Files
summercms/docs/database/attachments.md
Jakub Zych efb35a2d35 feat(11.1-04): add the Database section and the core Services pages
- docs/database: models, migrations, queries and pagination, relations,
  casts and validation, attachments and transactions (lagoon.Transaction,
  lagoon.AfterCommit, nested savepoints, lagoon.OnDatabase)
- docs/services: configuration, events, routing with auth groups, rate
  limiting, authentication, the OAuth server, mail and localization
- runnable Examples for lagoon, attach, compass, surf, wire, bouncer,
  wristband, postcard, phrasebook and festival; lagoon TestDocs* regions
  run on the package's Postgres harness through DocsDB
- 15 new required pages
2026-09-30 22:59:25 +02:00

4.5 KiB

title, description, section, order
title description section order
Attachments Attach files to models through WinterCMS-compatible system_files rows, serve originals and thumbnails, and delete blobs only after the transaction commits. database 60

Attachments

WinterCMS attaches files to models with $attachOne and $attachMany, storing a row per file in system_files and the bytes on a disk. The attach package of lagoon ports the same table and storage layout, so files uploaded to a WinterCMS site keep working after a data copy. The bytes live in a gocloud.dev bucket, configured with the storage.uploads.* keys.

The system_files row

attach.File is the system_files row. It links a file to its owner with the WinterCMS polymorphic columns: AttachmentType holds the owner's morph name and AttachmentID its ID, as a string, and Field names the relation, such as cover or gallery. The system_files table is created by the framework migrations, so a plugin does not migrate it.

An owner model implements attach.Owner. Its MorphName returns the PHP class name, so the attachment_type values copied from WinterCMS still match:

func (Post) MorphName() string { return `Acme\Blog\Models\Post` }

Storing and serving files

The original is stored under WinterCMS's partitioned key: attach.PartitionDirectory splits the first nine characters of the random disk_name into three directories, and attach.BlobKey appends the name. attach.File.Thumb returns the public URL of a thumbnail, generating it in the same partition on first use and reusing it afterwards:

ctx := context.Background()
dir, err := os.MkdirTemp("", "acme-config")
if err != nil {
	fmt.Println(err)
	return
}
defer os.RemoveAll(dir)

// storage.uploads.bucket_url is a file:// URL in production; mem:// keeps
// the example in memory.
cfg, err := compass.Open(compass.Options{Dir: dir, Env: "development", Environ: []string{}})
if err != nil {
	fmt.Println(err)
	return
}
_ = cfg.Set("storage.uploads.bucket_url", "mem://")
bucket, err := attach.OpenBucket(ctx, cfg)
if err != nil {
	fmt.Println(err)
	return
}
defer bucket.Close()

// The system_files row of a post's cover image.
var owner attach.Owner = Post{ID: 1}
f := attach.File{
	ID:             7,
	DiskName:       "5f1d0c2e9a7b4c3d8e6f.jpg",
	FileName:       "cover.jpg",
	ContentType:    "image/jpeg",
	Field:          "cover",
	AttachmentType: owner.MorphName(),
	AttachmentID:   "1",
}

// The original is stored under its partitioned WinterCMS key.
var img bytes.Buffer
if err := jpeg.Encode(&img, image.NewRGBA(image.Rect(0, 0, 640, 480)), nil); err != nil {
	fmt.Println(err)
	return
}
if err := bucket.WriteAll(ctx, attach.BlobKey(f.DiskName), img.Bytes(), nil); err != nil {
	fmt.Println(err)
	return
}
fmt.Println(attach.BlobKey(f.DiskName))

// A thumbnail is generated on first use and reused afterwards.
url, err := f.Thumb(ctx, bucket, 200, 200, "crop")
if err != nil {
	fmt.Println(err)
	return
}
fmt.Println(url)
// Output:
// 5f1/d0c/2e9/5f1d0c2e9a7b4c3d8e6f.jpg
// /storage/uploads/5f1/d0c/2e9/thumb_7_200_200_0_0_crop.jpg

URLs start with storage.uploads.public_path_prefix (/storage/uploads by default). attach.StaticHandler serves originals and thumbnails under that prefix; attach.StaticHandlerPublic does the same and answers 404 for a row whose is_public flag is false. Mount the gated handler when a bucket holds any private file.

Warning

Serve uploads from a separate origin, or at least never mount the ungated handler on the application's own origin. An uploaded file served with its own content type from the API's origin can run script in that origin.

Deleting files after commit

A rolled-back transaction can restore a row but not the bytes of a deleted blob. Deleting an owner's files is therefore split in two:

  1. Inside the transaction that deletes the owner, attach.DeleteForOwner deletes the owner's system_files rows and passes their blob keys to a callback. The callback only records the keys; it must not delete anything.
  2. After the transaction commits, attach.DeleteKeys deletes the originals and their thumbnails from the bucket.

A soft-deleted owner keeps its rows and files, so only a force delete (Unscoped().Delete) runs this. lagoon.AfterCommit is a natural place for the second step; see Transactions.