1 """GnuMed database object business class.
2
3 Overview
4 --------
5 This class wraps a source relation (table, view) which
6 represents an entity that makes immediate business sense
7 such as a vaccination or a medical document. In many if
8 not most cases this source relation is a denormalizing
9 view. The data in that view will in most cases, however,
10 originate from several normalized tables. One instance
11 of this class represents one row of said source relation.
12
13 Note, however, that this class does not *always* simply
14 wrap a single table or view. It can also encompass several
15 relations (views, tables, sequences etc) that taken together
16 form an object meaningful to *business* logic.
17
18 Initialization
19 --------------
20 There are two ways to initialize an instance with values.
21 One way is to pass a "primary key equivalent" object into
22 __init__(). Refetch_payload() will then pull the data from
23 the backend. Another way would be to fetch the data outside
24 the instance and pass it in via the <row> argument. In that
25 case the instance will not initially connect to the databse
26 which may offer a great boost to performance.
27
28 Values API
29 ----------
30 Field values are cached for later access. They can be accessed
31 by a dictionary API, eg:
32
33 old_value = object['field']
34 object['field'] = new_value
35
36 The field names correspond to the respective column names
37 in the "main" source relation. Accessing non-existant field
38 names will raise an error, so does trying to set fields not
39 listed in self.__class__._updatable_fields. To actually
40 store updated values in the database one must explicitly
41 call save_payload().
42
43 The class will in many cases be enhanced by accessors to
44 related data that is not directly part of the business
45 object itself but are closely related, such as codes
46 linked to a clinical narrative entry (eg a diagnosis). Such
47 accessors in most cases start with get_*. Related setters
48 start with set_*. The values can be accessed via the
49 object['field'] syntax, too, but they will be cached
50 independantly.
51
52 Concurrency handling
53 --------------------
54 GnuMed connections always run transactions in isolation level
55 "serializable". This prevents transactions happening at the
56 *very same time* to overwrite each other's data. All but one
57 of them will abort with a concurrency error (eg if a
58 transaction runs a select-for-update later than another one
59 it will hang until the first transaction ends. Then it will
60 succeed or fail depending on what the first transaction
61 did). This is standard transactional behaviour.
62
63 However, another transaction may have updated our row
64 between the time we first fetched the data and the time we
65 start the update transaction. This is noticed by getting the
66 XMIN system column for the row when initially fetching the
67 data and using that value as a where condition value when
68 updating the row later. If the row had been updated (xmin
69 changed) or deleted (primary key disappeared) in the
70 meantime the update will touch zero rows (as no row with
71 both PK and XMIN matching is found) even if the query itself
72 syntactically succeeds.
73
74 When detecting a change in a row due to XMIN being different
75 one needs to be careful how to represent that to the user.
76 The row may simply have changed but it also might have been
77 deleted and a completely new and unrelated row which happens
78 to have the same primary key might have been created ! This
79 row might relate to a totally different context (eg. patient,
80 episode, encounter).
81
82 One can offer all the data to the user:
83
84 self.original_payload
85 - contains the data at the last successful refetch
86
87 self.modified_payload
88 - contains the modified payload just before the last
89 failure of save_payload() - IOW what is currently
90 in the database
91
92 self._payload
93 - contains the currently active payload which may or
94 may not contain changes
95
96 For discussion on this see the thread starting at:
97
98 http://archives.postgresql.org/pgsql-general/2004-10/msg01352.php
99
100 and here
101
102 http://groups.google.com/group/pgsql.general/browse_thread/thread/e3566ba76173d0bf/6cf3c243a86d9233
103 (google for "XMIN semantic at peril")
104
105 Problem cases with XMIN:
106
107 1) not unlikely
108 - a very old row is read with XMIN
109 - vacuum comes along and sets XMIN to FrozenTransactionId
110 - now XMIN changed but the row actually didn't !
111 - an update with "... where xmin = old_xmin ..." fails
112 although there is no need to fail
113
114 2) quite unlikely
115 - a row is read with XMIN
116 - a long time passes
117 - the original XMIN gets frozen to FrozenTransactionId
118 - another writer comes along and changes the row
119 - incidentally the exact same old row gets the old XMIN *again*
120 - now XMIN is (again) the same but the data changed !
121 - a later update fails to detect the concurrent change !!
122
123 TODO:
124 The solution is to use our own column for optimistic locking
125 which gets updated by an AFTER UPDATE trigger.
126 """
127
128 __version__ = "$Revision: 1.60 $"
129 __author__ = "K.Hilbert <Karsten.Hilbert@gmx.net>"
130 __license__ = "GPL"
131
132 import sys, copy, types, inspect, logging
133
134
135 if __name__ == '__main__':
136 sys.path.insert(0, '../../')
137 from Gnumed.pycommon import gmExceptions, gmPG2
138
139
140 _log = logging.getLogger('gm.db')
141 _log.info(__version__)
142
144 """Represents business objects in the database.
145
146 Rules:
147 - instances ARE ASSUMED TO EXIST in the database
148 - PK construction (aPK_obj): DOES verify its existence on instantiation
149 (fetching data fails)
150 - Row construction (row): allowed by using a dict of pairs
151 field name: field value (PERFORMANCE improvement)
152 - does NOT verify FK target existence
153 - does NOT create new entries in the database
154 - does NOT lazy-fetch fields on access
155
156 Class scope SQL commands and variables:
157
158 <_cmd_fetch_payload>
159 - must return exactly one row
160 - where clause argument values are expected
161 in self.pk_obj (taken from __init__(aPK_obj))
162 - must return xmin of all rows that _cmds_store_payload
163 will be updating, so views must support the xmin columns
164 of their underlying tables
165
166 <_cmds_store_payload>
167 - one or multiple "update ... set ... where xmin_* = ..." statements
168 which actually update the database from the data in self._payload,
169 - the last query must refetch the XMIN values needed to detect
170 concurrent updates, their field names had better be the same as
171 in _cmd_fetch_payload
172
173 <_updatable_fields>
174 - a list of fields available for update via object['field']
175
176 """
177
178 - def __init__(self, aPK_obj=None, row=None):
179 """Init business object.
180
181 Call from child classes:
182
183 super(cChildClass, self).__init__(aPK_obj = aPK_obj, row = row)
184 """
185
186
187
188 self.pk_obj = '<uninitialized>'
189 self._idx = {}
190 self._payload = []
191 self._ext_cache = {}
192 self._is_modified = False
193
194
195 self.__class__._cmd_fetch_payload
196 self.__class__._cmds_store_payload
197 self.__class__._updatable_fields
198
199 if aPK_obj is not None:
200 self.__init_from_pk(aPK_obj=aPK_obj)
201 else:
202 self._init_from_row_data(row=row)
203
204 self._is_modified = False
205
207 """Creates a new clinical item instance by its PK.
208
209 aPK_obj can be:
210 - a simple value
211 * the primary key WHERE condition must be
212 a simple column
213 - a dictionary of values
214 * the primary key where condition must be a
215 subselect consuming the dict and producing
216 the single-value primary key
217 """
218 self.pk_obj = aPK_obj
219 result = self.refetch_payload()
220 if result is True:
221 self.original_payload = {}
222 for field in self._idx.keys():
223 self.original_payload[field] = self._payload[self._idx[field]]
224 return True
225
226 if result is False:
227 raise gmExceptions.ConstructorError, "[%s:%s]: error loading instance" % (self.__class__.__name__, self.pk_obj)
228
230 """Creates a new clinical item instance given its fields.
231
232 row must be a dict with the fields:
233 - pk_field: the name of the primary key field
234 - idx: a dict mapping field names to position
235 - data: the field values in a list (as returned by
236 cursor.fetchone() in the DB-API)
237
238 row = {'data': row, 'idx': idx, 'pk_field': 'the PK column name'}
239
240 rows, idx = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}], get_col_idx = True)
241 objects = [ cChildClass(row = {'data': r, 'idx': idx, 'pk_field': 'the PK column name'}) for r in rows ]
242 """
243 try:
244 self._idx = row['idx']
245 self._payload = row['data']
246 self.pk_obj = self._payload[self._idx[row['pk_field']]]
247 except:
248 _log.exception('faulty <row> argument structure: %s' % row)
249 raise gmExceptions.ConstructorError, "[%s:??]: error loading instance from row data" % self.__class__.__name__
250
251 if len(self._idx.keys()) != len(self._payload):
252 _log.critical('field index vs. payload length mismatch: %s field names vs. %s fields' % (len(self._idx.keys()), len(self._payload)))
253 _log.critical('faulty <row> argument structure: %s' % row)
254 raise gmExceptions.ConstructorError, "[%s:??]: error loading instance from row data" % self.__class__.__name__
255
256 self.original_payload = {}
257 for field in self._idx.keys():
258 self.original_payload[field] = self._payload[self._idx[field]]
259
261 if self.__dict__.has_key('_is_modified'):
262 if self._is_modified:
263 _log.critical('[%s:%s]: loosing payload changes' % (self.__class__.__name__, self.pk_obj))
264 _log.debug('original: %s' % self.original_payload)
265 _log.debug('modified: %s' % self._payload)
266
268 tmp = []
269 try:
270 for attr in self._idx.keys():
271 if self._payload[self._idx[attr]] is None:
272 tmp.append(u'%s: NULL' % attr)
273 else:
274 tmp.append('%s: >>%s<<' % (attr, self._payload[self._idx[attr]]))
275 return '[%s:%s]: %s' % (self.__class__.__name__, self.pk_obj, str(tmp))
276 except:
277 return 'nascent [%s @ %s], cannot show payload and primary key' %(self.__class__.__name__, id(self))
278
280
281
282
283 try:
284 return self._payload[self._idx[attribute]]
285 except KeyError:
286 pass
287
288
289 getter = getattr(self, 'get_%s' % attribute, None)
290 if not callable(getter):
291 _log.warning('[%s]: no attribute [%s]' % (self.__class__.__name__, attribute))
292 _log.warning('[%s]: valid attributes: %s' % (self.__class__.__name__, str(self._idx.keys())))
293 _log.warning('[%s]: no getter method [get_%s]' % (self.__class__.__name__, attribute))
294 methods = filter(lambda x: x[0].startswith('get_'), inspect.getmembers(self, inspect.ismethod))
295 _log.warning('[%s]: valid getter methods: %s' % (self.__class__.__name__, str(methods)))
296 raise gmExceptions.NoSuchBusinessObjectAttributeError, '[%s]: cannot access [%s]' % (self.__class__.__name__, attribute)
297
298 self._ext_cache[attribute] = getter()
299 return self._ext_cache[attribute]
300
302
303
304 if attribute in self.__class__._updatable_fields:
305 try:
306 if self._payload[self._idx[attribute]] != value:
307 self._payload[self._idx[attribute]] = value
308 self._is_modified = True
309 return
310 except KeyError:
311 _log.warning('[%s]: cannot set attribute <%s> despite marked settable' % (self.__class__.__name__, attribute))
312 _log.warning('[%s]: supposedly settable attributes: %s' % (self.__class__.__name__, str(self.__class__._updatable_fields)))
313 raise gmExceptions.NoSuchBusinessObjectAttributeError, '[%s]: cannot access [%s]' % (self.__class__.__name__, attribute)
314
315
316 if hasattr(self, 'set_%s' % attribute):
317 setter = getattr(self, "set_%s" % attribute)
318 if not callable(setter):
319 raise gmExceptions.NoSuchBusinessObjectAttributeError, '[%s] setter [set_%s] not callable' % (self.__class__.__name__, attribute)
320 try:
321 del self._ext_cache[attribute]
322 except KeyError:
323 pass
324 if type(value) is types.TupleType:
325 if setter(*value):
326 self._is_modified = True
327 return
328 raise gmExceptions.BusinessObjectAttributeNotSettableError, '[%s]: setter [%s] failed for [%s]' % (self.__class__.__name__, setter, value)
329 if setter(value):
330 self._is_modified = True
331 return
332
333
334 _log.error('[%s]: cannot find attribute <%s> or setter method [set_%s]' % (self.__class__.__name__, attribute, attribute))
335 _log.warning('[%s]: settable attributes: %s' % (self.__class__.__name__, str(self.__class__._updatable_fields)))
336 methods = filter(lambda x: x[0].startswith('set_'), inspect.getmembers(self, inspect.ismethod))
337 _log.warning('[%s]: valid setter methods: %s' % (self.__class__.__name__, str(methods)))
338 raise gmExceptions.BusinessObjectAttributeNotSettableError, '[%s]: cannot set [%s]' % (self.__class__.__name__, attribute)
339
340
341
343 raise NotImplementedError('comparison between [%s] and [%s] not implemented' % (self, another_object))
344
346 return self._is_modified
347
349 return self._idx.keys()
350
353
355 _log.error('[%s:%s]: forgot to override get_patient()' % (self.__class__.__name__, self.pk_obj))
356 return None
357
359 """Fetch field values from backend.
360 """
361 if self._is_modified:
362 if ignore_changes:
363 _log.critical('[%s:%s]: loosing payload changes' % (self.__class__.__name__, self.pk_obj))
364 _log.debug('original: %s' % self.original_payload)
365 _log.debug('modified: %s' % self._payload)
366 else:
367 _log.critical('[%s:%s]: cannot reload, payload changed' % (self.__class__.__name__, self.pk_obj))
368 return False
369
370 if type(self.pk_obj) == types.DictType:
371 arg = self.pk_obj
372 else:
373 arg = [self.pk_obj]
374 rows, self._idx = gmPG2.run_ro_queries (
375 queries = [{'cmd': self.__class__._cmd_fetch_payload, 'args': arg}],
376 get_col_idx = True
377 )
378 if len(rows) == 0:
379 _log.error('[%s:%s]: no such instance' % (self.__class__.__name__, self.pk_obj))
380 return False
381 self._payload = rows[0]
382 return True
383
386
387 - def save(self, conn=None):
389
391 """Store updated values (if any) in database.
392
393 Optionally accepts a pre-existing connection
394 - returns a tuple (<True|False>, <data>)
395 - True: success
396 - False: an error occurred
397 * data is (error, message)
398 * for error meanings see gmPG2.run_rw_queries()
399 """
400 if not self._is_modified:
401 return (True, None)
402
403 args = {}
404 for field in self._idx.keys():
405 args[field] = self._payload[self._idx[field]]
406 self.modified_payload = args
407
408 close_conn = self.__noop
409 if conn is None:
410 conn = gmPG2.get_connection(readonly=False)
411 close_conn = conn.close
412
413
414
415
416
417
418 queries = []
419 for query in self.__class__._cmds_store_payload:
420 queries.append({'cmd': query, 'args': args})
421 rows, idx = gmPG2.run_rw_queries (
422 link_obj = conn,
423 queries = queries,
424 return_data = True,
425 get_col_idx = True
426 )
427
428
429 row = rows[0]
430 for key in idx:
431 try:
432 self._payload[self._idx[key]] = row[idx[key]]
433 except KeyError:
434 conn.rollback()
435 close_conn()
436 _log.error('[%s:%s]: cannot update instance, XMIN refetch key mismatch on [%s]' % (self.__class__.__name__, self.pk_obj, key))
437 _log.error('payload keys: %s' % str(self._idx))
438 _log.error('XMIN refetch keys: %s' % str(idx))
439 _log.error(args)
440 raise
441
442 conn.commit()
443 close_conn()
444
445 self._is_modified = False
446
447 self.original_payload = {}
448 for field in self._idx.keys():
449 self.original_payload[field] = self._payload[self._idx[field]]
450
451 return (True, None)
452
453 if __name__ == '__main__':
454
455 if len(sys.argv) < 2:
456 sys.exit()
457
458 if sys.argv[1] != u'test':
459 sys.exit()
460
461
472
473 from Gnumed.pycommon import gmI18N
474 gmI18N.activate_locale()
475 gmI18N.install_domain()
476
477 data = {
478 'pk_field': 'bogus_pk',
479 'idx': {'bogus_pk': 0, 'bogus_field': 1},
480 'data': [-1, 'bogus_data']
481 }
482 obj = cTestObj(row=data)
483
484 obj['wrong_field'] = 1
485
486
487