diff --git a/freezer_api/db/sqlalchemy/api.py b/freezer_api/db/sqlalchemy/api.py index 76513622..f1c8e2a5 100644 --- a/freezer_api/db/sqlalchemy/api.py +++ b/freezer_api/db/sqlalchemy/api.py @@ -279,7 +279,7 @@ def get_recursively(source_dict, search_keys): provided. """ search_keys_found = {} - for key, value in source_dict.iteritems(): + for key, value in source_dict.items(): if key in search_keys.keys(): search_keys_found[key] = value elif isinstance(value, dict): @@ -297,44 +297,55 @@ def get_recursively(source_dict, search_keys): return search_keys_found +def valid_and_get_search_option(search=None): + if not search: + return {} + match_opt = search.get('match', {}) + match_not_opt = search.get('match_not', {}) + if not match_opt and not match_not_opt: + return {} + if len(match_opt) and '_all' in match_opt[0].keys(): + error_msg = ''' + Freezer-api supported by sqlalchemy: + The option --search in freezer cmd needs a list about {key, value}, + such as: + '[{"client_id": "node1_2"}]' + '[{"client_id": "node1_2"}, {"status": "completed"}]' + ''' + search = match_opt[0].get('_all') + try: + search = json_utils.json_decode(search) + except Exception as e: + msg = "{0} \n json_decode error: {1}".\ + format(error_msg, e) + LOG.error(error_msg) + raise Exception(msg) + if not isinstance(search, list): + LOG.error(error_msg) + raise Exception(error_msg) + search = {'match': search} + if not isinstance(search, dict): + search = {} + return search + + def filter_tuple_by_search_opt(tuples, search=None): search = search or {} # search opt is null, all tuples will be filtered in. if len(search) == 0: return tuples - # - match_opt = search.get('match') - if len(match_opt) and '_all' in match_opt[0].keys(): - # From python-freezer lient cli, if cli --search is - # '{match_not": [{"status": completed], - # match: [{client_id": node1_2"}]}, it will be converted to - # {'match': [{'_all': '{"match_not": [{"status": "completed"}], - # "match": [{"client_id": "node1_2"}]}'}]} - search = match_opt[0].get('_all') - try: - search = json_utils.json_decode(search) - except Exception as e: - msg = "search opt error, please input correct format." \ - " search opt {0} will be not use. error msg: {1}".\ - format(search, e) - LOG.error(msg) - raise Exception(msg) - search = {} - if not isinstance(search, dict): - search = {} - + search = valid_and_get_search_option(search=search) if len(search) == 0: return tuples - result_last = [] search_key = {} for m in search.get('match', []): - key = m.keys()[0] - search_key[key] = m.get(key) + for key, value in m.items(): + search_key[key] = value for m in search.get('match_not', []): - key = m.keys()[0] - search_key[key] = m.get(key) + for key, value in m.items(): + search_key[key] = value for tuple in tuples: filter_out = False @@ -342,19 +353,17 @@ def filter_tuple_by_search_opt(tuples, search=None): # If all keys and values are in search_keys_found, this tuple will be # filtered in, otherwise filtered out for m in search.get('match', []): - key = m.keys()[0] - value = m.get(key) - if value != search_keys_found.get(key): - filter_out = True - break + for key, value in m.items(): + if value != search_keys_found.get(key): + filter_out = True + break # If one keys and values are in search_keys_found, this tuple will be # filtered out, otherwise filtered in. for m in search.get('match_not', []): - key = m.keys()[0] - value = m.get(key) - if value == search_keys_found.get(key): - filter_out = True - break + for key, value in m.items(): + if value == search_keys_found.get(key): + filter_out = True + break if not filter_out: result_last.append(tuple) return result_last @@ -729,7 +738,12 @@ def search_job(user_id, project_id=None, offset=0, job.get('job_actions')) jobs.append(jobmap) - jobs = filter_tuple_by_search_opt(jobs, search) + # If search opt is wrong, filter will not work, + # return all tuples. + try: + jobs = filter_tuple_by_search_opt(jobs, search) + except Exception as e: + LOG.error(e) return jobs @@ -1043,5 +1057,10 @@ def search_session(user_id, project_id=None, offset=0, if jobt is not None: sessionmap['jobs'] = json_utils.json_decode(sessiont.get('job')) sessions.append(sessionmap) - sessions = filter_tuple_by_search_opt(sessions, search) + # If search opt is wrong, filter will not work, + # return all tuples. + try: + sessions = filter_tuple_by_search_opt(sessions, search) + except Exception as e: + LOG.error(e) return sessions diff --git a/freezer_api/tests/unit/sqlalchemy/test_job.py b/freezer_api/tests/unit/sqlalchemy/test_job.py index b13b3f5c..6ff6a2bb 100644 --- a/freezer_api/tests/unit/sqlalchemy/test_job.py +++ b/freezer_api/tests/unit/sqlalchemy/test_job.py @@ -159,7 +159,7 @@ class DbJobTestCase(base.DbTestCase): self.assertEqual(result, self.fake_job_id) - def test_add_and_search_job(self): + def test_job_list_without_search(self): count = 0 jobids = [] while (count < 20): @@ -185,3 +185,103 @@ class DbJobTestCase(base.DbTestCase): for index in range(len(result)): jobmap = result[index] self.assertEqual(jobids[index], jobmap['job_id']) + + def test_job_list_with_search_match_and_match_not(self): + count = 0 + jobids = [] + while (count < 20): + doc = copy.deepcopy(self.fake_job_3) + if count in [0, 4, 8, 12, 16]: + doc['client_id'] = "node1" + if count in [4, 12]: + doc['job_schedule']['schedule_interval'] = '10 days' + job_id = self.dbapi.add_job(user_id=self.fake_job_3. + get('user_id'), + doc=doc, + project_id=self.fake_project_id) + self.assertIsNotNone(job_id) + jobids.append(job_id) + count += 1 + search_opt = {'match_not': [{'schedule_interval': '10 days'}], + 'match': [{'client_id': 'node1'}]} + result = self.dbapi.search_job(user_id=self.fake_job_3. + get('user_id'), + project_id=self.fake_project_id, + offset=0, + limit=20, + search=search_opt) + + self.assertIsNotNone(result) + self.assertEqual(len(result), 3) + for index in range(len(result)): + jobmap = result[index] + self.assertEqual('node1', jobmap['client_id']) + self.assertEqual('14 days', + jobmap['job_schedule']['schedule_interval']) + + def test_job_list_with_search_match_list(self): + count = 0 + jobids = [] + while (count < 20): + doc = copy.deepcopy(self.fake_job_3) + if count in [0, 4, 8, 12, 16]: + doc['client_id'] = "node1" + if count in [4, 12]: + doc['job_schedule']['schedule_interval'] = '10 days' + job_id = self.dbapi.add_job(user_id=self.fake_job_3. + get('user_id'), + doc=doc, + project_id=self.fake_project_id) + self.assertIsNotNone(job_id) + jobids.append(job_id) + count += 1 + search_opt = {'match': [{'client_id': 'node1'}, + {'schedule_interval': '10 days'}]} + result = self.dbapi.search_job(user_id=self.fake_job_3. + get('user_id'), + project_id=self.fake_project_id, + offset=0, + limit=20, + search=search_opt) + + self.assertIsNotNone(result) + self.assertEqual(len(result), 2) + for index in range(len(result)): + jobmap = result[index] + self.assertEqual('node1', jobmap['client_id']) + self.assertEqual('10 days', + jobmap['job_schedule']['schedule_interval']) + + def test_job_list_with_search_match_not_list(self): + count = 0 + jobids = [] + while (count < 20): + doc = copy.deepcopy(self.fake_job_3) + if count in [0, 4, 8, 12, 16]: + doc['client_id'] = "node1" + if count in [4, 12]: + doc['job_schedule']['schedule_interval'] = '10 days' + job_id = self.dbapi.add_job(user_id=self.fake_job_3. + get('user_id'), + doc=doc, + project_id=self.fake_project_id) + self.assertIsNotNone(job_id) + jobids.append(job_id) + count += 1 + search_opt = {'match_not': + [{'schedule_interval': '10 days'}, + {'client_id': 'mytenantid_myhostname2'}]} + result = self.dbapi.search_job(user_id=self.fake_job_3. + get('user_id'), + project_id=self.fake_project_id, + offset=0, + limit=20, + search=search_opt) + + self.assertIsNotNone(result) + self.assertEqual(len(result), 3) + for index in range(len(result)): + jobmap = result[index] + self.assertEqual('node1', jobmap['client_id']) + self.assertEqual('14 days', + jobmap['job_schedule']['schedule_interval'])