migrations: Fix migration 0576 local backfill path and repeat.

Unfortunately, because this migration has already been run for many
installations, we need to ship another copy of the migration.

It should be a noop when repeated.

(cherry picked from commit 66537c8bf8)

zerver/migrations/0576_backfill_imageattachment.py

@@ -77,7 +77,8 @@ def backfill_imageattachment(apps: StateApps, schema_editor: BaseDatabaseSchemaE
                 source: pyvips.Source = pyvips.SourceCustom()
                 source.on_read(partial(s3_read, metadata["Body"]))
             else:
-                attachment_path = os.path.join(settings.LOCAL_UPLOADS_DIR, attachment.path_id)
+                assert settings.LOCAL_FILES_DIR is not None
+                attachment_path = os.path.join(settings.LOCAL_FILES_DIR, attachment.path_id)
                 if not os.path.exists(attachment_path):
                     print(f"{attachment.path_id}: Missing!")
                     continue

zerver/migrations/0622_backfill_imageattachment_again.py

@@ -0,0 +1,131 @@
+# Duplicate of database migration 0576, because the original used the
+# wrong path for servers using the local file upload backend, and
+# many servers had already upgraded to 9.2 where it was backported.
+
+import os
+from functools import reduce
+from operator import or_
+
+import boto3
+import pyvips
+from botocore.client import Config
+from botocore.exceptions import ClientError
+from botocore.response import StreamingBody
+from django.conf import settings
+from django.db import migrations
+from django.db.backends.base.schema import BaseDatabaseSchemaEditor
+from django.db.migrations.state import StateApps
+from django.db.models import Exists, OuterRef, Q
+
+from zerver.lib.partial import partial
+
+
+def backfill_imageattachment(apps: StateApps, schema_editor: BaseDatabaseSchemaEditor) -> None:
+    ImageAttachment = apps.get_model("zerver", "ImageAttachment")
+    Attachment = apps.get_model("zerver", "Attachment")
+
+    if settings.LOCAL_UPLOADS_DIR is None:
+        upload_bucket = boto3.resource(
+            "s3",
+            aws_access_key_id=settings.S3_KEY,
+            aws_secret_access_key=settings.S3_SECRET_KEY,
+            region_name=settings.S3_REGION,
+            endpoint_url=settings.S3_ENDPOINT_URL,
+            config=Config(
+                signature_version=None,
+                s3={"addressing_style": settings.S3_ADDRESSING_STYLE},
+            ),
+        ).Bucket(settings.S3_AUTH_UPLOADS_BUCKET)
+
+    # Historical attachments do not have a mime_type value, so we used
+    # to rely on the file extension.  We replicate that when
+    # backfilling.  This is the value from zerver.lib.markdown:
+    IMAGE_EXTENSIONS = [".bmp", ".gif", ".jpe", ".jpeg", ".jpg", ".png", ".webp"]
+
+    extension_limits = Q()
+    extension_limits = reduce(
+        or_,
+        [Q(file_name__endswith=extension) for extension in IMAGE_EXTENSIONS],
+        extension_limits,
+    )
+
+    min_id: int | None = 0
+    while True:
+        attachments = (
+            Attachment.objects.alias(
+                has_imageattachment=Exists(
+                    ImageAttachment.objects.filter(path_id=OuterRef("path_id"))
+                )
+            )
+            .filter(extension_limits, has_imageattachment=False, id__gt=min_id)
+            .order_by("id")
+        )[:1000]
+
+        min_id = None
+        for attachment in attachments:
+            min_id = attachment.id
+
+            if settings.LOCAL_UPLOADS_DIR is None:
+                try:
+                    metadata = upload_bucket.Object(attachment.path_id).get()
+                except ClientError:
+                    print(f"{attachment.path_id}: Missing!")
+                    continue
+
+                def s3_read(streamingbody: StreamingBody, size: int) -> bytes:
+                    return streamingbody.read(amt=size)
+
+                # We use the streaming body to only pull down as much
+                # of the image as we need to examine the headers --
+                # generally about 40k
+                source: pyvips.Source = pyvips.SourceCustom()
+                source.on_read(partial(s3_read, metadata["Body"]))
+            else:
+                assert settings.LOCAL_FILES_DIR is not None
+                attachment_path = os.path.join(settings.LOCAL_FILES_DIR, attachment.path_id)
+                if not os.path.exists(attachment_path):
+                    print(f"{attachment.path_id}: Missing!")
+                    continue
+                source = pyvips.Source.new_from_file(attachment_path)
+            try:
+                image = pyvips.Image.new_from_source(source, "", access="sequential")
+
+                # "original_width_px" and "original_height_px" here are
+                # _as rendered_, after applying the orientation
+                # information which the image may contain.
+                if (
+                    "orientation" in image.get_fields()
+                    and image.get("orientation") >= 5
+                    and image.get("orientation") <= 8
+                ):
+                    (width, height) = (image.height, image.width)
+                else:
+                    (width, height) = (image.width, image.height)
+
+                ImageAttachment.objects.create(
+                    realm_id=attachment.realm_id,
+                    path_id=attachment.path_id,
+                    original_width_px=width,
+                    original_height_px=height,
+                    frames=image.get_n_pages(),
+                    thumbnail_metadata=[],
+                )
+            except pyvips.Error:
+                pass
+
+        if min_id is None:
+            break
+
+
+class Migration(migrations.Migration):
+    atomic = False
+    dependencies = [
+        # Because this will be backported to 9.x, we only depend on the last migration in 9.x
+        ("zerver", "0576_backfill_imageattachment"),
+    ]
+
+    operations = [
+        migrations.RunPython(
+            backfill_imageattachment, reverse_code=migrations.RunPython.noop, elidable=True
+        )
+    ]