ClickHouse/utils/github/query.py

414 lines
16 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
import requests
class Query:
2019-04-11 15:08:25 +00:00
'''Implements queries to the Github API using GraphQL
'''
def __init__(self, token, max_page_size=100, min_page_size=5):
self._token = token
self._max_page_size = max_page_size
self._min_page_size = min_page_size
2019-09-06 12:42:15 +00:00
self.api_costs = {}
_MEMBERS = '''
organization(login: "{organization}") {{
team(slug: "{team}") {{
members(first: {max_page_size} {next}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
login
}}
}}
}}
}}
'''
def get_members(self, organization, team):
'''Get all team members for organization
Returns:
logins: a list of members' logins
'''
logins = []
not_end = True
query = Query._MEMBERS.format(organization=organization,
team=team,
max_page_size=self._max_page_size,
next='')
while not_end:
result = self._run(query)['organization']['team']
if result is None:
break
result = result['members']
not_end = result['pageInfo']['hasNextPage']
query = Query._MEMBERS.format(organization=organization,
team=team,
max_page_size=self._max_page_size,
next=f'after: "{result["pageInfo"]["endCursor"]}"')
logins += [node['login'] for node in result['nodes']]
return logins
2019-04-11 15:08:25 +00:00
_LABELS = '''
repository(owner: "yandex" name: "ClickHouse") {{
pullRequest(number: {number}) {{
labels(first: {max_page_size} {next}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
name
color
}}
}}
}}
}}
'''
2019-04-11 15:08:25 +00:00
def get_labels(self, pull_request):
'''Fetchs all labels for given pull-request
Args:
pull_request: JSON object returned by `get_pull_requests()`
2019-04-11 15:08:25 +00:00
Returns:
labels: a list of JSON nodes with the name and color fields
'''
labels = [label for label in pull_request['labels']['nodes']]
not_end = pull_request['labels']['pageInfo']['hasNextPage']
query = Query._LABELS.format(number = pull_request['number'],
max_page_size = self._max_page_size,
next=f'after: "{pull_request["labels"]["pageInfo"]["endCursor"]}"')
2019-04-08 19:28:08 +00:00
while not_end:
result = self._run(query)['repository']['pullRequest']['labels']
not_end = result['pageInfo']['hasNextPage']
query = Query._LABELS.format(number=pull_request['number'],
max_page_size=self._max_page_size,
next=f'after: "{result["pageInfo"]["endCursor"]}"')
2019-04-11 15:08:25 +00:00
labels += [label for label in result['nodes']]
return labels
_TIMELINE = '''
repository(owner: "yandex" name: "ClickHouse") {{
pullRequest(number: {number}) {{
timeline(first: {max_page_size} {next}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
... on CrossReferencedEvent {{
isCrossRepository
source {{
... on PullRequest {{
number
baseRefName
merged
labels(first: {max_page_size}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
name
color
}}
}}
}}
}}
target {{
... on PullRequest {{
number
}}
}}
}}
}}
}}
}}
}}
'''
def get_timeline(self, pull_request):
'''Fetchs all cross-reference events from pull-request's timeline
Args:
pull_request: JSON object returned by `get_pull_requests()`
Returns:
events: a list of JSON nodes for CrossReferenceEvent
'''
events = [event for event in pull_request['timeline']['nodes'] if event and event['source']]
not_end = pull_request['timeline']['pageInfo']['hasNextPage']
query = Query._TIMELINE.format(number = pull_request['number'],
max_page_size = self._max_page_size,
next=f'after: "{pull_request["timeline"]["pageInfo"]["endCursor"]}"')
while not_end:
result = self._run(query)['repository']['pullRequest']['timeline']
not_end = result['pageInfo']['hasNextPage']
query = Query._TIMELINE.format(number=pull_request['number'],
max_page_size=self._max_page_size,
next=f'after: "{result["pageInfo"]["endCursor"]}"')
events += [event for event in result['nodes'] if event and event['source']]
return events
2019-04-11 15:08:25 +00:00
_PULL_REQUESTS = '''
repository(owner: "yandex" name: "ClickHouse") {{
defaultBranchRef {{
name
target {{
... on Commit {{
history(first: {max_page_size} {next}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
oid
associatedPullRequests(first: {min_page_size}) {{
2019-04-11 15:08:25 +00:00
totalCount
nodes {{
... on PullRequest {{
id
2019-04-11 15:08:25 +00:00
number
author {{
login
}}
bodyText
mergedBy {{
login
}}
2019-04-11 15:08:25 +00:00
url
baseRefName
baseRepository {{
nameWithOwner
}}
mergeCommit {{
oid
}}
labels(first: {min_page_size}) {{
2019-04-11 15:08:25 +00:00
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
name
color
}}
}}
timeline(first: {min_page_size}) {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
... on CrossReferencedEvent {{
isCrossRepository
source {{
... on PullRequest {{
number
baseRefName
merged
labels(first: 0) {{
nodes {{
name
}}
}}
}}
}}
target {{
... on PullRequest {{
number
}}
}}
}}
}}
}}
2019-04-11 15:08:25 +00:00
}}
}}
}}
}}
}}
}}
}}
}}
}}
'''
def get_pull_requests(self, before_commit, login):
2019-04-11 15:08:25 +00:00
'''Get all merged pull-requests from the HEAD of default branch to the last commit (excluding)
Args:
before_commit (string-convertable): commit sha of the last commit (excluding)
login (string): filter pull-requests by user login
2019-04-11 15:08:25 +00:00
Returns:
pull_requests: a list of JSON nodes with pull-requests' details
'''
pull_requests = []
not_end = True
query = Query._PULL_REQUESTS.format(max_page_size=self._max_page_size,
min_page_size=self._min_page_size,
next='')
2019-04-11 15:08:25 +00:00
while not_end:
result = self._run(query)['repository']['defaultBranchRef']
2019-04-11 15:08:25 +00:00
default_branch_name = result['name']
result = result['target']['history']
not_end = result['pageInfo']['hasNextPage']
query = Query._PULL_REQUESTS.format(max_page_size=self._max_page_size,
min_page_size=self._min_page_size,
next=f'after: "{result["pageInfo"]["endCursor"]}"')
2019-04-11 15:08:25 +00:00
for commit in result['nodes']:
if str(commit['oid']) == str(before_commit):
not_end = False
break
2019-04-11 16:16:57 +00:00
# TODO: fetch all pull-requests that were merged in a single commit.
assert commit['associatedPullRequests']['totalCount'] <= self._min_page_size, \
2019-04-11 15:08:25 +00:00
f'there are {commit["associatedPullRequests"]["totalCount"]} pull-requests merged in commit {commit["oid"]}'
2019-04-11 15:08:25 +00:00
for pull_request in commit['associatedPullRequests']['nodes']:
2019-04-08 19:28:08 +00:00
if(pull_request['baseRepository']['nameWithOwner'] == 'yandex/ClickHouse' and
pull_request['baseRefName'] == default_branch_name and
2019-04-11 16:16:57 +00:00
pull_request['mergeCommit']['oid'] == commit['oid'] and
(not login or pull_request['author']['login'] == login)):
pull_requests.append(pull_request)
return pull_requests
2019-04-11 15:08:25 +00:00
_DEFAULT = '''
repository(owner: "yandex", name: "ClickHouse") {
defaultBranchRef {
name
}
}
'''
def get_default_branch(self):
2019-04-11 15:08:25 +00:00
'''Get short name of the default branch
2019-04-11 15:08:25 +00:00
Returns:
name (string): branch name
'''
return self._run(Query._DEFAULT)['repository']['defaultBranchRef']['name']
2019-04-11 16:16:57 +00:00
_GET_LABEL = '''
repository(owner: "yandex" name: "ClickHouse") {{
labels(first: {max_page_size} {next} query: "{name}") {{
pageInfo {{
hasNextPage
endCursor
}}
nodes {{
id
name
color
}}
}}
}}
'''
_SET_LABEL = '''
addLabelsToLabelable(input: {{ labelableId: "{pr_id}", labelIds: "{label_id}" }}) {{
clientMutationId
}}
'''
def set_label(self, pull_request, label_name):
'''Set label by name to the pull request
Args:
pull_request: JSON object returned by `get_pull_requests()`
label_name (string): label name
'''
labels = []
not_end = True
query = Query._GET_LABEL.format(name=label_name,
max_page_size=self._max_page_size,
next='')
while not_end:
result = self._run(query)['repository']['labels']
not_end = result['pageInfo']['hasNextPage']
query = Query._GET_LABEL.format(name=label_name,
max_page_size=self._max_page_size,
next=f'after: "{result["pageInfo"]["endCursor"]}"')
labels += [label for label in result['nodes']]
if not labels:
return
query = Query._SET_LABEL.format(pr_id = pull_request['id'], label_id = labels[0]['id'])
self._run(query, is_mutation=True)
pull_request['labels']['nodes'].append(labels[0])
def _run(self, query, is_mutation=False):
2019-04-11 15:08:25 +00:00
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry
def requests_retry_session(
retries=3,
backoff_factor=0.3,
status_forcelist=(500, 502, 504),
session=None,
):
session = session or requests.Session()
retry = Retry(
total=retries,
read=retries,
connect=retries,
backoff_factor=backoff_factor,
status_forcelist=status_forcelist,
)
adapter = HTTPAdapter(max_retries=retry)
session.mount('http://', adapter)
session.mount('https://', adapter)
return session
headers = {'Authorization': f'bearer {self._token}'}
if is_mutation:
query = f'''
mutation {{
{query}
2019-09-06 12:42:15 +00:00
}}
'''
else:
query = f'''
query {{
{query}
rateLimit {{
cost
remaining
}}
}}
'''
2019-04-11 15:08:25 +00:00
request = requests_retry_session().post('https://api.github.com/graphql', json={'query': query}, headers=headers)
if request.status_code == 200:
result = request.json()
if 'errors' in result:
raise Exception(f'Errors occured: {result["errors"]}')
2019-09-06 12:42:15 +00:00
if not is_mutation:
import inspect
caller = inspect.getouterframes(inspect.currentframe(), 2)[1][3]
if caller not in self.api_costs.keys():
self.api_costs[caller] = 0
self.api_costs[caller] += result['data']['rateLimit']['cost']
2019-09-06 12:42:15 +00:00
return result['data']
else:
import json
raise Exception(f'Query failed with code {request.status_code}:\n{json.dumps(request.json(), indent=4)}')