter_post_id.py
new file mode 100644
index 0000000..8f61c9b
--- /dev/null
+++ b/printpub/post/migrations/0002_alter_post_id.py
@@ -0,0 +1,25 @@
+# Generated by Django 4.2.3 on 2023-07-10 23:41
+
+from django.db import migrations
+
+import printpub.post.id
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("post", "0001_initial"),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name="post",
+            name="id",
+            field=printpub.post.id.IDField(
+                auto_created=True,
+                default=printpub.post.id.generate,
+                primary_key=True,
+                serialize=False,
+                verbose_name="ID",
+            ),
+        ),
+    ]
diff --git a/tests/post/id_test.py b/tests/post/id_test.py
new file mode 100644
index 0000000..cfb056c
--- /dev/null
+++ b/tests/post/id_test.py
@@ -0,0 +1,16 @@
+import unittest.mock
+
+import printpub.post.id
+
+
+def test_id_generation_has_unique_and_nonunique_part():
+    now_seconds = 1689031817.358745
+    with unittest.mock.patch("time.time", return_value=now_seconds):
+        id1 = printpub.post.id.generate()
+        id2 = printpub.post.id.generate()
+
+        time_mask = int("1" * 42 + "0" * (64 - 42), 2)
+        # The first 42 bits should be the time
+        assert id1 & time_mask == id2 & time_mask
+        # The last bits should be random
+        assert id1 & (~time_mask) != id2 & (~time_mask)