Arc
Arc is a flexible file upload library for Elixir with straightforward integrations for Amazon S3 and ImageMagick.
Browse the readme below, or jump to a full example.
Content
- Installation
- Getting Started
- Full example
Installation
Add the latest stable release to your mix.exs
file, along with the required dependencies for ExAws
if appropriate:
defp deps do
[
arc: "~> 0.11.0",
# If using Amazon S3:
ex_aws: "~> 2.0",
ex_aws_s3: "~> 2.0",
hackney: "~> 1.6",
poison: "~> 3.1",
sweet_xml: "~> 0.6"
]
end
Then run mix deps.get
in your shell to fetch the dependencies.
Configuration
Arc expects certain properties to be configured at the application level:
config :arc,
storage: Arc.Storage.S3, # or Arc.Storage.Local
bucket: {:system, "AWS_S3_BUCKET"} # if using Amazon S3
Along with any configuration necessary for ExAws.
Storage Providers
Arc ships with integrations for Local Storage and S3. Alternative storage providers may be supported by the community:
- Rackspace - https://github.com/lokalebasen/arc_rackspace
- Manta - https://github.com/onyxrev/arc_manta
- OVH - https://github.com/stephenmoloney/arc_ovh
- Google Cloud Storage - https://github.com/martide/arc_gcs
- Microsoft Azure Storage - https://github.com/phil-a/arc_azure
Usage with Ecto
Arc comes with a companion package for use with Ecto. If you intend to use Arc with Ecto, it is highly recommended you also add the arc_ecto
dependency. Benefits include:
- Changeset integration
- Versioned urls for cache busting (
.../thumb.png?v=63601457477
)
Getting Started: Defining your Upload
Arc requires a definition module which contains the relevant configuration to store and retrieve your files.
This definition module contains relevant functions to determine:
- Optional transformations of the uploaded file
- Where to put your files (the storage directory)
- What to name your files
- How to secure your files (private? Or publicly accessible?)
- Default placeholders
To start off, generate an attachment definition:
mix arc.g avatar
This should give you a basic file in:
web/uploaders/avatar.ex
Check this file for descriptions of configurable options.
Basics
There are two supported use-cases of Arc currently:
- As a general file store, or
- As an attachment to another model (the attached model is referred to as a
scope
)
The upload definition file responds to Avatar.store/1
which accepts either:
- A path to a local file
- A path to a remote
http
orhttps
file - A map with a filename and path keys (eg, a
%Plug.Upload{}
) - A map with a filename and binary keys (eg,
%{filename: "image.png", binary: <<255,255,255,...>>}
) - A two-tuple consisting of one of the above file formats as well as a scope object.
Example usage as general file store:
# Store any locally accessible file
Avatar.store("/path/to/my/file.png") #=> {:ok, "file.png"}
# Store any remotely accessible file
Avatar.store("http://example.com/file.png") #=> {:ok, "file.png"}
# Store a file directly from a `%Plug.Upload{}`
Avatar.store(%Plug.Upload{filename: "file.png", path: "/a/b/c"}) #=> {:ok, "file.png"}
# Store a file from a connection body
{:ok, data, _conn} = Plug.Conn.read_body(conn)
Avatar.store(%{filename: "file.png", binary: data})
Example usage as a file attached to a scope
:
scope = Repo.get(User, 1)
Avatar.store({%Plug.Upload{}, scope}) #=> {:ok, "file.png"}
This scope will be available throughout the definition module to be used as an input to the storage parameters (eg, store files in /uploads/#{scope.id}
).
Transformations
Arc can be used to facilitate transformations of uploaded files via any system executable. Some common operations you may want to take on uploaded files include resizing an uploaded avatar with ImageMagick or extracting a still image from a video with FFmpeg.
To transform an image, the definition module must define a transform/2
function which accepts a version atom and a tuple consisting of the uploaded file and corresponding scope.
This transform handler accepts the version atom, as well as the file/scope argument, and is responsible for returning one of the following:
:noaction
- The original file will be stored as-is.:skip
- Nothing will be stored for the provided version.{executable, args}
- Theexecutable
will be called withSystem.cmd
with the format#{original_file_path} #{args} #{transformed_file_path}
.{executable, fn(input, output) -> args end}
- If your executable expects arguments in a format other than the above, you may supply a function to the conversion tuple which will be invoked to generate the arguments. The arguments can be returned as a string (e.g. –" #{input} -strip -thumbnail 10x10 #{output}"
) or a list (e.g. –[input, "-strip", "-thumbnail", "10x10", output]
) for even more control.{executable, args, output_extension}
- If your transformation changes the file extension (eg, converting topng
), then the new file extension must be explicit.
ImageMagick transformations
As images are one of the most commonly uploaded filetypes, Arc has a recommended integration with ImageMagick's convert
tool for manipulation of images. Each upload definition may specify as many versions as desired, along with the corresponding transformation for each version.
The expected return value of a transform
function call must either be :noaction
, in which case the original file will be stored as-is, :skip
, in which case nothing will be stored, or {:convert, transformation}
in which the original file will be processed via ImageMagick's convert
tool with the corresponding transformation parameters.
The following example stores the original file, as well as a squared 100x100 thumbnail version which is stripped of comments (eg, GPS coordinates):
defmodule Avatar do
use Arc.Definition
@versions [:original, :thumb]
def transform(:thumb, _) do
{:convert, "-strip -thumbnail 100x100^ -gravity center -extent 100x100"}
end
end
Other examples:
# Change the file extension through ImageMagick's `format` parameter:
{:convert, "-strip -thumbnail 100x100^ -gravity center -extent 100x100 -format png", :png}
# Take the first frame of a gif and process it into a square jpg:
{:convert, fn(input, output) -> "#{input}[0] -strip -thumbnail 100x100^ -gravity center -extent 100x100 -format jpg #{output}", :jpg}
For more information on defining your transformation, please consult ImageMagick's convert documentation.
Note: Keep this transformation function simple and deterministic based on the version, file name, and scope object. The
transform
function is subsequently called during URL generation, and the transformation is scanned for the output file format. As such, if you conditionally format the image as apng
orjpg
depending on the time of day, you will be displeased with the result of Arc's URL generation.
System Resources: If you are accepting arbitrary uploads on a public site, it may be prudent to add system resource limits to prevent overloading your system resources from malicious or nefarious files. Since all processing is done directly in ImageMagick, you may pass in system resource restrictions through the -limit flag. One such example might be:
-limit area 10MB -limit disk 100MB
.
FFmpeg transformations
Common transformations of uploaded videos can be also defined through your definition module:
# To take a thumbnail from a video:
{:ffmpeg, fn(input, output) -> "-i #{input} -f jpg #{output}" end, :jpg}
# To convert a video to an animated gif
{:ffmpeg, fn(input, output) -> "-i #{input} -f gif #{output}" end, :gif}
Complex Transformations
Arc
requires the output of your transformation to be located at a predetermined path. However, the transformation may be done completely outside of Arc
. For fine-grained transformations, you should create an executable wrapper in your $PATH (eg. bash script) which takes these proper arguments, runs your transformation, and then moves the file into the correct location.
For example, to use soffice
to convert a doc to an html file, you should place the following bash script in your $PATH:
#!/usr/bin/env sh
# `soffice` doesn't allow for output file path option, and arc can't find the
# temporary file to process and copy. This script has a similar argument list as
# what arc expects. See https://github.com/stavro/arc/issues/77.
set -e
set -o pipefail
function convert {
soffice \
--headless \
--convert-to html \
--outdir $TMPDIR \
"$1"
}
function filter_new_file_name {
awk -F$TMPDIR '{print $2}' \
| awk -F" " '{print $1}' \
| awk -F/ '{print $2}'
}
converted_file_name=$(convert "$1" | filter_new_file_name)
cp $TMPDIR/$converted_file_name "$2"
rm $TMPDIR/$converted_file_name
And perform the transformation as such:
def transform(:html, _) do
{:soffice_wrapper, fn(input, output) -> [input, output] end, :html}
end
Asynchronous File Uploading
If you specify multiple versions in your definition module, each version is processed and stored concurrently as independent Tasks. To prevent an overconsumption of system resources, each Task is given a specified timeout to wait, after which the process will fail. By default this is 15 seconds
.
If you wish to change the time allocated to version transformation and storage, you may add a configuration parameter:
config :arc,
:version_timeout, 15_000 # milliseconds
To disable asynchronous processing, add @async false
to your upload definition.
Storage of files
Arc currently supports Amazon S3 and local destinations for file uploads.
Local Configuration
To store your attachments locally, override the __storage
function in your definition module to Arc.Storage.Local
. You may wish to optionally override the storage directory as well, as outlined below.
defmodule Avatar do
use Arc.Definition
def __storage, do: Arc.Storage.Local # Add this
end
S3 Configuration
ExAws is used to support Amazon S3.
To store your attachments in Amazon S3, you'll need to provide a bucket destination in your application config:
config :arc,
bucket: "uploads"
You may also set the bucket from an environment variable:
config :arc,
bucket: {:system, "S3_BUCKET"}
In addition, ExAws must be configured with the appropriate Amazon S3 credentials.
ExAws has by default the following configuration (which you may override if you wish):
config :ex_aws,
access_key_id: [{:system, "AWS_ACCESS_KEY_ID"}, :instance_role],
secret_access_key: [{:system, "AWS_SECRET_ACCESS_KEY"}, :instance_role]
This means it will first look for the AWS standard AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables, and fall back using instance meta-data if those don't exist. You should set those environment variables to your credentials, or configure an instance that this library runs on to have an iam role.
Storage Directory
Configuration Option
arc[:storage_dir]
- The storage directory to place files. Defaults touploads
, but can be overwritten via configuration options:storage_dir
config :arc,
storage_dir: "my/dir"
The storage dir can also be overwritten on an individual basis, in each separate definition. A common pattern for user profile pictures is to store each user's uploaded images in a separate subdirectory based on their primary key:
def storage_dir(version, {file, scope}) do
"uploads/users/avatars/#{scope.id}"
end
Note: If you are "attaching" a file to a record on creation (eg, while inserting the record at the same time), then you cannot use the model's
id
as a path component. You must either (1) use a different storage path format, such as UUIDs, or (2) attach and update the model after an id has been given.
Note: The storage directory is used for both local filestorage (as the relative or absolute directory), and S3 storage, as the path name (not including the bucket).
Specify multiple buckets
Arc lets you specify a bucket on a per definition basis. In case you want to use multiple buckets, you can specify a bucket in the uploader definition file like this:
def bucket, do: :some_custom_bucket_name
Specify multiple asset hosts
Arc lets you specify an asset host on a per definition basis. In case you want to use multiple hosts, you can specify an asset_host in the uploader definition file like this:
def asset_host, do: "https://example.com"
Access Control Permissions
Arc defaults all uploads to private
. In cases where it is desired to have your uploads public, you may set the ACL at the module level (which applies to all versions):
@acl :public_read
Or you may have more granular control over each version. As an example, you may wish to explicitly only make public a thumbnail version of the file:
def acl(:thumb, _), do: :public_read
Supported access control lists for Amazon S3 are:
ACL | Permissions Added to ACL |
---|---|
:private |
Owner gets FULL_CONTROL . No one else has access rights (default). |
:public_read |
Owner gets FULL_CONTROL . The AllUsers group gets READ access. |
:public_read_write |
Owner gets FULL_CONTROL . The AllUsers group gets READ and WRITE access. Granting this on a bucket is generally not recommended. |
:authenticated_read |
Owner gets FULL_CONTROL . The AuthenticatedUsers group gets READ access. |
:bucket_owner_read |
Object owner gets FULL_CONTROL . Bucket owner gets READ access. |
:bucket_owner_full_control |
Both the object owner and the bucket owner get FULL_CONTROL over the object. |
For more information on the behavior of each of these, please consult Amazon's documentation for Access Control List (ACL) Overview.
S3 Object Headers
The definition module may specify custom headers to pass through to S3 during object creation. The available custom headers include:
- :cache_control
- :content_disposition
- :content_encoding
- :content_length
- :content_type
- :expect
- :expires
- :storage_class
- :website_redirect_location
- :encryption (set to "AES256" for encryption at rest)
As an example, to explicitly specify the content-type of an object, you may define a s3_object_headers/2
function in your definition, which returns a Keyword list, or Map of desired headers.
def s3_object_headers(version, {file, scope}) do
[content_type: MIME.from_path(file.file_name)] # for "image.png", would produce: "image/png"
end
File Validation
While storing files on S3 (rather than your harddrive) eliminates some malicious attack vectors, it is strongly encouraged to validate the extensions of uploaded files as well.
Arc delegates validation to a validate/1
function with a tuple of the file and scope. As an example, to validate that an uploaded file conforms to popular image formats, you may use:
defmodule Avatar do
use Arc.Definition
@extension_whitelist ~w(.jpg .jpeg .gif .png)
def validate({file, _}) do
file_extension = file.file_name |> Path.extname() |> String.downcase()
Enum.member?(@extension_whitelist, file_extension)
end
end
Any uploaded file failing validation will return {:error, :invalid_file}
when passed through to Avatar.store
.
File Names
It may be undesirable to retain original filenames (eg, it may contain personally identifiable information, vulgarity, vulnerabilities with Unicode characters, etc).
You may specify the destination filename for uploaded versions through your definition module.
A common pattern is to combine directories scoped to a particular model's primary key, along with static filenames. (eg: user_avatars/1/thumb.png
)
Examples:
# To retain the original filename, but prefix the version and user id:
def filename(version, {file, scope}) do
file_name = Path.basename(file.file_name, Path.extname(file.file_name))
"#{scope.id}_#{version}_#{file_name}"
end
# To make the destination file the same as the version:
def filename(version, _), do: version
Object Deletion
After an object is stored through Arc, you may optionally remove it. To remove a stored object, pass the same path identifier and scope from which you stored the object.
Example:
# Without a scope:
{:ok, original_filename} = Avatar.store("/Images/me.png")
:ok = Avatar.delete(original_filename)
# With a scope:
user = Repo.get! User, 1
{:ok, original_filename} = Avatar.store({"/Images/me.png", user})
:ok = Avatar.delete({original_filename, user})
# or
user = Repo.get!(User, 1)
{:ok, original_filename} = Avatar.store({"/Images/me.png", user})
user = Repo.get!(User, 1)
:ok = Avatar.delete({user.avatar, user})
Url Generation
Saving your files is only the first half of any decent storage solution. Straightforward access to your uploaded files is equally as important as storing them in the first place.
Often times you will want to regain access to the stored files. As such, Arc
facilitates the generation of urls.
# Given some user record
user = %{id: 1}
Avatar.store({%Plug.Upload{}, user}) #=> {:ok, "selfie.png"}
# To generate a regular, unsigned url (defaults to the first version):
Avatar.url({"selfie.png", user}) #=> "https://bucket.s3.amazonaws.com/uploads/1/original.png"
# To specify the version of the upload:
Avatar.url({"selfie.png", user}, :thumb) #=> "https://bucket.s3.amazonaws.com/uploads/1/thumb.png"
# To generate a signed url:
Avatar.url({"selfie.png", user}, :thumb, signed: true) #=> "https://bucket.s3.amazonaws.com/uploads/1/thumb.png?AWSAccessKeyId=AKAAIPDF14AAX7XQ&Signature=5PzIbSgD1V2vPLj%2B4WLRSFQ5M%3D&Expires=1434395458"
# To generate urls for all versions:
Avatar.urls({"selfie.png", user}) #=> %{original: "https://.../original.png", thumb: "https://.../thumb.png"}
Default url
In cases where a placeholder image is desired when an uploaded file is not present, Arc allows the definition of a default image to be returned gracefully when requested with a nil
file.
def default_url(version) do
MyApp.Endpoint.url <> "/images/placeholders/profile_image.png"
end
Avatar.url(nil) #=> "http://example.com/images/placeholders/profile_image.png"
Avatar.url({nil, scope}) #=> "http://example.com/images/placeholders/profile_image.png"
Virtual Host
To support AWS regions other than US Standard, it may be required to generate urls in the virtual_host
style. This will generate urls in the style: https://#{bucket}.s3.amazonaws.com
instead of https://s3.amazonaws.com/#{bucket}
.
To use this style of url generation, your bucket name must be DNS compliant.
This can be enabled with:
config :arc,
virtual_host: true
When using virtual hosted–style buckets with SSL, the SSL wild card certificate only matches buckets that do not contain periods. To work around this, use HTTP or write your own certificate verification logic.
Asset Host
You may optionally specify an asset host rather than using the default bucket.s3.amazonaws.com
format.
In your application configuration, you'll need to provide an asset_host
value:
config :arc,
asset_host: "https://d3gav2egqolk5.cloudfront.net", # For a value known during compilation
asset_host: {:system, "ASSET_HOST"} # For a value not known until runtime
Alternate S3 configuration example
If you are using a region other than US-Standard, it is necessary to specify the correct configuration for ex_aws
. A full example configuration for both arc and ex_aws is as follows:
config :arc,
bucket: "my-frankfurt-bucket"
config :ex_aws,
access_key_id: "my_access_key_id",
secret_access_key: "my_secret_access_key",
region: "eu-central-1",
s3: [
scheme: "https://",
host: "s3.eu-central-1.amazonaws.com",
region: "eu-central-1"
]
For your host configuration, please examine the approved AWS Hostnames. There are often multiple hostname formats for AWS regions, and it will not work unless you specify the correct one.
Full Example
defmodule Avatar do
use Arc.Definition
@versions [:original, :thumb]
@extension_whitelist ~w(.jpg .jpeg .gif .png)
def acl(:thumb, _), do: :public_read
def validate({file, _}) do
file_extension = file.file_name |> Path.extname |> String.downcase
Enum.member?(@extension_whitelist, file_extension)
end
def transform(:thumb, _) do
{:convert, "-thumbnail 100x100^ -gravity center -extent 100x100 -format png", :png}
end
def filename(version, _) do
version
end
def storage_dir(_, {file, user}) do
"uploads/avatars/#{user.id}"
end
def default_url(:thumb) do
"https://placehold.it/100x100"
end
end
# Given some current_user record
current_user = %{id: 1}
# Store any accessible file
Avatar.store({"/path/to/my/selfie.png", current_user}) #=> {:ok, "selfie.png"}
# ..or store directly from the `params` of a file upload within your controller
Avatar.store({%Plug.Upload{}, current_user}) #=> {:ok, "selfie.png"}
# and retrieve the url later
Avatar.url({"selfie.png", current_user}, :thumb) #=> "https://s3.amazonaws.com/bucket/uploads/avatars/1/thumb.png"
Roadmap
Contributions are welcome. Here is my current roadmap:
- Ease migration for version (or acl) changes
- Alternative storage destinations (eg, Filesystem)
- Solidify public API
Contribution
Open source contributions are welcome. All pull requests must have corresponding unit tests.
To execute all tests locally, make sure the following system environment variables are set prior to running tests (if you wish to test s3_test.exs
)
ARC_TEST_BUCKET
ARC_TEST_S3_KEY
ARC_TEST_S3_SECRET
Then execute mix test
.
License
Copyright 2015 Sean Stavropoulos
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.