Documentation
¶
Overview ¶
Package s3store provides a storage backend using AWS S3 or compatible servers.
Configuration ¶
In order to allow this backend to function properly, the user accessing the bucket must have at least following AWS IAM policy permissions for the bucket and all of its subresources:
s3:AbortMultipartUpload s3:DeleteObject s3:GetObject s3:ListMultipartUploadParts s3:PutObject
While this package uses the official AWS SDK for Go, S3Store is able to work with any S3-compatible service such as Riak CS. In order to change the HTTP endpoint used for sending requests to, consult the AWS Go SDK (http://docs.aws.amazon.com/sdk-for-go/api/aws/Config.html#WithEndpoint-instance_method).
Implementation ¶
Once a new tus upload is initiated, multiple objects in S3 are created:
First of all, a new info object is stored which contains a JSON-encoded blob of general information about the upload including its size and meta data. This kind of objects have the suffix ".info" in their key.
In addition a new multipart upload (http://docs.aws.amazon.com/AmazonS3/latest/dev/uploadobjusingmpu.html) is created. Whenever a new chunk is uploaded to tusd using a PATCH request, a new part is pushed to the multipart upload on S3.
If meta data is associated with the upload during creation, it will be added to the multipart upload and after finishing it, the meta data will be passed to the final object. However, the metadata which will be attached to the final object can only contain ASCII characters and every non-ASCII character will be replaced by a question mark (for example, "Menü" will be "Men?"). However, this does not apply for the metadata returned by the GetInfo function since it relies on the info object for reading the metadata. Therefore, HEAD responses will always contain the unchanged metadata, Base64- encoded, even if it contains non-ASCII characters.
Once the upload is finished, the multipart upload is completed, resulting in the entire file being stored in the bucket. The info object, containing meta data is not deleted. It is recommended to copy the finished upload to another bucket to avoid it being deleted by the Termination extension.
If an upload is about to being terminated, the multipart upload is aborted which removes all of the uploaded parts from the bucket. In addition, the info object is also deleted. If the upload has been finished already, the finished object containing the entire upload is also removed.
Considerations ¶
In order to support tus' principle of resumable upload, S3's Multipart-Uploads are internally used. For each incoming PATCH request (a call to WriteChunk), a new part is uploaded to S3. However, each part of a multipart upload, except the last one, must be 5MB or bigger. This introduces a problem, since in tus' perspective it's totally fine to upload just a few kilobytes in a single request.
Therefore, a few special conditions have been implemented:
Each PATCH request must contain a body of, at least, 5MB. If the size is smaller than this limit, the entire request will be dropped and not even passed to the storage server. If your server supports a different limit, you can adjust this value using S3Store.MinPartSize.
When receiving a PATCH request, its body will be temporarily stored on disk. This requirement has been made to ensure the minimum size of a single part and to allow the calculating of a checksum. Once the part has been uploaded to S3, the temporary file will be removed immediately. Therefore, please ensure that the server running this storage backend has enough disk space available to hold these caches.
In addition, it must be mentioned that AWS S3 only offers eventual consistency (https://docs.aws.amazon.com/AmazonS3/latest/dev/Introduction.html#ConsistencyModel). Therefore, it is required to build additional measurements in order to prevent concurrent access to the same upload resources which may result in data corruption. See tusd.LockerDataStore for more information.
Index ¶
- type S3API
- type S3Store
- func (store S3Store) ConcatUploads(dest string, partialUploads []string) error
- func (store S3Store) FinishUpload(id string) error
- func (store S3Store) GetInfo(id string) (info tusd.FileInfo, err error)
- func (store S3Store) GetReader(id string) (io.Reader, error)
- func (store S3Store) NewUpload(info tusd.FileInfo) (id string, err error)
- func (store S3Store) Terminate(id string) error
- func (store S3Store) UseIn(composer *tusd.StoreComposer)
- func (store S3Store) WriteChunk(id string, offset int64, src io.Reader) (int64, error)
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type S3API ¶
type S3API interface {
PutObject(input *s3.PutObjectInput) (*s3.PutObjectOutput, error)
ListParts(input *s3.ListPartsInput) (*s3.ListPartsOutput, error)
UploadPart(input *s3.UploadPartInput) (*s3.UploadPartOutput, error)
GetObject(input *s3.GetObjectInput) (*s3.GetObjectOutput, error)
CreateMultipartUpload(input *s3.CreateMultipartUploadInput) (*s3.CreateMultipartUploadOutput, error)
AbortMultipartUpload(input *s3.AbortMultipartUploadInput) (*s3.AbortMultipartUploadOutput, error)
DeleteObjects(input *s3.DeleteObjectsInput) (*s3.DeleteObjectsOutput, error)
CompleteMultipartUpload(input *s3.CompleteMultipartUploadInput) (*s3.CompleteMultipartUploadOutput, error)
UploadPartCopy(input *s3.UploadPartCopyInput) (*s3.UploadPartCopyOutput, error)
}
type S3Store ¶
type S3Store struct {
// Bucket used to store the data in, e.g. "tusdstore.example.com"
Bucket string
// Service specifies an interface used to communicate with the S3 backend.
// Usually, this is an instance of github.com/aws/aws-sdk-go/service/s3.S3
// (http://docs.aws.amazon.com/sdk-for-go/api/service/s3/S3.html).
Service S3API
// MaxPartSize specifies the maximum size of a single part uploaded to S3
// in bytes. This value must be bigger than MinPartSize! In order to
// choose the correct number, two things have to be kept in mind:
//
// If this value is too big and uploading the part to S3 is interrupted
// expectedly, the entire part is discarded and the end user is required
// to resume the upload and re-upload the entire big part. In addition, the
// entire part must be written to disk before submitting to S3.
//
// If this value is too low, a lot of requests to S3 may be made, depending
// on how fast data is coming in. This may result in an eventual overhead.
MaxPartSize int64
// MinPartSize specifies the minimum size of a single part uploaded to S3
// in bytes. This number needs to match with the underlying S3 backend or else
// uploaded parts will be reject. AWS S3, for example, uses 5MB for this value.
MinPartSize int64
// MaxMultipartParts is the maximum number of parts an S3 multipart upload is
// allowed to have according to AWS S3 API specifications.
// See: http://docs.aws.amazon.com/AmazonS3/latest/dev/qfacts.html
MaxMultipartParts int64
// MaxObjectSize is the maximum size an S3 Object can have according to S3
// API specifications. See link above.
MaxObjectSize int64
}
See the tusd.DataStore interface for documentation about the different methods.
func (S3Store) ConcatUploads ¶
func (S3Store) FinishUpload ¶
func (S3Store) UseIn ¶
func (store S3Store) UseIn(composer *tusd.StoreComposer)
UseIn sets this store as the core data store in the passed composer and adds all possible extension to it.
Source Files