Synnefo

.. _dev-guide:

Synnefo Developer's Guide

^^^^^^^^^^^^^^^^^^^^^^^^^

This is the complete Synnefo Developer's Guide

Tying it all up with kamaki

===========================

kamaki

------

IM API (Astakos)

================

This is the Identity Management API:

.. toctree::

   :maxdepth: 2

   IM API <astakos-api-guide>

Compute API (Cyclades)

======================

This is the Cyclades Compute API:

.. toctree::

   :maxdepth: 2

   Compute API <cyclades-api-guide>

Network API (Cyclades)

======================

The Network API is currently implemented inside Cyclades. Please consult the

:ref:`Compute API <cyclades-api-guide>` for more details.

Images API (Plankton)

=====================

This is the Plankton Image API:

.. toctree::

   :maxdepth: 2

   Image API <plankton-api-guide>

Storage API (Pithos+)

=====================

This is the Pithos+ Object Storage API:

.. toctree::

   :maxdepth: 2

   Object Storage API <pithos-api-guide>

Implementing new clients

========================

In this section we discuss implementation guidelines, that a developer should

take into account before writing his own client for the above APIs. Before,

starting your client implementation, make sure you have thoroughly read the

corresponding Synnefo API.

Pithos+ clients

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

User Experience

~~~~~~~~~~~~~~~

Hopefully this API will allow for a multitude of client implementations, each

supporting a different device or operating system. All clients will be able to

manipulate containers and objects - even software only designed for OOS API

compatibility. But a Pithos interface should not be only about showing

containers and folders. There are some extra user interface elements and

functionalities that should be common to all implementations.

Upon entrance to the service, a user is presented with the following elements -

which can be represented as folders or with other related icons:

 * The ``home`` element, which is used as the default entry point to the user's

   "files". Objects under ``home`` are represented in the usual hierarchical

   organization of folders and files.

 * The ``trash`` element, which contains files that have been marked for

   deletion, but can still be recovered.

 * The ``shared`` element, which contains all objects shared by the user to

   other users of the system.

 * The ``others`` element, which contains all objects that other users share

   with the user.

 * The ``groups`` element, which contains the names of groups the user has

   defined. Each group consists of a user list. Group creation, deletion, and

   manipulation is carried out by actions originating here.

 * The ``history`` element, which allows browsing past instances of ``home``

   and - optionally - ``trash``.

Objects in Pithos+ can be:

 * Moved to trash and then deleted.

 * Shared with specific permissions.

 * Made public (shared with non-Pithos+ users).

 * Restored from previous versions.

Some of these functions are performed by the client software and some by the

Pithos+ server.

In the first version of Pithos, objects could also be assigned custom tags.

This is no longer supported. Existing deployments can migrate tags into a

specific metadata value, i.e. ``X-Object-Meta-Tags``.

Implementation Guidelines

~~~~~~~~~~~~~~~~~~~~~~~~~

Pithos+ clients should use the ``pithos`` and ``trash`` containers for active

and inactive objects respectively. If any of these containers is not found, the

client software should create it, without interrupting the user's workflow. The

``home`` element corresponds to ``pithos`` and the ``trash`` element to

``trash``. Use ``PUT`` with the ``X-Move-From`` header, or ``MOVE`` to transfer

objects from one container to the other. Use ``DELETE`` to remove from

``pithos`` without trashing, or to remove from ``trash``. When moving objects,

detect naming conflicts with the ``If-Match`` or ``If-None-Match`` headers.

Such conflicts should be resolved by the user.

Object names should use the ``/`` delimiter to impose a hierarchy of folders

and files.

The ``shared`` element should be implemented as a read-only view of the

``pithos`` container, using the ``shared`` parameter when listing objects. The

``others`` element, should start with a top-level ``GET`` to retrieve the list

of accounts accessible to the user. It is suggested that the client software

hides the next step of navigation - the container - if it only includes

``pithos`` and forwards the user directly to the objects.

Public objects are not included in ``shared`` and ``others`` listings. It is

suggested that they are marked in a visually distinctive way in ``pithos``

listings (for example using an icon overlay).

A special application menu, or a section in application preferences, should be

devoted to managing groups (the ``groups`` element). All group-related actions

are implemented at the account level.

Browsing past versions of objects should be available both at the object and

the container level. At the object level, a list of past versions can be

included in the screen showing details or more information on the object

(metadata, permissions, etc.). At the container level, it is suggested that

clients use a ``history`` element, which presents to the user a read-only,

time-variable view of ``pithos`` contents. This can be accomplished via the

``until`` parameter in listings. Optionally, ``history`` may include ``trash``.

Uploading and downloading data

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

By using hashmaps to upload and download objects the corresponding operations

can complete much faster.

In the case of an upload, only the missing blocks will be submitted to the

server:

 * Calculate the hash value for each block of the object to be uploaded. Use

   the hash algorithm and block size of the destination container.

 * Send a hashmap ``PUT`` request for the object.

   * Server responds with status ``201`` (Created):

     * Blocks are already on the server. The object has been created. Done.

   * Server responds with status ``409`` (Conflict):

     * Server's response body contains the hashes of the blocks that do not

       exist on the server.

     * For each hash value in the server's response (or all hashes together):

       * Send a ``POST`` request to the destination container with the

         corresponding data.

 * Repeat hashmap ``PUT``. Fail if the server's response is not ``201``.

Consulting hashmaps when downloading allows for resuming partially transferred

objects. The client should retrieve the hashmap from the server and compare it

with the hashmap computed from the respective local file. Any missing parts can

be downloaded with ``GET`` requests with the additional ``Range`` header.

Syncing

~~~~~~~

Consider the following algorithm for synchronizing a local folder with the

server. The "state" is the complete object listing, with the corresponding

attributes.

.. code-block:: python

  # L: Local State, the last synced state of the object.

  # Stored locally (e.g. in an SQLite database)

  # C: Current State, the current local state of the object

  # Returned by the filesystem

  # S: Server State, the current server state of the object

  # Returned by the server (HTTP request)

  def sync(path):

      L = get_local_state(path)   # Database action

      C = get_current_state(path) # Filesystem action

      S = get_server_state(path)  # Network action

      if C == L:

          # No local changes

          if S == L:

              # No remote changes, nothing to do

              return

          else:

              # Update local state to match that of the server

              download(path)

              update_local_state(path, S)

      else:

          # Local changes exist

          if S == L:

              # No remote changes, update the server and the local state

              upload(path)

              update_local_state(path, C)

          else:

              # Both local and server changes exist

              if C == S:

                  # We were lucky, both did the same

                  update_local_state(path, C)

              else:

                  # Conflicting changes exist

                  conflict()

Notes:

 * States represent file hashes (it is suggested to use Merkle). Deleted or

   non-existing files are assumed to have a magic hash (e.g. empty string).