diff options
author | grakiss <grakiss.wanglei@huawei.com> | 2017-09-28 03:47:54 -0400 |
---|---|---|
committer | grakiss <grakiss.wanglei@huawei.com> | 2017-09-28 05:15:01 -0400 |
commit | 0cf6b232ac9cf128ee9183a27c08f4f74ab2e2e6 (patch) | |
tree | 7be233b8f8f65fa968c7b93f1d1e75b691952ed9 /cvp/opnfv_testapi/tests/unit/fake_pymongo.py | |
parent | 63c2e2aa4b8d86349a767f611123ceafc19fa6d6 (diff) |
add api&web services for cvp
JIRA: DOVETAIL-512
add api&web services for cvp
Change-Id: I9ef9525e980fe61dc3108035ef9a3ff8783b2697
Signed-off-by: grakiss <grakiss.wanglei@huawei.com>
Diffstat (limited to 'cvp/opnfv_testapi/tests/unit/fake_pymongo.py')
-rw-r--r-- | cvp/opnfv_testapi/tests/unit/fake_pymongo.py | 284 |
1 files changed, 284 insertions, 0 deletions
diff --git a/cvp/opnfv_testapi/tests/unit/fake_pymongo.py b/cvp/opnfv_testapi/tests/unit/fake_pymongo.py new file mode 100644 index 00000000..0ca83df6 --- /dev/null +++ b/cvp/opnfv_testapi/tests/unit/fake_pymongo.py @@ -0,0 +1,284 @@ +############################################################################## +# Copyright (c) 2016 ZTE Corporation +# feng.xiaowei@zte.com.cn +# All rights reserved. This program and the accompanying materials +# are made available under the terms of the Apache License, Version 2.0 +# which accompanies this distribution, and is available at +# http://www.apache.org/licenses/LICENSE-2.0 +############################################################################## +from operator import itemgetter + +from bson.objectid import ObjectId +from concurrent.futures import ThreadPoolExecutor + + +def thread_execute(method, *args, **kwargs): + with ThreadPoolExecutor(max_workers=2) as executor: + result = executor.submit(method, *args, **kwargs) + return result + + +class MemCursor(object): + def __init__(self, collection): + self.collection = collection + self.length = len(self.collection) + self.sorted = [] + + def _is_next_exist(self): + return self.length != 0 + + @property + def fetch_next(self): + return thread_execute(self._is_next_exist) + + def next_object(self): + self.length -= 1 + return self.collection.pop() + + def sort(self, key_or_list): + for k, v in key_or_list.iteritems(): + if v == -1: + reverse = True + else: + reverse = False + + self.collection = sorted(self.collection, + key=itemgetter(k), reverse=reverse) + return self + + def limit(self, limit): + if limit != 0 and limit < len(self.collection): + self.collection = self.collection[0: limit] + self.length = limit + return self + + def skip(self, skip): + if skip < self.length and (skip > 0): + self.collection = self.collection[self.length - skip: -1] + self.length -= skip + elif skip >= self.length: + self.collection = [] + self.length = 0 + return self + + def _count(self): + return self.length + + def count(self): + return thread_execute(self._count) + + +class MemDb(object): + + def __init__(self, name): + self.name = name + self.contents = [] + pass + + def _find_one(self, spec_or_id=None, *args): + if spec_or_id is not None and not isinstance(spec_or_id, dict): + spec_or_id = {"_id": spec_or_id} + if '_id' in spec_or_id: + spec_or_id['_id'] = str(spec_or_id['_id']) + cursor = self._find(spec_or_id, *args) + for result in cursor: + return result + return None + + def find_one(self, spec_or_id=None, *args): + return thread_execute(self._find_one, spec_or_id, *args) + + def _insert(self, doc_or_docs, check_keys=True): + + docs = doc_or_docs + return_one = False + if isinstance(docs, dict): + return_one = True + docs = [docs] + + if check_keys: + for doc in docs: + self._check_keys(doc) + + ids = [] + for doc in docs: + if '_id' not in doc: + doc['_id'] = str(ObjectId()) + if not self._find_one(doc['_id']): + ids.append(doc['_id']) + self.contents.append(doc_or_docs) + + if len(ids) == 0: + return None + if return_one: + return ids[0] + else: + return ids + + def insert(self, doc_or_docs, check_keys=True): + return thread_execute(self._insert, doc_or_docs, check_keys) + + @staticmethod + def _compare_date(spec, value): + gte = True + lt = False + for k, v in spec.iteritems(): + if k == '$gte' and value < v: + gte = False + elif k == '$lt' and value < v: + lt = True + return gte and lt + + def _in(self, content, *args): + if self.name == 'scenarios': + return self._in_scenarios(content, *args) + else: + return self._in_others(content, *args) + + def _in_scenarios_installer(self, installer, content): + hit = False + for s_installer in content['installers']: + if installer == s_installer['installer']: + hit = True + + return hit + + def _in_scenarios_version(self, version, content): + hit = False + for s_installer in content['installers']: + for s_version in s_installer['versions']: + if version == s_version['version']: + hit = True + return hit + + def _in_scenarios_project(self, project, content): + hit = False + for s_installer in content['installers']: + for s_version in s_installer['versions']: + for s_project in s_version['projects']: + if project == s_project['project']: + hit = True + + return hit + + def _in_scenarios(self, content, *args): + for arg in args: + for k, v in arg.iteritems(): + if k == 'installers': + for inner in v.values(): + for i_k, i_v in inner.iteritems(): + if i_k == 'installer': + return self._in_scenarios_installer(i_v, + content) + elif i_k == 'versions.version': + return self._in_scenarios_version(i_v, + content) + elif i_k == 'versions.projects.project': + return self._in_scenarios_project(i_v, + content) + elif content.get(k, None) != v: + return False + + return True + + def _in_others(self, content, *args): + for arg in args: + for k, v in arg.iteritems(): + if k == 'start_date': + if not MemDb._compare_date(v, content.get(k)): + return False + elif k == 'trust_indicator.current': + if content.get('trust_indicator').get('current') != v: + return False + elif not isinstance(v, dict) and content.get(k, None) != v: + return False + return True + + def _find(self, *args): + res = [] + for content in self.contents: + if self._in(content, *args): + res.append(content) + + return res + + def find(self, *args): + return MemCursor(self._find(*args)) + + def _aggregate(self, *args, **kwargs): + res = self.contents + print args + for arg in args[0]: + for k, v in arg.iteritems(): + if k == '$match': + res = self._find(v) + cursor = MemCursor(res) + for arg in args[0]: + for k, v in arg.iteritems(): + if k == '$sort': + cursor = cursor.sort(v) + elif k == '$skip': + cursor = cursor.skip(v) + elif k == '$limit': + cursor = cursor.limit(v) + return cursor + + def aggregate(self, *args, **kwargs): + return self._aggregate(*args, **kwargs) + + def _update(self, spec, document, check_keys=True): + updated = False + + if check_keys: + self._check_keys(document) + + for index in range(len(self.contents)): + content = self.contents[index] + if self._in(content, spec): + for k, v in document.iteritems(): + updated = True + content[k] = v + self.contents[index] = content + return updated + + def update(self, spec, document, check_keys=True): + return thread_execute(self._update, spec, document, check_keys) + + def _remove(self, spec_or_id=None): + if spec_or_id is None: + self.contents = [] + if not isinstance(spec_or_id, dict): + spec_or_id = {'_id': spec_or_id} + for index in range(len(self.contents)): + content = self.contents[index] + if self._in(content, spec_or_id): + del self.contents[index] + return True + return False + + def remove(self, spec_or_id=None): + return thread_execute(self._remove, spec_or_id) + + def clear(self): + self._remove() + + def _check_keys(self, doc): + for key in doc.keys(): + if '.' in key: + raise NameError('key {} must not contain .'.format(key)) + if key.startswith('$'): + raise NameError('key {} must not start with $'.format(key)) + if isinstance(doc.get(key), dict): + self._check_keys(doc.get(key)) + + +def __getattr__(name): + return globals()[name] + + +pods = MemDb('pods') +projects = MemDb('projects') +testcases = MemDb('testcases') +results = MemDb('results') +scenarios = MemDb('scenarios') +tokens = MemDb('tokens') |