root / docs / dev-guide.rst @ 991d0507
History | View | Annotate | Download (9.7 kB)
1 |
.. _dev-guide: |
---|---|
2 |
|
3 |
Synnefo Developer's Guide |
4 |
^^^^^^^^^^^^^^^^^^^^^^^^^ |
5 |
|
6 |
This is the complete Synnefo Developer's Guide. Here, we document all Synnefo |
7 |
REST APIs, to allow external developers write independent tools that interact |
8 |
with Synnefo. |
9 |
|
10 |
Synnefo exposes the OpenStack APIs for all its operations. Also, extensions |
11 |
have been written for advanced operations wherever needed, and minor changes |
12 |
for things that were missing or change frequently. |
13 |
|
14 |
All Synnefo services have the analogous OpenStack API: |
15 |
|
16 |
| Cyclades/Compute Service -> OpenStack Compute API |
17 |
| Cyclades/Network Service -> OpenStack Compute/Network API (not Quantum yet) |
18 |
| Cyclades/Image Service -> OpenStack Compute/Image API |
19 |
| Cyclades/Plankton/Image Service -> OpenStack Glance API |
20 |
| Pithos/Storage Service -> OpenStack Object Store API |
21 |
| Astakos/Identity Service -> Proprietary, moving to OpenStack Keystone API |
22 |
|
23 |
Below, we will describe all Synnefo APIs with conjuction to the OpenStack APIs. |
24 |
|
25 |
|
26 |
Identity Service API (Astakos) |
27 |
============================== |
28 |
|
29 |
Currently, Astakos which is the Identity Management Service of Synnefo, has a |
30 |
proprietary API, but we are moving to the OpenStack Keystone API. |
31 |
|
32 |
The current Identity Management API is: |
33 |
|
34 |
.. toctree:: |
35 |
:maxdepth: 2 |
36 |
|
37 |
Identity API <astakos-api-guide> |
38 |
|
39 |
|
40 |
Compute Service API (Cyclades) |
41 |
============================== |
42 |
|
43 |
The Compute part of Cyclades exposes the OpenStack Compute API with minor |
44 |
changes wherever needed. |
45 |
|
46 |
This is the Cyclades/Compute API: |
47 |
|
48 |
.. toctree:: |
49 |
:maxdepth: 2 |
50 |
|
51 |
Compute API <cyclades-api-guide> |
52 |
|
53 |
|
54 |
Network Service API (Cyclades) |
55 |
============================== |
56 |
|
57 |
The Network Service is implemented inside Cyclades. It exposes the part of the |
58 |
OpenStack Compute API that has to do with Networks. The OpenStack Quantum API |
59 |
is not implemented yet. |
60 |
|
61 |
Please consult the :ref:`Cyclades/Network API <cyclades-api-guide>` for more |
62 |
details. |
63 |
|
64 |
|
65 |
Images Service API (Cyclades/Plankton) |
66 |
====================================== |
67 |
|
68 |
Plankton is the Image Service of Synnefo, currently implemented inside |
69 |
Cyclades. Plankton exposes the OpenStack Glance API with minor changes wherever |
70 |
needed. |
71 |
|
72 |
This is the Cyclades/Plankton Image API: |
73 |
|
74 |
.. toctree:: |
75 |
:maxdepth: 2 |
76 |
|
77 |
Image API <plankton-api-guide> |
78 |
|
79 |
|
80 |
Storage Service API (Pithos) |
81 |
============================ |
82 |
|
83 |
Pithos is the Storage Service of Synnefo and it exposes the OpenStack Object |
84 |
Storage API with extensions for advanced operations, e.g., syncing. |
85 |
|
86 |
This is the Pithos Object Storage API: |
87 |
|
88 |
.. toctree:: |
89 |
:maxdepth: 2 |
90 |
|
91 |
Object Storage API <pithos-api-guide> |
92 |
|
93 |
|
94 |
Implementing new clients |
95 |
======================== |
96 |
|
97 |
In this section we discuss implementation guidelines, that a developer should |
98 |
take into account before writing his own client for the above APIs. Before, |
99 |
starting your client implementation, make sure you have thoroughly read the |
100 |
corresponding Synnefo API. |
101 |
|
102 |
Pithos clients |
103 |
-------------- |
104 |
|
105 |
User Experience |
106 |
~~~~~~~~~~~~~~~ |
107 |
|
108 |
Hopefully this API will allow for a multitude of client implementations, each |
109 |
supporting a different device or operating system. All clients will be able to |
110 |
manipulate containers and objects - even software only designed for OOS API |
111 |
compatibility. But a Pithos interface should not be only about showing |
112 |
containers and folders. There are some extra user interface elements and |
113 |
functionalities that should be common to all implementations. |
114 |
|
115 |
Upon entrance to the service, a user is presented with the following elements - |
116 |
which can be represented as folders or with other related icons: |
117 |
|
118 |
* The ``home`` element, which is used as the default entry point to the user's |
119 |
"files". Objects under ``home`` are represented in the usual hierarchical |
120 |
organization of folders and files. |
121 |
* The ``trash`` element, which contains files that have been marked for |
122 |
deletion, but can still be recovered. |
123 |
* The ``shared`` element, which contains all objects shared by the user to |
124 |
other users of the system. |
125 |
* The ``others`` element, which contains all objects that other users share |
126 |
with the user. |
127 |
* The ``groups`` element, which contains the names of groups the user has |
128 |
defined. Each group consists of a user list. Group creation, deletion, and |
129 |
manipulation is carried out by actions originating here. |
130 |
* The ``history`` element, which allows browsing past instances of ``home`` |
131 |
and - optionally - ``trash``. |
132 |
|
133 |
Objects in Pithos can be: |
134 |
|
135 |
* Moved to trash and then deleted. |
136 |
* Shared with specific permissions. |
137 |
* Made public (shared with non-Pithos users). |
138 |
* Restored from previous versions. |
139 |
|
140 |
Some of these functions are performed by the client software and some by the |
141 |
Pithos server. |
142 |
|
143 |
In the first version of Pithos, objects could also be assigned custom tags. |
144 |
This is no longer supported. Existing deployments can migrate tags into a |
145 |
specific metadata value, i.e. ``X-Object-Meta-Tags``. |
146 |
|
147 |
Implementation Guidelines |
148 |
~~~~~~~~~~~~~~~~~~~~~~~~~ |
149 |
|
150 |
Pithos clients should use the ``pithos`` and ``trash`` containers for active |
151 |
and inactive objects respectively. If any of these containers is not found, the |
152 |
client software should create it, without interrupting the user's workflow. The |
153 |
``home`` element corresponds to ``pithos`` and the ``trash`` element to |
154 |
``trash``. Use ``PUT`` with the ``X-Move-From`` header, or ``MOVE`` to transfer |
155 |
objects from one container to the other. Use ``DELETE`` to remove from |
156 |
``pithos`` without trashing, or to remove from ``trash``. When moving objects, |
157 |
detect naming conflicts with the ``If-Match`` or ``If-None-Match`` headers. |
158 |
Such conflicts should be resolved by the user. |
159 |
|
160 |
Object names should use the ``/`` delimiter to impose a hierarchy of folders |
161 |
and files. |
162 |
|
163 |
The ``shared`` element should be implemented as a read-only view of the |
164 |
``pithos`` container, using the ``shared`` parameter when listing objects. The |
165 |
``others`` element, should start with a top-level ``GET`` to retrieve the list |
166 |
of accounts accessible to the user. It is suggested that the client software |
167 |
hides the next step of navigation - the container - if it only includes |
168 |
``pithos`` and forwards the user directly to the objects. |
169 |
|
170 |
Public objects are not included in ``shared`` and ``others`` listings. It is |
171 |
suggested that they are marked in a visually distinctive way in ``pithos`` |
172 |
listings (for example using an icon overlay). |
173 |
|
174 |
A special application menu, or a section in application preferences, should be |
175 |
devoted to managing groups (the ``groups`` element). All group-related actions |
176 |
are implemented at the account level. |
177 |
|
178 |
Browsing past versions of objects should be available both at the object and |
179 |
the container level. At the object level, a list of past versions can be |
180 |
included in the screen showing details or more information on the object |
181 |
(metadata, permissions, etc.). At the container level, it is suggested that |
182 |
clients use a ``history`` element, which presents to the user a read-only, |
183 |
time-variable view of ``pithos`` contents. This can be accomplished via the |
184 |
``until`` parameter in listings. Optionally, ``history`` may include ``trash``. |
185 |
|
186 |
Uploading and downloading data |
187 |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
188 |
|
189 |
By using hashmaps to upload and download objects the corresponding operations |
190 |
can complete much faster. |
191 |
|
192 |
In the case of an upload, only the missing blocks will be submitted to the |
193 |
server: |
194 |
|
195 |
* Calculate the hash value for each block of the object to be uploaded. Use |
196 |
the hash algorithm and block size of the destination container. |
197 |
* Send a hashmap ``PUT`` request for the object. |
198 |
|
199 |
* Server responds with status ``201`` (Created): |
200 |
|
201 |
* Blocks are already on the server. The object has been created. Done. |
202 |
|
203 |
* Server responds with status ``409`` (Conflict): |
204 |
|
205 |
* Server's response body contains the hashes of the blocks that do not |
206 |
exist on the server. |
207 |
* For each hash value in the server's response (or all hashes together): |
208 |
|
209 |
* Send a ``POST`` request to the destination container with the |
210 |
corresponding data. |
211 |
|
212 |
* Repeat hashmap ``PUT``. Fail if the server's response is not ``201``. |
213 |
|
214 |
Consulting hashmaps when downloading allows for resuming partially transferred |
215 |
objects. The client should retrieve the hashmap from the server and compare it |
216 |
with the hashmap computed from the respective local file. Any missing parts can |
217 |
be downloaded with ``GET`` requests with the additional ``Range`` header. |
218 |
|
219 |
Syncing |
220 |
~~~~~~~ |
221 |
|
222 |
Consider the following algorithm for synchronizing a local folder with the |
223 |
server. The "state" is the complete object listing, with the corresponding |
224 |
attributes. |
225 |
|
226 |
.. code-block:: python |
227 |
|
228 |
# L: Local State, the last synced state of the object. |
229 |
# Stored locally (e.g. in an SQLite database) |
230 |
|
231 |
# C: Current State, the current local state of the object |
232 |
# Returned by the filesystem |
233 |
|
234 |
# S: Server State, the current server state of the object |
235 |
# Returned by the server (HTTP request) |
236 |
|
237 |
def sync(path): |
238 |
L = get_local_state(path) # Database action |
239 |
C = get_current_state(path) # Filesystem action |
240 |
S = get_server_state(path) # Network action |
241 |
|
242 |
if C == L: |
243 |
# No local changes |
244 |
if S == L: |
245 |
# No remote changes, nothing to do |
246 |
return |
247 |
else: |
248 |
# Update local state to match that of the server |
249 |
download(path) |
250 |
update_local_state(path, S) |
251 |
else: |
252 |
# Local changes exist |
253 |
if S == L: |
254 |
# No remote changes, update the server and the local state |
255 |
upload(path) |
256 |
update_local_state(path, C) |
257 |
else: |
258 |
# Both local and server changes exist |
259 |
if C == S: |
260 |
# We were lucky, both did the same |
261 |
update_local_state(path, C) |
262 |
else: |
263 |
# Conflicting changes exist |
264 |
conflict() |
265 |
|
266 |
|
267 |
Notes: |
268 |
|
269 |
* States represent file hashes (it is suggested to use Merkle). Deleted or |
270 |
non-existing files are assumed to have a magic hash (e.g. empty string). |