Antoine R. Dumont
swh-deposit

Repository



= swh-deposit (draft) =

This is SWH's SWORD Server implementation.

SWORD (Simple Web-Service Offering Repository Deposit) is an
interoperability standard for digital file deposit.

This protocol will be used to interact between a client (a repository)
and a server (swh repository) to permit the deposit of software
tarballs.

In this document, we will refer to a client (e.g. HAL server) and a
server (SWH's).


== Use cases ==

=== First deposit ===

From client's deposit repository server to SWH's repository server
(aka deposit).

1. The client requests for the server's abilities.
(GET query to the *service document uri*)

2. The server answers the client with the service document

3. The client sends the deposit (an archive -> .zip, .tar.gz)
through the deposit *creation uri*.
(one or more POST requests since the archive and metadata can be sent
in multiple requests)


4. The server notifies the client it acknowledged the
client's request. ('http 201 Created' with a deposit receipt id in
the Location header of the response)


=== Updating an existing archive ===

5. Client updates existing archive through the deposit *update uri*
(one or more PUT requests, in effect chunking the artifact to deposit)

=== Deleting an existing archive ===

6. Document deletion will not be implemented,
cf. limitation paragraph for detail

=== Client asks for operation status and repository id ===

NOTE: add specifictions about operation status and injection

== API overview ==

API access is over HTTPS.

service document accessible at:
https://archive.softwareheritage.org/api/1/servicedocument/

API endpoints:

  - without a specific collection, are rooted at
      https://archive.softwareheritage.org/api/1/deposit/.

  - with a specific and unique collection dubbed 'software', are rooted at
      https://archive.softwareheritage.org/api/1/software/.


IMPORTANT: Determine which one of those solutions according to sword possibilities
(cf. 'unclear points' chapter below)

== Limitations ==

Applying the SWORD protocol procedure will result with voluntary implementation
shortcomings during the first iteration:

- upload limitation of 200Mib
- only tarballs (.zip, .tar.gz) will be accepted
- no removal (implementation-wise, this will possibly be a means
  to hide the origin).
- no mediation (we do not know the other system's users)
- basic http authentication enforced at the application layer
  on a per client basis (authentication:
  http://swordapp.github.io/SWORDv2-Profile/SWORDProfile.html#authenticationmediateddeposit)

== Unclear points ==

- SWORD defines a 'collection' concept. should we apply the 'collection' concept
   even thought SWH is software archive having one 'software' collection?
   - option A:
      The collection refers to a group of documents to which the document sent
      (aka deposit) is part of. In this process with HAL, HAL is the collection,
      maybe tomorrow we will do the same with MIT and MIT could be
      the collection (the logic of the answer above is a result of this
      link: https://hal.inria.fr/USPC  for the USPC collection)

      **result**: 1 client being equivalent as 1 collection in this case.
        The is client pushes us software in 'their' one collection.
        The collection name could show up in the uri endpoint.

    - option B:
      Define none? (is it possible? i don't think it is due to the service
      document part listing the collection to act upon...)

      **result**: the deposited software has no other entry point via
      collection name


== <a name="scenarios"> Scenarios ==
=== 1. Client request for Service Document ===

This is the endpoint permitting the client to ask the server's abilities.


==== API endpoint ====

GET api/1/servicedocument/

Answer:
> 200, Content-Type: application/atomserv+xml: OK, with the body
  described below

==== Sample request:====

```lang=shell
GET https://archive.softwareheritage.org/api/1/servicedocument HTTP/1.1
Host: archive.softwareheritage.org
```

=== 2. Sever respond for Service Document ===

The server returns its abilities with the service document in xml format:
- protocol sword version v2
- accepted mime types: application/zip, application/gzip
- upload max size accepted, beyond that, it's expected the client
  chunk the tarball into multiple ones
- the collections the client can act upon (swh supports only one software collection)
- mediation not supported

==== Sample answer:====
``` lang=xml
<?xml version="1.0" ?>
<service xmlns:dcterms="http://purl.org/dc/terms/"
    xmlns:sword="http://purl.org/net/sword/terms/"
    xmlns:atom="http://www.w3.org/2005/Atom"
    xmlns="http://www.w3.org/2007/app">

    <sword:version>2.0</sword:version>
    <sword:maxUploadSize>${max_upload_size}</sword:maxUploadSize>

    <workspace>
        <atom:title>The SWH archive</atom:title>

        <collection href="https://archive.softwareherigage.org/api/1/deposit/">
            <atom:title>SWH Collection</atom:title>
            <accept>application/gzip</accept>
            <accept alternate="multipart-related">application/gzip</accept>
            <dcterms:abstract>Software Heritage Archive Deposit</dcterms:abstract>
            <sword:mediation>false</sword:mediation>
            <sword:acceptPackaging>http://purl.org/net/sword/package/SimpleZip</sword:acceptPackaging>
        </collection>
    </workspace>
</service>
```


== Deposit Creation: client point of view ==

Process of deposit creation:

-> [3] client request

  - [3.1] server validation
  - [3.2] server temporary upload
  - [3.3] server injects deposit into archive*

<- [4] server returns deposit receipt id


NOTE: [3.3] Asynchronously, the server will inject the archive uploaded and the
associated metadata. The operation status mentioned
earlier is a reference to that injection operation.

The image bellow represent only the communication and creation of
a deposit:
{F2403754}

=== [3] client request  ===

The client can send a deposit through one request deposit or multiple
requests deposit.

The deposit can contain:
- an archive holding the software source code (binary upload)
- an envelop with metadata describing information regarding a deposit
  (atom entry deposit)
- or both (multipart deposit, exactly one archive and one envelop).

the client can deposit a binary file, supplying the following headers:
- Content-Type (text): accepted mimetype
- Content-Length (int): tarball size
- Content-MD5 (text): md5 checksum hex encoded of the tarball
- Content-Disposition (text): attachment; filename=[filename] ; the filename
  parameter must be text (ascii)
- Packaging (IRI): http://purl.org/net/sword/package/SimpleZip
- In-Progress (bool): true to specify it's not the last request, false
  to specify it's a final request and the server can go on with
  processing the request's information.

WARNING: if In-Progress is not present the server MUST assume that it is false

==== API endpoint ====

POST /api/1/deposit/<client-name>

=== Archive deposit ===

This is a single zip archive deposit. Almost no metadata is associated
with the archive except for the unique external identifier.

Note: This kind of deposit should be partial (In-Progress: True) as
almost no metadata can be associated with the uploaded archive.

==== sample request for binary upload deposit ====

```
curl -i --data-binary @swh/deposit.zip \
    -H 'In-Progress: false' -H 'Content-MD5: 0faa1ecbf9224b9bf48a7c691b8c2b6f' \
    -H 'Content-Disposition: attachment; filename=[deposit.zip]' \
    -H 'Slug: some-external-id' \
    -H 'Packaging: http://purl.org/net/sword/package/SimpleZIP' \
    -H 'Content-type: application/zip' \
    -XPOST http://127.0.0.1:8000/deposit/hal
```

=== Atom entry deposit ===

TBD

Note: This kind of deposit should be partial (In-Progress: True) since
no archive will be associated to those metadata.

==== One request deposit / Multipart deposit ====

The one request deposit is a single request containing both the
metadata (as atom entry attachment) and the archive (as payload
attachment). Thus, it is a multipart deposit.

Client provides:
- Content-Disposition (text): header of type 'attachment' on the Entry
  Part with a name parameter set to 'atom'
- Content-Disposition (text): header of type 'attachment' on the Media
  Part with a name parameter set to payload and a filename parameter
  (the filename will be expressed in ASCII).
- Content-MD5 (text): md5 checksum hex encoded of the tarball
- Packaging (text): http://purl.org/net/sword/package/SimpleZip
  (packaging format used on the Media Part)
- In-Progress (bool): true|false; true means partial upload and we can expect
  other requests in the future, false means the deposit is done.
- add metadata formats or foreign markup to the atom:entry element

==== sample request for multipart deposit: ====

``` lang=xml
POST deposit HTTP/1.1
Host: archive.softwareheritage.org
Content-Length: [content length]
Content-Type: multipart/related;
            boundary="===============1605871705==";
            type="application/atom+xml"
In-Progress: false
MIME-Version: 1.0

Media Post
--===============1605871705==
Content-Type: application/atom+xml; charset="utf-8"
Content-Disposition: attachment; name="atom"
MIME-Version: 1.0

<?xml version="1.0"?>
<entry xmlns="http://www.w3.org/2005/Atom"
        xmlns:dcterms="http://purl.org/dc/terms/">
    <title>Title</title>
    <id>hal-or-other-archive-id</id>
    <updated>2005-10-07T17:17:08Z</updated>
    <author><name>Contributor</name></author>

    <!-- some embedded metadata TO BE DEFINED -->

</entry>
--===============1605871705==
Content-Type: application/zip
Content-Disposition: attachment; name=payload; filename=[filename]
Packaging: http://purl.org/net/sword/package/SimpleZip
Content-MD5: [md5-digest]
MIME-Version: 1.0

[...binary package data...]
--===============1605871705==--
```

== Deposit Creation - server point of view ==

The server receives the request and:

=== [3.1] Validation of the header and body request ===


=== [3.2] Server uploads the content in a temporary location ==
(deposit table in a separated DB).
- saves the archives in a temporary location
- executes a md5 checksum on that archive and check it against the
  same header information
- adds a deposit entry and retrieves the associated id


=== [4] Servers answers the client ===
an 'http 201 Created' with a deposit receipt id in the Location header of
the response.

The server possible answers are:
- OK: '201 created' + one header 'Location' holding the deposit receipt
  id
- KO: with the error status code and associated message
  (cf. [possible errors paragraph](#possible errors)).


=== [5] Deposit Update ===

The client previously uploaded an archive and wants to add either new
metadata information or a new version for that previous deposit
(possibly in multiple steps as well).  The important thing to note
here is that for swh, this will result in a new version of the
previous deposit in any case.

Providing the identifier of the previous version deposit received from
the status URI, the client executes a PUT request on the same URI as
the deposit one.

After validation of the body request, the server:
- uploads such content in a temporary location (to be defined).

- answers the client an 'http 204 (No content)'. In the Location
  header of the response lies a deposit receipt id permitting the
  client to check back the operation status later on.

- Asynchronously, the server will inject the archive uploaded and the
  associated metadata. The operation status mentioned earlier is a
  reference to that injection operation. The fact that the version is
  a new one is dealt with at the injection level.

  URL: PUT /1/deposit/<previous-swh-id>

=== [6] Deposit Removal ===

[#limitation](As explained in the limitation paragraph), removal won't
be implemented.  Nothing is removed from the SWH archive.

The server answers a '405 Method not allowed' error.

=== Operation Status ===

Providing a deposit receipt id, the client asks the operation status
of a prior upload.

  URL: GET /1/collection/{deposit_receipt}

or

  GET /1/deposit/{deposit_receipt}

NOTE: depends of the decision taken about collections

## <a name="errors"> Possible errors

### sword:ErrorContent

IRI: http://purl.org/net/sword/error/ErrorContent

The supplied format is not the same as that identified in the
Packaging header and/or that supported by the server Associated HTTP

Status: 415 (Unsupported Media Type) or 406 (Not Acceptable)

### sword:ErrorChecksumMismatch

IRI: http://purl.org/net/sword/error/ErrorChecksumMismatch

Checksum sent does not match the calculated checksum. The server MUST
also return a status code of 412 Precondition Failed

### sword:ErrorBadRequest

IRI: http://purl.org/net/sword/error/ErrorBadRequest

Some parameters sent with the POST/PUT were not understood. The server
MUST also return a status code of 400 Bad Request.

### sword:MediationNotAllowed

IRI: http://purl.org/net/sword/error/MediationNotAllowed

Used where a client has attempted a mediated deposit, but this is not
supported by the server. The server MUST also return a status code of
412 Precondition Failed.

### sword:MethodNotAllowed

IRI: http://purl.org/net/sword/error/MethodNotAllowed

Used when the client has attempted one of the HTTP update verbs (POST,
PUT, DELETE) but the server has decided not to respond to such
requests on the specified resource at that time. The server MUST also
return a status code of 405 Method Not Allowed

### sword:MaxUploadSizeExceeded

IRI: http://purl.org/net/sword/error/MaxUploadSizeExceeded

Used when the client has attempted to supply to the server a file
which exceeds the server's maximum upload size limit

Associated HTTP Status: 413 (Request Entity Too Large)

---------------

== Tarball Injection ==

Providing we use indeed synthetic revision to represent a version of a
tarball injected through the sword use case, this needs to be improved
so that the synthetic revision is created with a parent revision (the
previous known one for the same 'origin').


=== Injection mapping ===
| origin                              |      https://hal.inria.fr/hal-id      |
|-------------------------------------|---------------------------------------|
| origin_visit                        |           1 :reception_date           |
| occurrence &amp; occurrence_history | branch: client's version n° (e.g hal) |
| revision                            |      synthetic_revision (tarball)     |
| directory                           | upper level of the uncompressed archive|


=== Questions raised concerning injection: ===
- A deposit has one origin, yet an origin can have multiple deposits ?

No, an origin can have multiple requests for the same deposit,
which should end up in one single deposit (when the client pushes its final
request saying deposit 'done' through the header In-Progress).

When an update of a deposit is requested,
the new version is identified with the external_id.

Illustration First deposit injection:

HAL's deposit 01535619 = SWH's deposit **01535619-1**

    + 1 origin with url:https://hal.inria.fr/medihal-01535619

    + 1 synthetic revision

    + 1 directory

HAL's update on deposit 01535619 = SWH's deposit **01535619-2**

(*with HAL updates can only be on the metadata and a new version is required
if the content changes)
    + 1 origin with url:https://hal.inria.fr/medihal-01535619

    + new synthetic revision (with new metadata)

    + same directory

HAL's deposit 01535619-v2 = SWH's deposit **01535619-v2-1**

    + same origin

    + new revision

    + new directory


== Technical details ==

We will need:
- one dedicated db to store state - swh-deposit

- one dedicated temporary storage to store archives before injection

- one client to test the communication with SWORD protocol

=== Deposit reception schema ===

- **deposit** table:
  - id (bigint): deposit receipt id

  - external id (text): client's internal identifier (e.g hal's id, etc...).

  - origin id : null before injection
  - swh_id : swh identifier result once the injection is complete

  - reception_date: first deposit date

  - complete_date: reception date of the last deposit which makes the deposit
  complete

  - status (enum):
```
      'partial',      -- the deposit is new or partially received since it
                      -- can be done in multiple requests
      'expired',      -- deposit has been there too long and is now deemed
                      -- ready to be garbage collected
      'ready',        -- deposit is fully received and ready for injection
      'scheduled',    -- injection is scheduled on swh's side
      'success',      -- injection successful
      'failure'       -- injection failure
```
- **deposit_request** table:
  - id (bigint): identifier
  - deposit_id: deposit concerned by the request
  - metadata: metadata associated to the request

- **client** table:
  - id (bigint): identifier
  - name (text): client's name (e.g HAL)
  - credentials


All metadata (declared metadata) are stored in deposit_request (with the
request they were sent with).
When the deposit is complete metadata fields are aggregated and sent
to injection. During injection the metadata is kept in the
origin_metadata table (see [metadata injection](#metadata-injection)).

The only update actions occurring on the deposit table are in regards of:
  - status changing
    - partial -> {expired/ready},
    - ready -> scheduled,
    - scheduled -> {success/failure}
  - complete_date when the deposit is finalized
  (when the status is changed to ready)
  - swh-id being populated once we have the result of the injection

==== SWH Identifier returned? ====

    swh-<client-name>-<synthetic-revision-id>

    e.g: swh-hal-47dc6b4636c7f6cba0df83e3d5490bf4334d987e

    We could have a specific dedicated 'client' table to reference client
    identifier.

=== Scheduling injection ===
All data and metadata separated with multiple requests should be aggregated
before injection.

TODO: injection modeling

=== Metadata injection ===
- the metadata received with the deposit should be kept in the origin_metadata
table before translation as part of the injection process and a indexation
process should be scheduled.

origin_metadata table:
```
origin                                  bigint        PK FK
discovery_date                          date          PK FK
translation_date                        date          PK FK
provenance_type                         text                  // (enum: 'publisher', 'lister' needs to be completed)
raw_metadata                            jsonb                 // before translation
indexer_configuration_id                bigint            FK  // tool used for translation
translated_metadata                     jsonb                 // with codemeta schema and terms
```

== Nomenclature ==

SWORD uses IRI. This means Internationalized Resource Identifier. In
this chapter, we will describe SWH's IRI.

=== SD-IRI - The Service Document IRI ===

This is the IRI from which the root service document can be
located.

=== Col-IRI - The Collection IRI ===

Only one collection of software is used in this repository.

NOTE:
This is the IRI to which the initial deposit will take place, and
which are listed in the Service Document.
Discuss to check if we want to implement this or not.

=== Cont-IRI - The Content IRI ===

This is the IRI from which the client will be able to retrieve
representations of the object as it resides in the SWORD server.

=== EM-IRI - The Atom Edit Media IRI ===

To simplify, this is the same as the Cont-IRI.

=== Edit-IRI - The Atom Entry Edit IRI ===

This is the IRI of the Atom Entry of the object, and therefore also of
the container within the SWORD server.

=== SE-IRI - The SWORD Edit IRI ===

This is the IRI to which clients may POST additional content to an
Atom Entry Resource. This MAY be the same as the Edit-IRI, but is
defined separately as it supports HTTP POST explicitly while the
Edit-IRI is defined by [AtomPub] as limited to GET, PUT and DELETE
operations.

=== State-IRI - The SWORD Statement IRI ===

This is the one of the IRIs which can be used to retrieve a
description of the object from the sword server, including the
structure of the object and its state. This will be used as the
operation status endpoint.

== Sources ==

- [SWORD v2 specification](http://swordapp.github.io/SWORDv2-Profile/SWORDProfile.html)
- [arxiv documentation](https://arxiv.org/help/submit_sword)
- [Dataverse example](http://guides.dataverse.org/en/4.3/api/sword.html)
- [SWORD used on HAL](https://api.archives-ouvertes.fr/docs/sword)
- [xml examples for CCSD](https://github.com/CCSDForge/HAL/tree/master/Sword)