File: graph.rst

package info (click to toggle)
python-cassandra-driver 3.29.2-5
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 5,144 kB
  • sloc: python: 51,532; ansic: 768; makefile: 136; sh: 13
file content (434 lines) | stat: -rw-r--r-- 16,343 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
DataStax Graph Queries
======================

The driver executes graph queries over the Cassandra native protocol. Use
:meth:`.Session.execute_graph` or :meth:`.Session.execute_graph_async` for 
executing gremlin queries in DataStax Graph.

The driver defines three Execution Profiles suitable for graph execution:

* :data:`~.cluster.EXEC_PROFILE_GRAPH_DEFAULT`
* :data:`~.cluster.EXEC_PROFILE_GRAPH_SYSTEM_DEFAULT`
* :data:`~.cluster.EXEC_PROFILE_GRAPH_ANALYTICS_DEFAULT`

See :doc:`getting_started` and :doc:`execution_profiles`
for more detail on working with profiles.

In DSE 6.8.0, the Core graph engine has been introduced and is now the default. It
provides a better unified multi-model, performance and scale. This guide
is for graphs that use the core engine. If you work with previous versions of 
DSE or existing graphs, see :doc:`classic_graph`.

Getting Started with Graph and the Core Engine
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

First, we need to create a graph in the system. To access the system API, we 
use the system execution profile ::

    from cassandra.cluster import Cluster, EXEC_PROFILE_GRAPH_SYSTEM_DEFAULT

    cluster = Cluster()
    session = cluster.connect()

    graph_name = 'movies'
    session.execute_graph("system.graph(name).create()", {'name': graph_name},
                          execution_profile=EXEC_PROFILE_GRAPH_SYSTEM_DEFAULT)


Graphs that use the core engine only support GraphSON3. Since they are Cassandra tables under
the hood, we can automatically configure the execution profile with the proper options
(row_factory and graph_protocol) when executing queries. You only need to make sure that
the `graph_name` is set and GraphSON3 will be automatically used::

    from cassandra.cluster import Cluster, GraphExecutionProfile, EXEC_PROFILE_GRAPH_DEFAULT

    graph_name = 'movies'
    ep = GraphExecutionProfile(graph_options=GraphOptions(graph_name=graph_name))
    cluster = Cluster(execution_profiles={EXEC_PROFILE_GRAPH_DEFAULT: ep})
    session = cluster.connect()
    session.execute_graph("g.addV(...)")


Note that this graph engine detection is based on the metadata. You might experience
some query errors if the graph has been newly created and is not yet in the metadata. This
would result to a badly configured execution profile. If you really want to avoid that,
configure your execution profile explicitly::

    from cassandra.cluster import Cluster, GraphExecutionProfile, EXEC_PROFILE_GRAPH_DEFAULT
    from cassandra.graph import GraphOptions, GraphProtocol, graph_graphson3_row_factory

    graph_name = 'movies'
    ep_graphson3 = GraphExecutionProfile(
        row_factory=graph_graphson3_row_factory,
        graph_options=GraphOptions(
            graph_protocol=GraphProtocol.GRAPHSON_3_0,
            graph_name=graph_name))

    cluster = Cluster(execution_profiles={'core': ep_graphson3})
    session = cluster.connect()
    session.execute_graph("g.addV(...)", execution_profile='core')


We are ready to configure our graph schema. We will create a simple one for movies::

    # A Vertex represents a "thing" in the world.
    # Create the genre vertex
    query = """
        schema.vertexLabel('genre')
            .partitionBy('genreId', Int)
            .property('name', Text)
            .create()
    """
    session.execute_graph(query)

    # Create the person vertex
    query = """
        schema.vertexLabel('person')
            .partitionBy('personId', Int)
            .property('name', Text)
            .create()
    """
    session.execute_graph(query)

    # Create the movie vertex
    query = """
        schema.vertexLabel('movie')
            .partitionBy('movieId', Int)
            .property('title', Text)
            .property('year', Int)
            .property('country', Text)
            .create()
    """
    session.execute_graph(query)

    # An edge represents a relationship between two vertices
    # Create our edges
    queries = """
    schema.edgeLabel('belongsTo').from('movie').to('genre').create();
    schema.edgeLabel('actor').from('movie').to('person').create();
    """
    session.execute_graph(queries)

    # Indexes to execute graph requests efficiently

    # If you have a node with the search workload enabled (solr), use the following:
    indexes = """
        schema.vertexLabel('genre').searchIndex()
            .by("name")
            .create();

        schema.vertexLabel('person').searchIndex()
            .by("name")
            .create();

        schema.vertexLabel('movie').searchIndex()
            .by('title')
            .by("year")
            .create();
    """
    session.execute_graph(indexes)

    # Otherwise, use secondary indexes:
    indexes = """
        schema.vertexLabel('genre')
            .secondaryIndex('by_genre')
            .by('name')
            .create()

        schema.vertexLabel('person')
            .secondaryIndex('by_name')
            .by('name')
            .create()

        schema.vertexLabel('movie')
            .secondaryIndex('by_title')
            .by('title')
            .create()
    """
    session.execute_graph(indexes)

Add some edge indexes (materialized views)::

    indexes = """
        schema.edgeLabel('belongsTo')
            .from('movie')
            .to('genre')
            .materializedView('movie__belongsTo__genre_by_in_genreId')
            .ifNotExists()
            .partitionBy(IN, 'genreId')
            .clusterBy(OUT, 'movieId', Asc)
            .create()

        schema.edgeLabel('actor')
            .from('movie')
            .to('person')
            .materializedView('movie__actor__person_by_in_personId')
            .ifNotExists()
            .partitionBy(IN, 'personId')
            .clusterBy(OUT, 'movieId', Asc)
            .create()
    """
    session.execute_graph(indexes)

Next, we'll add some data::

    session.execute_graph("""
        g.addV('genre').property('genreId', 1).property('name', 'Action').next();
        g.addV('genre').property('genreId', 2).property('name', 'Drama').next();
        g.addV('genre').property('genreId', 3).property('name', 'Comedy').next();
        g.addV('genre').property('genreId', 4).property('name', 'Horror').next();
    """)

    session.execute_graph("""
        g.addV('person').property('personId', 1).property('name', 'Mark Wahlberg').next();
        g.addV('person').property('personId', 2).property('name', 'Leonardo DiCaprio').next();
        g.addV('person').property('personId', 3).property('name', 'Iggy Pop').next();
    """)

    session.execute_graph("""
        g.addV('movie').property('movieId', 1).property('title', 'The Happening').
            property('year', 2008).property('country', 'United States').next();
        g.addV('movie').property('movieId', 2).property('title', 'The Italian Job').
            property('year', 2003).property('country', 'United States').next();

        g.addV('movie').property('movieId', 3).property('title', 'Revolutionary Road').
            property('year', 2008).property('country', 'United States').next();
        g.addV('movie').property('movieId', 4).property('title', 'The Man in the Iron Mask').
            property('year', 1998).property('country', 'United States').next();

        g.addV('movie').property('movieId', 5).property('title', 'Dead Man').
            property('year', 1995).property('country', 'United States').next();
    """)

Now that our genre, actor and movie vertices are added, we'll create the relationships (edges) between them::

    session.execute_graph("""
        genre_horror = g.V().hasLabel('genre').has('name', 'Horror').id().next();
        genre_drama = g.V().hasLabel('genre').has('name', 'Drama').id().next();
        genre_action = g.V().hasLabel('genre').has('name', 'Action').id().next();

        leo  = g.V().hasLabel('person').has('name', 'Leonardo DiCaprio').id().next();
        mark = g.V().hasLabel('person').has('name', 'Mark Wahlberg').id().next();
        iggy = g.V().hasLabel('person').has('name', 'Iggy Pop').id().next();

        the_happening = g.V().hasLabel('movie').has('title', 'The Happening').id().next();
        the_italian_job = g.V().hasLabel('movie').has('title', 'The Italian Job').id().next();
        rev_road = g.V().hasLabel('movie').has('title', 'Revolutionary Road').id().next();
        man_mask = g.V().hasLabel('movie').has('title', 'The Man in the Iron Mask').id().next();
        dead_man = g.V().hasLabel('movie').has('title', 'Dead Man').id().next();

        g.addE('belongsTo').from(__.V(the_happening)).to(__.V(genre_horror)).next();
        g.addE('belongsTo').from(__.V(the_italian_job)).to(__.V(genre_action)).next();
        g.addE('belongsTo').from(__.V(rev_road)).to(__.V(genre_drama)).next();
        g.addE('belongsTo').from(__.V(man_mask)).to(__.V(genre_drama)).next();
        g.addE('belongsTo').from(__.V(man_mask)).to(__.V(genre_action)).next();
        g.addE('belongsTo').from(__.V(dead_man)).to(__.V(genre_drama)).next();

        g.addE('actor').from(__.V(the_happening)).to(__.V(mark)).next();
        g.addE('actor').from(__.V(the_italian_job)).to(__.V(mark)).next();
        g.addE('actor').from(__.V(rev_road)).to(__.V(leo)).next();
        g.addE('actor').from(__.V(man_mask)).to(__.V(leo)).next();
        g.addE('actor').from(__.V(dead_man)).to(__.V(iggy)).next();
    """)

We are all set. You can now query your graph. Here are some examples::

    # Find all movies of the genre Drama
    for r in session.execute_graph("""
      g.V().has('genre', 'name', 'Drama').in('belongsTo').valueMap();"""):
        print(r)

    # Find all movies of the same genre than the movie 'Dead Man'
    for r in session.execute_graph("""
      g.V().has('movie', 'title', 'Dead Man').out('belongsTo').in('belongsTo').valueMap();"""):
        print(r)

    # Find all movies of Mark Wahlberg
    for r in session.execute_graph("""
      g.V().has('person', 'name', 'Mark Wahlberg').in('actor').valueMap();"""):
        print(r)

To see a more graph examples, see `DataStax Graph Examples <https://github.com/datastax/graph-examples/>`_.

Graph Types for the Core Engine
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Here are the supported graph types with their python representations:

============   =================
DSE Graph      Python Driver
============   =================
text           str
boolean        bool
bigint         long
int            int
smallint       int
varint         long
double         float
float          float
uuid           UUID
bigdecimal     Decimal
duration       Duration (cassandra.util)
inet           str or IPV4Address/IPV6Address (if available)
timestamp      datetime.datetime
date           datetime.date
time           datetime.time
polygon        Polygon
point          Point
linestring     LineString
blob           bytearray, buffer (PY2), memoryview (PY3), bytes (PY3)
list           list
map            dict
set            set or list
               (Can return a list due to numerical values returned by Java)
tuple          tuple
udt            class or namedtuple
============   =================

Named Parameters
~~~~~~~~~~~~~~~~

Named parameters are passed in a dict to :meth:`.cluster.Session.execute_graph`::

    result_set = session.execute_graph('[a, b]', {'a': 1, 'b': 2}, execution_profile=EXEC_PROFILE_GRAPH_SYSTEM_DEFAULT)
    [r.value for r in result_set]  # [1, 2]

All python types listed in `Graph Types for the Core Engine`_ can be passed as named parameters and will be serialized
automatically to their graph representation:

Example::

    session.execute_graph("""
      g.addV('person').
      property('name', text_value).
      property('age', integer_value).
      property('birthday', timestamp_value).
      property('house_yard', polygon_value).next()
    """, {
      'text_value': 'Mike Smith',
      'integer_value': 34,
      'timestamp_value': datetime.datetime(1967, 12, 30),
      'polygon_value': Polygon(((30, 10), (40, 40), (20, 40), (10, 20), (30, 10)))
    })


As with all Execution Profile parameters, graph options can be set in the cluster default (as shown in the first example)
or specified per execution::

    ep = session.execution_profile_clone_update(EXEC_PROFILE_GRAPH_DEFAULT,
                                                graph_options=GraphOptions(graph_name='something-else'))
    session.execute_graph(statement, execution_profile=ep)

CQL collections, Tuple and UDT
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This is a very interesting feature of the core engine: we can use all CQL data types, including
list, map, set, tuple and udt. Here is an example using all these types::

    query = """
        schema.type('address')
            .property('address', Text)
            .property('city', Text)
            .property('state', Text)
            .create();
    """
    session.execute_graph(query)

    # It works the same way than normal CQL UDT, so we
    # can create an udt class and register it
    class Address(object):
        def __init__(self, address, city, state):
            self.address = address
            self.city = city
            self.state = state

    session.cluster.register_user_type(graph_name, 'address', Address)

    query = """
        schema.vertexLabel('person')
            .partitionBy('personId', Int)
            .property('address', typeOf('address'))
            .property('friends', listOf(Text))
            .property('skills', setOf(Text))
            .property('scores', mapOf(Text, Int))
            .property('last_workout', tupleOf(Text, Date))
            .create()
    """
    session.execute_graph(query)

    # insertion example
    query = """
         g.addV('person')
            .property('personId', pid)
            .property('address', address)
            .property('friends', friends)
            .property('skills', skills)
            .property('scores', scores)
            .property('last_workout', last_workout)
            .next()
    """

    session.execute_graph(query, {
        'pid': 3,
        'address': Address('42 Smith St', 'Quebec', 'QC'),
        'friends': ['Al', 'Mike', 'Cathy'],
        'skills': {'food', 'fight', 'chess'},
        'scores': {'math': 98, 'french': 3},
        'last_workout': ('CrossFit', datetime.date(2018, 11, 20))
    })

Limitations
-----------

Since Python is not a strongly-typed language and the UDT/Tuple graphson representation is, you might 
get schema errors when trying to write numerical data. Example::

    session.execute_graph("""
        schema.vertexLabel('test_tuple').partitionBy('id', Int).property('t', tupleOf(Text, Bigint)).create()
    """)

    session.execute_graph("""
        g.addV('test_tuple').property('id', 0).property('t', t)
          """, 
          {'t': ('Test', 99))}
    )

    # error: [Invalid query] message="Value component 1 is of type int, not bigint"

This is because the server requires the client to include a GraphSON schema definition
with every UDT or tuple query. In the general case, the driver can't determine what Graph type
is meant by, e.g., an int value, and so it can't serialize the value with the correct type in the schema.
The driver provides some numerical type-wrapper factories that you can use to specify types:

* :func:`~.to_int`
* :func:`~.to_bigint`
* :func:`~.to_smallint`
* :func:`~.to_float`
* :func:`~.to_double`

Here's the working example of the case above::

    from cassandra.graph import to_bigint

     session.execute_graph("""
        g.addV('test_tuple').property('id', 0).property('t', t)
          """, 
          {'t': ('Test', to_bigint(99))}
    )

Continuous Paging
~~~~~~~~~~~~~~~~~

This is another nice feature that comes with the core engine: continuous paging with
graph queries. If all nodes of the cluster are >= DSE 6.8.0, it is automatically
enabled under the hood to get the best performance. If you want to explicitly
enable/disable it, you can do it through the execution profile::

    # Disable it
    ep = GraphExecutionProfile(..., continuous_paging_options=None))
    cluster = Cluster(execution_profiles={EXEC_PROFILE_GRAPH_DEFAULT: ep})

    # Enable with a custom max_pages option
    ep = GraphExecutionProfile(...,
        continuous_paging_options=ContinuousPagingOptions(max_pages=10)))
    cluster = Cluster(execution_profiles={EXEC_PROFILE_GRAPH_DEFAULT: ep})