# Copyright 2011 OpenStack, LLC. # Copyright 2012 Hewlett-Packard Development Company, L.P. # Copyright 2023-2025 Acme Gating, LLC # # Licensed under the Apache License, Version 2.0 (the "License"); you may # not use this file except in compliance with the License. You may obtain # a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the # License for the specific language governing permissions and limitations # under the License. import collections import copy import datetime import enum import itertools import json import logging import paramiko import pprint import re import re2 import requests import shlex import threading import time import urllib import urllib.parse import weakref import cachetools from typing import Dict, List from uuid import uuid4 from opentelemetry import trace from zuul import version as zuul_version from zuul.connection import ( BaseConnection, BaseThreadPoolEventConnector, ZKBranchCacheMixin, ZKChangeCacheMixin, ) from zuul.driver.gerrit.auth import FormAuth from zuul.driver.gerrit.gcloudauth import GCloudAuth from zuul.driver.gerrit.gerritmodel import GerritChange, GerritTriggerEvent from zuul.driver.gerrit.gerriteventssh import GerritSSHEventListener from zuul.driver.gerrit.gerriteventchecks import GerritChecksPoller from zuul.driver.gerrit.gerriteventkafka import GerritKafkaEventListener from zuul.driver.gerrit.gerriteventawskinesis import ( GerritAWSKinesisEventListener, ) from zuul.driver.gerrit.gerriteventgcloudpubsub import ( GerritGcloudPubsubEventListener, ) from zuul.driver.git.gitwatcher import GitWatcher from zuul.lib import tracing from zuul.lib.logutil import get_annotated_logger from zuul.model import Ref, Tag, Branch, Project from zuul.zk.branch_cache import BranchCache, BranchFlag, BranchInfo from zuul.zk.change_cache import ( AbstractChangeCache, ChangeKey, ConcurrentUpdateError, ) from zuul.zk.event_queues import ConnectionEventQueue # HTTP timeout in seconds TIMEOUT = 30 # SSH connection timeout SSH_TIMEOUT = TIMEOUT # commentSizeLimit default set by Gerrit. Gerrit is a bit # vague about what this means, it says # # Comments which exceed this size will be rejected ... Size # computation is approximate and may be off by roughly 1% ... # Default is 16k # # This magic number is int((16 << 10) * 0.98). Robot comments # are accounted for separately. GERRIT_HUMAN_MESSAGE_LIMIT = 16056 class HTTPConflictException(Exception): message = "Received response 409" class HTTPBadRequestException(Exception): pass class HTTPNotFoundException(Exception): pass class GerritEventProcessingException(Exception): pass class GerritChangeCache(AbstractChangeCache): log = logging.getLogger("zuul.driver.GerritChangeCache") CHANGE_TYPE_MAP = { "Ref": Ref, "Tag": Tag, "Branch": Branch, "GerritChange": GerritChange, } class ChangeNetworkFuture: """A set of changes related by dependencies""" def __init__(self, min_ltime): self.changes = set() self.event = threading.Event() self.query_results = {} # We will not accept cached data older than this time self.min_ltime = min_ltime def add(self, change_key): self.changes.add(change_key) def setComplete(self): self.event.set() def wait(self): self.event.wait() def hasNumber(self, number): for change_key in self.changes: if number == change_key.stable_id: return True return False def addQueryResult(self, number, data): existing = self.query_results.get(number) existing_time = existing and existing.zuul_query_ltime or 0 if existing_time < data.zuul_query_ltime: self.query_results[number] = data def getQueryResult(self, number): data = self.query_results.get(number) return data def mergeQueryResults(self, future): for (number, data) in future.query_results.items(): if number not in self.query_results: self.query_results[number] = data else: own_data = self.query_results[number] if own_data.zuul_query_ltime < data.zuul_query_ltime: self.query_results[number] = data class ChangeNetworkConflict(Exception): """Raised when two threads are found to be walking the same change graph. The associated ChangeNetworkFuture is the winner of the conflict and the caller should wait for it. """ def __init__(self, future): self.future = future class ChangeNetworkManager: """Prevent collisions between multiple threads querying changes""" def __init__(self): self.lock = threading.Lock() self.futures = [] self.query_locks = weakref.WeakValueDictionary() def getQueryLock(self, number): # TODO: verify that WeakValueDictionary.setdefault is atomic; # if so, we can remove this internal locking call. with self.lock: return self.query_locks.setdefault(number, threading.Lock()) def getQueryResult(self, number): with self.lock: for f in self.futures: if f.hasNumber(number): data = f.getQueryResult(number) if data: return data def updateQueryResult(self, number, data): with self.lock: for f in self.futures: if f.hasNumber(number): f.addQueryResult(number, data) def permissionToProceed(self, future): with self.lock: for f in self.futures: if f is future: continue if future.changes.intersection(f.changes): # The first thread to request permission for a # change always succeeds, second or later threads # connected to the same network always lose. if future in self.futures: self.futures.remove(future) # Let the winner benefit from any queries the # loser has performed. f.mergeQueryResults(future) raise ChangeNetworkConflict(f) # No conflicts if future not in self.futures: self.futures.append(future) return True def setComplete(self, future): if future is None: return with self.lock: try: self.futures.remove(future) except ValueError: pass future.setComplete() class GerritChangeData(object): """Compatability layer for SSH/HTTP This class holds the raw data returned from a change query over SSH or HTTP. Most of the work of parsing the data and storing it on the change is in the gerritmodel.GerritChange class, however this does perform a small amount of parsing of dependencies since they are handled outside of that class. This provides an API to that data independent of the source. """ SSH = 1 HTTP = 2 def __init__(self, fmt, data, related=None, files=None, commentable_files=None, zuul_query_ltime=None): self.format = fmt self.data = data self.files = files self.commentable_files = commentable_files self.zuul_query_ltime = zuul_query_ltime self.submit_type = 'MERGE_IF_NECESSARY' if fmt == self.SSH: self.parseSSH(data) else: self.parseHTTP(data) if related: self.parseRelatedHTTP(data, related) def parseSSH(self, data): self.needed_by = [] self.depends_on = None self.message = data['commitMessage'] self.current_patchset = str(data['currentPatchSet']['number']) self.number = str(data['number']) self.id = data['id'] if 'dependsOn' in data: parts = data['dependsOn'][0]['ref'].split('/') self.depends_on = (parts[3], parts[4]) for needed in data.get('neededBy', []): parts = needed['ref'].split('/') self.needed_by.append((parts[3], parts[4])) def parseHTTP(self, data): rev = data['revisions'][data['current_revision']] self.message = rev['commit']['message'] self.current_patchset = str(rev['_number']) self.number = str(data['_number']) self.id = data['change_id'] def parseRelatedHTTP(self, data, related): self.needed_by = [] self.depends_on = None current_rev = data['revisions'][data['current_revision']] self.submit_type = data.get('submit_type', self.submit_type) for change in related['changes']: # For cherry-pick only, allow the dependency to float to # the latest revision since Gerrit will not prohibit # merges with dependencies on older versions. We also # want Zuul to test the latest patchset if possible, since # that is what will end up being merged. if self.submit_type == 'CHERRY_PICK': ps = change['_current_revision_number'] else: ps = change['_revision_number'] for parent in current_rev['commit']['parents']: if change['commit']['commit'] == parent['commit']: self.depends_on = (change['_change_number'], ps) break else: self.needed_by.append((change['_change_number'], ps)) class QueryHistory: class Query(enum.Enum): SEEN = 1 # Not a real query, just that we've seen the change CHANGE = 2 # The main change query def __init__(self): self.queries = collections.defaultdict(lambda: dict()) def getByKey(self, query, change_key): if not isinstance(change_key, ChangeKey): raise Exception("Must supply a ChangeKey") key = (change_key.stable_id, change_key.revision) return self.queries[query].get(key) def add(self, query, change): if not isinstance(change, GerritChange): raise Exception("Must supply a GerritChange") key = (change.number, change.patchset) self.queries[query][key] = change class PeekQueue: # If we see events whose timestamp is this long since an event # we're waiting for, give up the wait. This is a 10 second propogation # delay. Replication timeouts are handled separately and are operator # configurable. timeout = 10 def __init__(self, handler, replication_timeout): self.queue = collections.deque() self.handler = handler self.replication_timeout = replication_timeout self.change_merged_cache = cachetools.LRUCache(128) def append(self, event): self.queue.append(event) def _getProjectAndRef(self, data): # We need to match up ref-updated, change-merged, and patchset-created # events to various replication events based on project, ref tuples. # Unfortunately gerrit events don't use consistent data structures to # present this data across the events. Pull the required info out # based on the event type with fallback. kind = data.get('type') if kind == 'ref-updated': refupdate = data.get('refUpdate', {}) return refupdate.get('project'), refupdate.get('refName') elif kind == 'patchset-created': change = data.get('change', {}) patchset = data.get('patchSet', {}) return change.get('project'), patchset.get('ref') elif kind == 'change-merged': change = data.get('change', {}) # change-merged events need to match up the branch name to # replication events and not the change ref (it is already # replicated and captured by patchset-created events). return change.get('project'), data.get('refName') elif (project := data.get('project')) and (ref := data.get('ref')): return project, ref else: return None, None def run(self): if not self.queue: return # Try to do two things: # Hold events until they have replicated and # Attempt to match ref-updated events with change-merged # events. ref_replication = collections.defaultdict(list) ref_updates = {} new_event_list = collections.deque() for event in self.queue: data = event["payload"] kind = data.get('type') refupdate = data.get('refUpdate', {}) ref = refupdate.get('refName') inserted = False if kind == 'ref-replication-scheduled': # Note we can get many ref-replication-scheduled events for # a single ref-replication-done event. We can also get # overlapping ref-replication-scheduled events for the same # ref. For this reason we don't use the -done events and # instead rely on counting matching pairs of -scheduled and # ref-replicated events. When all have paired up or we timeout # the related events are considered valid. # # Replication events don't use the same refUpdate and refName # conventions... ref = data.get('ref') project = data.get('project') ref_replication[(project, ref)].append(event) elif kind == 'ref-replicated': ref = data.get('ref') project = data.get('project') replication_events = ref_replication[(project, ref)] # TODO(clarkb) Is it necessary to wait for success? if (len(replication_events) and data.get("status") == "succeeded"): # Its possible we start listening when things have already # started replicating. In that case we'll empty the # replication events list early. replication_events.pop() elif (kind == 'ref-updated' and ((not ref.startswith('refs/')) or ref.startswith('refs/heads/'))): # This is a ref-updated event for a branch, we # want to find its change-merged event. newrev = refupdate.get('newRev') if newrev in self.change_merged_cache: # As of writing, this should not happen, but if it # does it means gerrit has reversed the order of # events! del self.change_merged_cache[newrev] else: ref_updates[newrev] = event elif kind == 'change-merged': newrev = data.get('newRev') if newrev in ref_updates: # This is a change-merged event that matches a # ref-updated event we're interested in. other_event = ref_updates.pop(newrev) idx = new_event_list.index(other_event) # Put our event immediately before the ref-updated event new_event_list.insert(idx, event) # Give both events the same earlier ltime so that # we don't have ltime going backwards. event.zuul_event_ltime = other_event.zuul_event_ltime inserted = True else: self.change_merged_cache[newrev] = True if not inserted: new_event_list.append(event) while new_event_list: event = new_event_list.popleft() data = event["payload"] project, ref = self._getProjectAndRef(data) ok = False delay = None now = time.time() time_since_event = now - event["timestamp"] # First check if replication is up to date. if self.replication_timeout <= 0: # We are not configured to look at Gerrit replication # targets. Ignore replication status. ok = True elif ref and project and ref_replication[(project, ref)]: if time_since_event >= self.replication_timeout: # Waited long enough for replication del ref_replication[(project, ref)] ok = True else: # If replication hasn't completed wait longer ok = False # Wait max one second for replication to complete. delay = min(self.replication_timeout - time_since_event, 1) else: # We have replicated or don't have replication events # corresponding to our project,ref. Proceed. ok = True if not ok: return delay # Now check if ref-updated and change-merged events need to be # reordered kind = data.get('type') if kind == 'ref-updated': refupdate = data.get('refUpdate', {}) newrev = refupdate.get('newRev') if newrev in ref_updates: # We're waiting on data for this one if time_since_event >= self.timeout: # It's been 10 seconds since we saw # the event, so it may be missing at this # point; release it. ok = True else: # Otherwise, we're still waiting ok = False delay = min(self.timeout - time_since_event, 1) else: # Not a branch ref-update ok = True else: # Not a ref-update and not waiting for replication ok = True if not ok: return delay self.queue.remove(event) self.handler(event) class GerritEventConnector(BaseThreadPoolEventConnector): """Move events from Gerrit to the scheduler.""" IGNORED_EVENTS = ( 'cache-eviction', # evict-cache plugin 'fetch-ref-replicated', 'fetch-ref-replication-scheduled', 'ref-replication-done' ) log = logging.getLogger("zuul.GerritEventConnector") def __init__(self, connection, replication_timeout): super().__init__(connection) self.replication_timeout = replication_timeout self._peek_queue = PeekQueue( self._peekQueueHandler, replication_timeout) def _getEventProcessor(self, event): return GerritEventProcessor(self, event).run def _dispatchEvents(self): # This is the first half of the event dispatcher. It reads # events from the ssh stream event queue and passes them to a # concurrent executor for pre-processing. # This overrides the superclass in order to add the peek queue. try: peek_ack_refs = [r.ack_ref for r in self._peek_queue.queue] event_id_offset = max( self.event_queue.eventIdFromAckRef(r) for r in list(self._events_in_progress) + peek_ack_refs) except ValueError: event_id_offset = None delay = None for event in self.event_queue.iter(event_id_offset): if self._shouldStop(): break self._peek_queue.append(event) delay = self._peek_queue.run() if delay: return delay return self._peek_queue.run() def _peekQueueHandler(self, event): # Called when the peek queue has decided an event should be processed processor = self._getEventProcessor(event) future = self._thread_pool.submit(processor) # Events are acknowledged in the event forwarder # loop after pre-processing. This way we can # ensure that no events are lost. self._events_in_progress.add(event.ack_ref) self._event_forward_queue.append(future) class GerritEventProcessor: tracer = trace.get_tracer("zuul") def __init__(self, connector, connection_event): self.connector = connector self.connection = connector.connection self.zk_client = self.connection.sched.zk_client self.connection_event = connection_event self.event_span = tracing.restoreSpanContext( self.connection_event.get("span_context")) logger = logging.getLogger("zuul.GerritEventProcessor") self.zuul_event_id = connection_event["zuul_event_id"] self.log = get_annotated_logger(logger, self.zuul_event_id) self.events = [] def run(self): if self.connector._stopped: return try: attributes = {"rel": "GerritEvent"} link = trace.Link(self.event_span.get_span_context(), attributes=attributes) with self.tracer.start_as_current_span( "GerritEventProcessing", links=[link]): self.events = self._handleEvent(self.connection_event) except GerritEventProcessingException as e: self.log.warning("Skipping event due to %s", e) except Exception: self.log.exception("Skipping event due to:") return self.events, self.connection_event def _handleEvent(self, connection_event): timestamp = connection_event["timestamp"] data = connection_event["payload"] event = GerritTriggerEvent.fromGerritEventDict( data, timestamp, self.connection, self.zuul_event_id) event_type = data.get("type") # Recheck events as we may have only cared about replication events # in preprocessing. if not self.connection.shouldProcessEvent(event_type, event, False): self.log.debug("Event did not match pre-filters after " "preprocessing %s", event) return [] min_change_ltime = self.zk_client.getCurrentLtime() # In order to perform connection hygene actions like those # below, the preFilter method must pass relevant events # through to get to this point. if event.default_branch_changed: self.log.debug('Updating default branch for %s to %s', event.project_name, event.branch) self.connection._branch_cache.setProjectDefaultBranch( event.project_name, event.branch) if event._accountfield_unknown: self.log.warning("Received unrecognized event type '%s' " "from Gerrit. Can not get account information." % (event.type,)) if event._branch_ref_update: self.connection.clearConnectionCacheOnBranchEvent(event) change = self._getChange(event, min_change_ltime) if (change and change.patchset and event.change_number and event.patch_number is None): event.patch_number = str(change.patchset) return [event] def _getChange(self, event, min_change_ltime): # Grab the change if we are managing the project or if it exists in the # cache as it may be a dependency change = None if event.change_number: refresh = True change_key = self.connection.source.getChangeKey(event) change = self.connection._change_cache.get(change_key) if change is None: refresh = False for tenant in self.connection.sched.abide.tenants.values(): # TODO(fungi): it would be better to have some simple means # of inferring the hostname from the connection, or at # least split this into separate method arguments, rather # than assembling and passing in a baked string. if (None, None) != tenant.getProject('/'.join(( self.connection.canonical_hostname, event.project_name))): refresh = True break else: # We have a cache entry for this change Get the # query ltime for the cache entry; if it's after the # event ltime, we don't need to refresh. if (change.zuul_query_ltime and change.zuul_query_ltime > min_change_ltime): refresh = False if refresh: # Call _getChange for the side effect of updating the # cache. Note that this modifies Change objects outside # the main thread. # NOTE(jhesketh): Ideally we'd just remove the change from the # cache to denote that it needs updating. However the change # object is already used by Items and hence BuildSets etc. and # we need to update those objects by reference so that they # have the correct/new information and also avoid hitting # gerrit multiple times. change = self.connection._getChange( change_key, refresh=True, event=event, allow_key_update=True, change=change, update_if_older_than=min_change_ltime) return change class GerritConnection(ZKChangeCacheMixin, ZKBranchCacheMixin, BaseConnection): driver_name = 'gerrit' log = logging.getLogger("zuul.GerritConnection") tracer = trace.get_tracer("zuul") iolog = logging.getLogger("zuul.GerritConnection.io") depends_on_re = re.compile(r"^Depends-On: (I[0-9a-f]{40})\s*$", re.MULTILINE | re.IGNORECASE) refname_bad_sequences = re2.compile( r"[ \\*\[?:^~\x00-\x1F\x7F]|" # Forbidden characters r"@{|\.\.|\.$|^@$|/$|^/|//+") # everything else we can check with re2 is_merged_replication_timeout = 300 replication_retry_interval = 5 _poller_class = GerritChecksPoller _ref_watcher_class = GitWatcher ref_watcher_poll_interval = 60 submit_retry_backoff = 10 EVENT_SOURCE_NONE = 'none' EVENT_SOURCE_STREAM_EVENTS = 'stream-events' EVENT_SOURCE_KAFKA = 'kafka' EVENT_SOURCE_KINESIS = 'kinesis' EVENT_SOURCE_GCLOUD_PUBSUB = 'gcloudpubsub' def __init__(self, driver, connection_name, connection_config): super(GerritConnection, self).__init__(driver, connection_name, connection_config) if 'server' not in self.connection_config: raise Exception('server is required for gerrit connections in ' '%s' % self.connection_name) if 'user' not in self.connection_config: raise Exception('user is required for gerrit connections in ' '%s' % self.connection_name) self.user = self.connection_config.get('user') self.server = self.connection_config.get('server') self.ssh_server = self.connection_config.get('ssh_server', self.server) self.canonical_hostname = self.connection_config.get( 'canonical_hostname', self.server) self.port = int(self.connection_config.get('port', 29418)) self.keyfile = self.connection_config.get('sshkey', None) self.keepalive = int(self.connection_config.get('keepalive', 60)) self.max_dependencies = self.connection_config.get( 'max_dependencies', None) if self.max_dependencies is not None: self.max_dependencies = int(self.max_dependencies) self.event_source = self.EVENT_SOURCE_NONE # TODO(corvus): Document this when the checks api is stable; # it's not useful without it. enable_stream_events = self.connection_config.get( 'stream_events', True) if enable_stream_events in [ 'true', 'True', '1', 1, 'TRUE', True]: self.event_source = self.EVENT_SOURCE_STREAM_EVENTS if self.connection_config.get('kafka_bootstrap_servers', None): self.event_source = self.EVENT_SOURCE_KAFKA elif self.connection_config.get('aws_kinesis_region', None): self.event_source = self.EVENT_SOURCE_KINESIS elif self.connection_config.get('gcloud_pubsub_project', None): self.event_source = self.EVENT_SOURCE_GCLOUD_PUBSUB self.replication_timeout = int(self.connection_config.get( 'replication_timeout', 0)) # Thread for whatever event source we use self.event_thread = None # Next two are only used by checks plugin self.poller_thread = None self.ref_watcher_thread = None self.client = None self.watched_checkers = [] self.project_checker_map = {} self.watched_event_filters = [] self.watched_event_filters_by_tenant = {} self.watched_event_filters_lock = threading.Lock() self.version = (0, 0, 0) self.submit_whole_topic = None self.ssh_timeout = SSH_TIMEOUT self.baseurl = self.connection_config.get( 'baseurl', 'https://%s' % self.server).rstrip('/') default_gitweb_url_template = '{baseurl}/gitweb?' \ 'p={project.name}.git;' \ 'a=commitdiff;h={sha}' url_template = self.connection_config.get('gitweb_url_template', default_gitweb_url_template) self.gitweb_url_template = url_template self.change_network_manager = ChangeNetworkManager() self.projects = {} self.gerrit_event_connector = None self.source = driver.getSource(self) self.session = None self.password = self.connection_config.get('password', None) self.git_over_ssh = self.connection_config.get('git_over_ssh', False) self.auth_type = self.connection_config.get('auth_type', None) self.anonymous_git = False if self.password or self.auth_type == 'gcloud_service': self.verify_ssl = self.connection_config.get('verify_ssl', True) if self.verify_ssl not in ['true', 'True', '1', 1, 'TRUE']: self.verify_ssl = False self.user_agent = 'Zuul/%s %s' % ( zuul_version.release_string, requests.utils.default_user_agent()) self.session = requests.Session() if self.auth_type == 'digest': authclass = requests.auth.HTTPDigestAuth elif self.auth_type == 'form': authclass = FormAuth elif self.auth_type == 'gcloud_service': authclass = GCloudAuth # The executors in google cloud may not have access # to the gerrit account credentials, so just use # anonymous http access for git self.anonymous_git = True else: authclass = requests.auth.HTTPBasicAuth self.auth = authclass(self.user, self.password) def setWatchedCheckers(self, checkers_to_watch): # TODO: This is not safe for multiple tenants # TODO: This does not remove checkers when tenants are deleted self.log.debug("Setting watched checkers to %s", checkers_to_watch) self.watched_checkers = set() self.project_checker_map = {} schemes_to_watch = set() uuids_to_watch = set() for x in checkers_to_watch: if 'scheme' in x: schemes_to_watch.add(x['scheme']) if 'uuid' in x: uuids_to_watch.add(x['uuid']) if schemes_to_watch: # get a list of all configured checkers try: configured_checkers = self.get('plugins/checks/checkers/') except Exception: self.log.exception("Unable to get checkers") configured_checkers = [] # filter it through scheme matches in checkers_to_watch for checker in configured_checkers: if checker['status'] != 'ENABLED': continue checker_scheme, checker_id = checker['uuid'].split(':') repo = checker['repository'] repo = self.canonical_hostname + '/' + repo # map scheme matches to project names if checker_scheme in schemes_to_watch: repo_checkers = self.project_checker_map.setdefault( repo, set()) repo_checkers.add(checker['uuid']) self.watched_checkers.add(checker['uuid']) # add uuids from checkers_to_watch for x in uuids_to_watch: self.watched_checkers.add(x) def setWatchedEventFilters(self, tenant_name, filters): # TODO: This does not remove filters when tenants are deleted self.log.debug("Setting watched event filters for %s to %s", tenant_name, filters) with self.watched_event_filters_lock: self.watched_event_filters_by_tenant[tenant_name] = filters new_filters = set() # Build a set of unique filters across all tenants for tenant_filters in \ self.watched_event_filters_by_tenant.values(): new_filters.update(tenant_filters) self.log.debug("Setting watched event filters to %s", new_filters) self.watched_event_filters = new_filters def toDict(self): d = super().toDict() d.update({ "baseurl": self.baseurl, "canonical_hostname": self.canonical_hostname, "server": self.server, "ssh_server": self.ssh_server, "port": self.port, }) return d def url(self, path): return self.baseurl + '/a/' + path def get(self, path): url = self.url(path) self.log.debug('GET: %s' % (url,)) r = self.session.get( url, verify=self.verify_ssl, auth=self.auth, timeout=TIMEOUT, headers={'User-Agent': self.user_agent}) self.iolog.debug('Received: %s %s' % (r.status_code, r.text,)) if r.status_code == 409: raise HTTPConflictException() elif r.status_code == 404: raise HTTPNotFoundException() elif r.status_code != 200: raise Exception("Received response %s" % (r.status_code,)) ret = None if r.text and len(r.text) > 4: try: ret = json.loads(r.text[4:]) except Exception: self.log.exception( "Unable to parse result %s from post to %s" % (r.text, url)) raise return ret def post(self, path, data): url = self.url(path) self.log.debug('POST: %s' % (url,)) self.log.debug('data: %s' % (data,)) r = self.session.post( url, data=json.dumps(data).encode('utf8'), verify=self.verify_ssl, auth=self.auth, timeout=TIMEOUT, headers={'Content-Type': 'application/json;charset=UTF-8', 'User-Agent': self.user_agent}) self.iolog.debug('Received: %s %s' % (r.status_code, r.text,)) if r.status_code == 409: raise HTTPConflictException() if r.status_code == 400: raise HTTPBadRequestException('Received response 400: %s' % r.text) elif r.status_code != 200: raise Exception("Received response %s: %s" % ( r.status_code, r.text)) ret = None if r.text and len(r.text) > 4: try: ret = json.loads(r.text[4:]) except Exception: self.log.exception( "Unable to parse result %s from post to %s" % (r.text, url)) raise return ret def getProject(self, name: str) -> Project: return self.projects.get(name) def addProject(self, project: Project) -> None: self.projects[project.name] = project def getChange(self, change_key, refresh=False, event=None): if change_key.connection_name != self.connection_name: return None if change_key.change_type == 'GerritChange': return self._getChange(change_key, refresh=refresh, event=event) elif change_key.change_type == 'Tag': return self._getTag(change_key, refresh=refresh, event=event) elif change_key.change_type == 'Branch': return self._getBranch(change_key, refresh=refresh, event=event) elif change_key.change_type == 'Ref': return self._getRef(change_key, refresh=refresh, event=event) def _checkMaxDependencies(self, change, history): if change and history: history.add(history.Query.SEEN, change) if (self.max_dependencies is not None and len(history.queries[history.Query.SEEN]) > self.max_dependencies): raise GerritEventProcessingException( f"Change {change} has too many dependencies") def _getChange(self, change_key, refresh=False, history=None, network_future=None, event=None, allow_key_update=False, change=None, update_if_older_than=None): # Ensure number and patchset are str if change is None: # We may be called from the event handler which has # already gotten a change from the cache but thinks it may # need to be updated. change = self._change_cache.get(change_key) self._checkMaxDependencies(change, history) if change and not refresh: return change if not change: change = GerritChange(None) change.number = change_key.stable_id change.patchset = change_key.revision self._checkMaxDependencies(change, history) network_start = network_future is None while True: if network_future is None: network_future = ChangeNetworkFuture(update_if_older_than) network_future.add(change_key) try: self.change_network_manager.permissionToProceed(network_future) return self._updateChange(change_key, change, event, history, network_future, allow_key_update) except ChangeNetworkConflict as e: if network_start: # This is the top of the stack. We wait for the # future and try again. e.future.wait() self.change_network_manager.setComplete(network_future) # Start a new future but copy over the query cache # results from the winner. In case we decide that # the winning network did not update our change(s) # recently enough, we may still be able to benefit # from some of the previously run queries. network_future = ChangeNetworkFuture(update_if_older_than) network_future.query_results = e.future.query_results change = self._change_cache.get(change_key) or change continue else: raise finally: if network_start: self.change_network_manager.setComplete(network_future) def _getTag(self, change_key, refresh=False, event=None): tag = change_key.stable_id change = self._change_cache.get(change_key) if change: if refresh: self._change_cache.updateChangeWithRetry( change_key, change, lambda c: None) return change project = self.source.getProject(change_key.project_name) change = Tag(project) change.tag = tag change.ref = f'refs/tags/{tag}' change.oldrev = change_key.oldrev change.newrev = change_key.newrev change.url = self._getWebUrl(project, sha=change.newrev) try: self._change_cache.set(change_key, change) except ConcurrentUpdateError: change = self._change_cache.get(change_key) return change def _getBranch(self, change_key, refresh=False, event=None): branch = change_key.stable_id change = self._change_cache.get(change_key) if change: if refresh: self._change_cache.updateChangeWithRetry( change_key, change, lambda c: None) return change project = self.source.getProject(change_key.project_name) change = Branch(project) change.branch = branch change.ref = f'refs/heads/{branch}' change.oldrev = change_key.oldrev change.newrev = change_key.newrev change.url = self._getWebUrl(project, sha=change.newrev) try: self._change_cache.set(change_key, change) except ConcurrentUpdateError: change = self._change_cache.get(change_key) return change def _getRef(self, change_key, refresh=False, event=None): change = self._change_cache.get(change_key) if change: if refresh: self._change_cache.updateChangeWithRetry( change_key, change, lambda c: None) return change project = self.source.getProject(change_key.project_name) change = Ref(project) change.ref = change_key.stable_id change.oldrev = change_key.oldrev change.newrev = change_key.newrev change.url = self._getWebUrl(project, sha=change.newrev) try: self._change_cache.set(change_key, change) except ConcurrentUpdateError: change = self._change_cache.get(change_key) return change def _getDependsOnFromCommit(self, message, change, event): log = get_annotated_logger(self.log, event) records = [] seen = set() for match in self.depends_on_re.findall(message): if match in seen: log.debug("Ignoring duplicate Depends-On: %s", match) continue seen.add(match) query = "change:%s" % (match,) log.debug("Updating %s: Running query %s to find needed changes", change, query) records.extend(self.simpleQuery(query, event=event)) return [(x.number, x.current_patchset) for x in records] def _getNeededByFromCommit(self, change_id, change, event): log = get_annotated_logger(self.log, event) records = [] seen = set() query = 'message:{%s}' % change_id log.debug("Updating %s: Running query %s to find changes needed-by", change, query) results = self.simpleQuery(query, event=event) for result in results: for match in self.depends_on_re.findall( result.message): if match != change_id: continue # Note: This is not a ChangeCache ChangeKey key = (result.number, result.current_patchset) if key in seen: continue log.debug("Updating %s: Found change %s,%s " "needs %s from commit", change, key[0], key[1], change_id) seen.add(key) records.append(result) return [(x.number, x.current_patchset) for x in records] def _updateChange(self, key, change, event, history, network_future, allow_key_update=False): log = get_annotated_logger(self.log, event) # In case this change is already in the history we have a # cyclic dependency and don't need to update ourselves again # as this gets done in a previous frame of the call stack. # NOTE(jeblair): The only case where this can still be hit is # when we get an event for a change with no associated # patchset; for instance, when the gerrit topic is changed. # In that case, we will update change 1234,None, which will be # inserted into the cache as its own entry, but then we will # resolve the patchset before adding it to the history list, # then if there are dependencies, we can walk down and then # back up to the version of this change with a patchset which # will match the history list but will have bypassed the # change cache because the previous object had a patchset of # None. All paths hit the change cache first. To be able to # drop history, we need to resolve the patchset on events with # no patchsets before adding the entry to the change cache. if history and change.number and change.patchset: history_change = history.getByKey(history.Query.CHANGE, key) if history_change: log.debug("Change %s is in history", change) return history_change if (network_future.min_ltime and change.zuul_query_ltime and change.zuul_query_ltime > network_future.min_ltime): # The change was updated in another thread while we were # processing this change network. log.debug("Change %s is up to date", change) return change log.info("Updating %s", change) data = self.queryChange(change.number, event=event, min_ltime=network_future.min_ltime) # Do a local update without updating the cache so that we can # reference this change when we recurse for dependencies. change.update(data, {}, self) # Get the dependencies for this change, and recursively update # dependent changes (recursively calling this method). if not change.is_merged: extra = self._updateChangeDependencies( log, key, change, data, event, history, network_future) else: extra = {} # Actually update this change in the change cache. def _update_change(c): return c.update(data, extra, self) change = self._change_cache.updateChangeWithRetry( key, change, _update_change, allow_key_update=allow_key_update) return change def _updateChangeDependencies(self, log, key, change, data, event, history, network_future): if history is None: history = QueryHistory() history.add(history.Query.CHANGE, change) needs_changes = set() git_needs_changes = [] if data.depends_on is not None: dep_num, dep_ps = data.depends_on log.debug("Updating %s: Getting git-dependent change %s,%s", change, dep_num, dep_ps) dep_key = ChangeKey(self.connection_name, None, 'GerritChange', str(dep_num), str(dep_ps)) dep = self._getChange(dep_key, history=history, network_future=network_future, event=event) # This is a git commit dependency. So we only ignore it if it is # already merged. So even if it is "ABANDONED", we should not # ignore it. if (not dep.is_merged) and dep not in needs_changes: git_needs_changes.append(dep_key.reference) needs_changes.add(dep_key.reference) compat_needs_changes = [] for (dep_num, dep_ps) in self._getDependsOnFromCommit( data.message, change, event): log.debug("Updating %s: Getting commit-dependent " "change %s,%s", change, dep_num, dep_ps) dep_key = ChangeKey(self.connection_name, None, 'GerritChange', str(dep_num), str(dep_ps)) dep = self._getChange(dep_key, history=history, network_future=network_future, event=event) if dep.open and dep not in needs_changes: compat_needs_changes.append(dep_key.reference) needs_changes.add(dep_key.reference) needed_by_changes = set() git_needed_by_changes = [] for (dep_num, dep_ps) in data.needed_by: try: log.debug("Updating %s: Getting git-needed change %s,%s", change, dep_num, dep_ps) dep_key = ChangeKey(self.connection_name, None, 'GerritChange', str(dep_num), str(dep_ps)) dep = self._getChange(dep_key, history=history, network_future=network_future, event=event) if (dep.open and dep.is_current_patchset and dep not in needed_by_changes): git_needed_by_changes.append(dep_key.reference) needed_by_changes.add(dep_key.reference) except ChangeNetworkConflict: raise # We ignore GerritEventProcessingExceptions (e.g. when exceeding # max_dependencies) here because they are not important on the # "needed_by" side of the dependency chain. Those are already # checked for on the "needs" side. except Exception: log.exception("Failed to get git-needed change %s,%s", dep_num, dep_ps) compat_needed_by_changes = [] for (dep_num, dep_ps) in self._getNeededByFromCommit( data.id, change, event): try: log.debug("Updating %s: Getting commit-needed change %s,%s", change, dep_num, dep_ps) # Because a commit needed-by may be a cross-repo # dependency, cause that change to refresh so that it will # reference the latest patchset of its Depends-On (this # change). In case the dep is already in history we already # refreshed this change so refresh is not needed in this case. dep_key = ChangeKey(self.connection_name, None, 'GerritChange', str(dep_num), str(dep_ps)) refresh = not history.getByKey(history.Query.CHANGE, dep_key) dep = self._getChange( dep_key, refresh=refresh, history=history, network_future=network_future, event=event) if (dep.open and dep.is_current_patchset and dep not in needed_by_changes): compat_needed_by_changes.append(dep_key.reference) needed_by_changes.add(dep_key.reference) except ChangeNetworkConflict: raise # We ignore GerritEventProcessingExceptions (e.g. when exceeding # max_dependencies) here because they are not important on the # "needed_by" side of the dependency chain. Those are already # checked for on the "needs" side. except Exception: log.exception("Failed to get commit-needed change %s,%s", dep_num, dep_ps) return dict( git_needs_changes=git_needs_changes, compat_needs_changes=compat_needs_changes, git_needed_by_changes=git_needed_by_changes, compat_needed_by_changes=compat_needed_by_changes, ) def isMerged(self, change, head=None): self.log.debug("Checking if change %s is merged" % change) if not change.number: self.log.debug("Change has no number; considering it merged") # Good question. It's probably ref-updated, which, ah, # means it's merged. return True data = self.queryChange(change.number) # Note: in the case of a cherry-pick there will be a new # patchset (i.e., change.patchset might be 3 while # data.current_patchset would be 4). We could add a new # change to the cache here, but that would be more expensive. # Also, the ssh-query code path with cherry-pick does not have # a way of determining the current patchset for dependencies, # so in that case, it is useful for any changes still pointing # at this old version to think it is merged. key = ChangeKey(self.connection_name, None, 'GerritChange', str(change.number), str(change.patchset)) def _update_change(c): c.update(data, {}, self) self._change_cache.updateChangeWithRetry(key, change, _update_change) if change.is_merged: self.log.debug("Change %s is merged" % (change,)) else: self.log.debug("Change %s is not merged" % (change,)) if not head: return change.is_merged if not change.is_merged: return False ref = 'refs/heads/' + change.branch self.log.debug("Waiting for %s to appear in git repo" % (change)) if not hasattr(change, '_ref_sha'): self.log.error("Unable to confirm change %s in git repo: " "the change has not been reported; " "this pipeline may be misconfigured " "(check for multiple Gerrit connections)." % (change,)) return False if self._waitForRefSha(change.project, ref, change._ref_sha): self.log.debug("Change %s is in the git repo" % (change)) return True self.log.debug("Change %s did not appear in the git repo" % (change)) return False def _waitForRefSha(self, project: Project, ref: str, old_sha: str='') -> bool: # Wait for the ref to show up in the repo start = time.time() while time.time() - start < self.is_merged_replication_timeout: sha = self.getRefSha(project, ref) if old_sha != sha: return True time.sleep(self.replication_retry_interval) return False def getRefSha(self, project: Project, ref: str) -> str: refs = {} # type: Dict[str, str] try: refs = self.getInfoRefs(project) except Exception: self.log.exception("Exception looking for ref %s" % ref) sha = refs.get(ref, '') return sha def canMerge(self, change, allow_needs, event=None): log = get_annotated_logger(self.log, event) if not change.number: log.debug("Change has no number; considering it merged") # Good question. It's probably ref-updated, which, ah, # means it's merged. return True if change.wip: return False missing_labels = change.missing_labels - set(allow_needs) if missing_labels: self.log.debug("Unable to merge due to " "missing labels: %s", missing_labels) return False for sr in change.submit_requirements: if sr.get('status') == 'UNSATISFIED': # Otherwise, we don't care and should skip. # We're going to look at each unsatisfied submit # requirement, and if one of the involved labels is an # "allow_needs" label, we will assume that Zuul may be # able to take an action which can cause the # requirement to be satisfied, and we will ignore it. # Otherwise, it is likely a requirement that Zuul can # not alter in which case the requirement should stand # and block merging. result = sr.get("submittability_expression_result", {}) expression = result.get("expression", '') expr_contains_allow = False for allow in allow_needs: if f'label:{allow}' in expression: expr_contains_allow = True break if not expr_contains_allow: self.log.debug("Unable to merge due to " "submit requirement: %s", sr) return False return True def getProjectOpenChanges(self, project: Project) -> List[GerritChange]: # This is a best-effort function in case Gerrit is unable to return # a particular change. It happens. query = "project:{%s} status:open" % (project.name,) self.log.debug("Running query %s to get project open changes" % (query,)) data = self.simpleQuery(query) changes = [] # type: List[GerritChange] for record in data: try: change_key = ChangeKey(self.connection_name, None, 'GerritChange', str(record.number), str(record.current_patchset)) changes.append(self._getChange(change_key)) except Exception: self.log.exception("Unable to query change %s", record.number) return changes @staticmethod def _checkRefFormat(refname: str) -> bool: # These are the requirements for valid ref names as per # man git-check-ref-format parts = refname.split('/') return \ (GerritConnection.refname_bad_sequences.search(refname) is None and len(parts) > 1 and not any(part.startswith('.') or part.endswith('.lock') for part in parts)) def _getProjectBranchesRequiredFlags( self, exclude_unprotected, exclude_locked): return BranchFlag.PRESENT def _filterProjectBranches( self, branch_infos, exclude_unprotected, exclude_locked): return branch_infos def _fetchProjectBranches(self, project, required_flags): refs = self.getInfoRefs(project) heads = [str(k[len('refs/heads/'):]) for k in refs if k.startswith('refs/heads/') and GerritConnection._checkRefFormat(k)] branch_infos = [BranchInfo(h, present=True) for h in heads] self.log.debug("Fetched %s branches for %s", len(heads), project) return BranchFlag.PRESENT, branch_infos def _fetchProjectDefaultBranch(self, project): if not self.session: return 'master' head = None for attempt in range(1, 4): try: head = self.get( 'projects/%s/HEAD' % ( urllib.parse.quote(project.name, safe=''), )) if head.startswith('refs/heads/'): head = head[len('refs/heads/'):] return head except HTTPNotFoundException: self.log.exception("Unable to get HEAD for %s", project) return head except Exception as e: if attempt >= 3: self.log.exception("Unable to get HEAD for %s", project) return head self.log.warning("Unable to get HEAD for %s, will retry: %s", project, e) time.sleep(1) def isBranchProtected(self, project_name, branch_name, zuul_event_id=None): # TODO: This could potentially be expanded to do something # with user-specific branches. return True def shouldProcessEvent(self, event_type, event, replication): valid_events = event._branch_ref_update or \ event.default_branch_changed or \ event.change_number if replication: valid_events = valid_events or \ event_type == "ref-replication-scheduled" or \ event_type == "ref-replicated" if not (valid_events): # Check if the events match specific filters and are # valid. for event_filter in self.watched_event_filters: r = event_filter.preFilter(event) if r: break else: return False return True def addEvent(self, data): # NOTE(mnaser): Certain plugins fire events which end up causing # an unrecognized event log *and* a traceback if they # do not contain full project information, we skip them # here to keep logs clean. event_type = data.get('type') if event_type in GerritEventConnector.IGNORED_EVENTS: return # Due to notedb, an high percentage of all events Zuul # processes are ref-updated of the /meta ref, and that is # unlikely to be used in Zuul. Skip those here so that we # reduce traffic on the event queue. if event_type == 'ref-updated': refname = data.get('refUpdate', {}).get('refName', '') if (refname.startswith('refs/changes/') and refname.endswith('/meta')): return # Partially realize a GerritTriggerEvent with enough # information to determine branches/refs/etc. event = GerritTriggerEvent.fromGerritEventDict( data, None, self, None) # If the event communicates replication info or might trigger # reconfiguration actions we keep it; otherwise check to see if # it matches our pre-filters: if not self.shouldProcessEvent(event_type, event, True): self.log.debug("Event did not match pre-filters %s", event) return event_uuid = uuid4().hex attributes = { "zuul_event_id": event_uuid, } # Gerrit events don't have an event id that could be used to globally # identify this event in the system so we have to generate one. with self.tracer.start_span( "GerritEvent", attributes=attributes) as span: event = { "timestamp": time.time(), "zuul_event_id": event_uuid, "span_context": tracing.getSpanContext(span), "payload": data, } self.event_queue.put(event) def review(self, item, change, message, submit, labels, checks_api, notify, file_comments, phase1, phase2, zuul_event_id=None): if self.session: meth = self.review_http else: meth = self.review_ssh return meth(item, change, message, submit, labels, checks_api, notify, file_comments, phase1, phase2, zuul_event_id=zuul_event_id) def review_ssh(self, item, change, message, submit, labels, checks_api, notify, file_comments, phase1, phase2, zuul_event_id=None): log = get_annotated_logger(self.log, zuul_event_id) if checks_api: log.error("Zuul is configured to report to the checks API, " "but no HTTP password is present for the connection " "in the configuration file.") project = change.project.name cmd = 'gerrit review --project %s' % project if notify: cmd += ' --notify %s' % shlex.quote(notify) if phase1: if message: b_len = len(message.encode('utf-8')) if b_len >= GERRIT_HUMAN_MESSAGE_LIMIT: log.info("Message truncated %d > %d" % (b_len, GERRIT_HUMAN_MESSAGE_LIMIT)) message = ("%s... (truncated)" % message[:GERRIT_HUMAN_MESSAGE_LIMIT - 20]) cmd += ' --message %s' % shlex.quote(message) for key, val in labels.items(): if val is True: cmd += ' --%s' % key else: cmd += ' --label %s=%s' % (key, val) if self.version >= (2, 13, 0): cmd += ' --tag autogenerated:zuul:%s' % ( item.manager.pipeline.name) if phase2 and submit: cmd += ' --submit' changeid = '%s,%s' % (change.number, change.patchset) cmd += ' %s' % changeid out, err = self._ssh(cmd, zuul_event_id=zuul_event_id) return err def report_checks(self, log, item, change, changeid, checkinfo): checkinfo = checkinfo.copy() uuid = checkinfo.pop('uuid', None) scheme = checkinfo.pop('scheme', None) if uuid is None: uuids = self.project_checker_map.get( change.project.canonical_name, set()) for u in uuids: if u.split(':')[0] == scheme: uuid = u break if uuid is None: log.error("Unable to find matching checker for %s %s", item, checkinfo) return def fmt(t): return str(datetime.datetime.fromtimestamp(t)) if item.enqueue_time: checkinfo['started'] = fmt(item.enqueue_time) if item.report_time: checkinfo['finished'] = fmt(item.report_time) url = item.formatItemUrl() if url: checkinfo['url'] = url if checkinfo: for x in range(1, 4): try: self.post('changes/%s/revisions/%s/checks/%s' % (changeid, change.commit, uuid), checkinfo) break except HTTPConflictException: log.exception("Conflict submitting check data to gerrit.") break except HTTPBadRequestException: log.exception( "Bad request submitting check data to gerrit.") break except Exception as e: log.exception("Error submitting check data to gerrit on " "attempt %s: %s", x, e) time.sleep(x * self.submit_retry_backoff) def review_http(self, item, change, message, submit, labels, checks_api, notify, file_comments, phase1, phase2, zuul_event_id=None): changeid = "%s~%s~%s" % ( urllib.parse.quote(str(change.project), safe=''), urllib.parse.quote(str(change.branch), safe=''), change.id) log = get_annotated_logger(self.log, zuul_event_id) b_len = len(message.encode('utf-8')) if b_len >= GERRIT_HUMAN_MESSAGE_LIMIT: log.info("Message truncated %d > %d" % (b_len, GERRIT_HUMAN_MESSAGE_LIMIT)) message = ("%s... (truncated)" % message[:GERRIT_HUMAN_MESSAGE_LIMIT - 20]) data = dict(strict_labels=False) if notify: data['notify'] = notify if phase1: data['message'] = message if change.is_current_patchset: if labels: data['labels'] = labels if file_comments: if self.version >= (2, 15, 0): file_comments = copy.deepcopy(file_comments) url = item.formatItemUrl() for comments in itertools.chain( file_comments.values()): for comment in comments: comment['robot_id'] = 'zuul' comment['robot_run_id'] = \ item.current_build_set.uuid if url: comment['url'] = url data['robot_comments'] = file_comments else: data['comments'] = file_comments if self.version >= (2, 13, 0): data['tag'] = 'autogenerated:zuul:%s' % ( item.manager.pipeline.name) if checks_api: self.report_checks(log, item, change, changeid, checks_api) if (message or data.get('labels') or data.get('comments') or data.get('robot_comments')): for x in range(1, 4): try: self.post('changes/%s/revisions/%s/review' % (changeid, change.commit), data) break except HTTPConflictException: log.exception("Conflict submitting data to gerrit.") break except HTTPBadRequestException: log.exception( "Bad request submitting check data to gerrit.") break except Exception as e: log.exception( "Error submitting data to gerrit " "on attempt %s: %s", x, e) time.sleep(x * self.submit_retry_backoff) if phase2 and change.is_current_patchset and submit: for x in range(1, 4): try: self.post('changes/%s/submit' % (changeid,), {}) break except HTTPConflictException: log.info("Conflict submitting data to gerrit, " "change may already be merged") break except HTTPBadRequestException: log.exception( "Bad request submitting check data to gerrit.") break except Exception as e: log.exception( "Error submitting data to gerrit on attempt %s: %s", x, e) time.sleep(x * self.submit_retry_backoff) def queryChangeSSH(self, number, event=None): args = '--all-approvals --comments --commit-message' args += ' --current-patch-set --dependencies --files' args += ' --patch-sets --submit-records' cmd = 'gerrit query --format json %s change:%s' % (args, number) out, err = self._ssh(cmd) if not out: return False lines = out.split('\n') if not lines: return False data = json.loads(lines[0]) if not data: return False iolog = get_annotated_logger(self.iolog, event) iolog.debug("Received data from Gerrit query: \n%s", pprint.pformat(data)) return data def queryChangeHTTP(self, number, event=None): query = ('changes/%s?o=DETAILED_ACCOUNTS&o=CURRENT_REVISION&' 'o=CURRENT_COMMIT&o=CURRENT_FILES&o=LABELS&' 'o=DETAILED_LABELS&o=ALL_REVISIONS' % (number,)) if self.version >= (3, 5, 0): query += '&o=SUBMIT_REQUIREMENTS' data = self.get(query) related = self.get('changes/%s/revisions/%s/related' % ( number, data['current_revision'])) files_query = 'changes/%s/revisions/%s/files' % ( number, data['current_revision']) commentable_files_query = None # We query with parent=1 to get the list of files compared to # the merge target, unless this is the initial commit to the # repository (there are no parents). if data['revisions'][data['current_revision']]['commit']['parents']: files_query += '?parent=1' if len(data['revisions'][data['current_revision']][ 'commit']['parents']) > 1: # If there is more than one parent, we also query the # list of files on this specific commit so we know # upon which we can leave comments. commentable_files_query = 'changes/%s/revisions/%s/files' % ( number, data['current_revision']) if commentable_files_query: commentable_files = self.get(commentable_files_query) else: commentable_files = None files = self.get(files_query) return data, related, files, commentable_files def queryChange(self, number, event=None, min_ltime=None): lock = self.change_network_manager.getQueryLock(number) with lock: if min_ltime: data = self.change_network_manager.getQueryResult(number) if data and data.zuul_query_ltime > min_ltime: return data for attempt in range(1, 4): # Get a query ltime -- any events before this point should be # included in our change data. zuul_query_ltime = self.sched.zk_client.getCurrentLtime() try: if self.session: data, related, files, commentable_files =\ self.queryChangeHTTP( number, event=event) ret = GerritChangeData( GerritChangeData.HTTP, data, related, files, commentable_files, zuul_query_ltime=zuul_query_ltime) else: data = self.queryChangeSSH(number, event=event) ret = GerritChangeData( GerritChangeData.SSH, data, zuul_query_ltime=zuul_query_ltime) self.change_network_manager.updateQueryResult(number, ret) return ret except HTTPNotFoundException as e: # do not retry on 404 results, instead skip further event # processing since we can't load a corresponding change raise GerritEventProcessingException( f"Did not find change for number {number}") from e except Exception as e: if attempt >= 3: raise # The internet is a flaky place try again. self.log.warning( "Error querying change %s, will retry: %s", number, e) time.sleep(1) def simpleQuerySSH(self, query, event=None): def _query_chunk(query, event): args = '--commit-message --current-patch-set' cmd = 'gerrit query --format json %s %s' % ( args, query) out, err = self._ssh(cmd) if not out: return False lines = out.split('\n') if not lines: return False # filter out blank lines data = [json.loads(line) for line in lines if line.startswith('{')] # check last entry for more changes more_changes = None if 'moreChanges' in data[-1]: more_changes = data[-1]['moreChanges'] # we have to remove the statistics line del data[-1] if not data: return False, more_changes iolog = get_annotated_logger(self.iolog, event) iolog.debug("Received data from Gerrit query: \n%s", pprint.pformat(data)) return data, more_changes # gerrit returns 500 results by default, so implement paging # for large projects like nova alldata = [] chunk, more_changes = _query_chunk(query, event) while chunk: alldata.extend(chunk) if more_changes is None: # continue sortKey based (before Gerrit 2.9) resume = "resume_sortkey:'%s'" % chunk[-1]["sortKey"] elif more_changes: # continue moreChanges based (since Gerrit 2.9) resume = "-S %d" % len(alldata) else: # no more changes break chunk, more_changes = _query_chunk( "%s %s" % (query, resume), event) return alldata def simpleQueryHTTP(self, query, event=None): iolog = get_annotated_logger(self.iolog, event) changes = [] sortkey = '' done = False offset = 0 query = urllib.parse.quote(query, safe='') while not done: # We don't actually want to limit to 500, but that's the # server-side default, and if we don't specify this, we # won't get a _more_changes flag. q = ('changes/?n=500%s&o=CURRENT_REVISION&o=CURRENT_COMMIT&' 'q=%s' % (sortkey, query)) iolog.debug('Query: %s', q) batch = self.get(q) iolog.debug("Received data from Gerrit query: \n%s", pprint.pformat(batch)) done = True if batch: changes += batch if '_more_changes' in batch[-1]: done = False if '_sortkey' in batch[-1]: sortkey = '&N=%s' % (batch[-1]['_sortkey'],) else: offset += len(batch) sortkey = '&start=%s' % (offset,) return changes def simpleQuery(self, query, event=None): if self.session: # None of the users of this method require dependency # data, so we only perform the change query and omit the # related changes query. alldata = self.simpleQueryHTTP(query, event=event) return [GerritChangeData(GerritChangeData.HTTP, data) for data in alldata] else: alldata = self.simpleQuerySSH(query, event=event) return [GerritChangeData(GerritChangeData.SSH, data) for data in alldata] def _uploadPack(self, project): if self.session and not self.git_over_ssh: url = ('%s/%s/info/refs?service=git-upload-pack' % (self.baseurl, project.name)) for attempt in range(1, 4): try: r = self.session.get( url, verify=self.verify_ssl, auth=self.auth, timeout=TIMEOUT, headers={'User-Agent': self.user_agent}) self.iolog.debug('Received: %s %s', r.status_code, r.text) if r.status_code == 409: raise HTTPConflictException() if r.status_code == 404: raise HTTPNotFoundException() elif r.status_code != 200: raise Exception("Received response %s" % ( r.status_code,)) except HTTPNotFoundException: raise except Exception as e: if attempt >= 3: self.log.exception( "Error getting refs for %s:", project) raise self.log.warning( "Error getting refs for %s, will retry: %s", project, e) time.sleep(1) out = r.text[r.text.find('\n') + 5:] else: cmd = "git-upload-pack %s" % project.name out, err = self._ssh(cmd, "0000") return out def _open(self): if self.client: # Paramiko needs explicit closes, its possible we will open even # with an unclosed client so explicitly close here. self.client.close() try: client = paramiko.SSHClient() client.load_system_host_keys() client.set_missing_host_key_policy(paramiko.WarningPolicy()) # SSH banner, handshake, and auth timeouts default to 15 # seconds, so we only set the socket timeout here. client.connect(self.ssh_server, username=self.user, port=self.port, key_filename=self.keyfile, timeout=self.ssh_timeout) transport = client.get_transport() transport.set_keepalive(self.keepalive) self.client = client except Exception: client.close() self.client = None raise def _ssh(self, command, stdin_data=None, zuul_event_id=None): log = get_annotated_logger(self.log, zuul_event_id) if not self.client: self._open() max_attempts = 2 for x in range(max_attempts): try: log.debug("SSH command:\n%s", command) stdin, stdout, stderr = self.client.exec_command( command, timeout=TIMEOUT) break except Exception: self._open() if x + 1 >= max_attempts: raise if stdin_data: stdin.write(stdin_data) out = stdout.read().decode('utf-8') self.iolog.debug("SSH received stdout:\n%s" % out) ret = stdout.channel.recv_exit_status() log.debug("SSH exit status: %s", ret) err = stderr.read().decode('utf-8') if err.strip(): log.debug("SSH received stderr:\n%s", err) if ret: log.debug("SSH received stdout:\n%s", out) raise Exception("Gerrit error executing %s" % command) return (out, err) def getInfoRefs(self, project: Project) -> Dict[str, str]: try: # Encode the UTF-8 data back to a byte array, as the size of # each record in the pack is in bytes, and so the slicing must # also be done on a byte-basis. data = self._uploadPack(project).encode("utf-8") except Exception: self.log.error("Cannot get references from %s" % project) raise # keeps error information ret = {} read_advertisement = False i = 0 while i < len(data): if len(data) - i < 4: raise Exception("Invalid length in info/refs") plen = int(data[i:i + 4], 16) i += 4 # It's the length of the packet, including the 4 bytes of the # length itself, unless it's null, in which case the length is # not included. if plen > 0: plen -= 4 if len(data) - i < plen: raise Exception("Invalid data in info/refs") # Once the pack data is sliced, we can safely decode it back # into a (UTF-8) string. line = data[i:i + plen].decode("utf-8") i += plen if not read_advertisement: read_advertisement = True continue if plen == 0: # The terminating null continue line = line.strip() revision, ref = line.split() ret[ref] = revision return ret def getGitUrl(self, project: Project) -> str: if self.anonymous_git: url = ('%s/%s' % (self.baseurl, project.name)) elif self.session and not self.git_over_ssh: baseurl = list(urllib.parse.urlparse(self.baseurl)) # Make sure we escape '/' symbols, otherwise git's url # parser will think the username is a hostname. baseurl[1] = '%s:%s@%s' % ( urllib.parse.quote(self.user, safe=''), urllib.parse.quote(self.password, safe=''), baseurl[1]) baseurl = urllib.parse.urlunparse(baseurl) url = ('%s/a/%s' % (baseurl, project.name)) else: url = 'ssh://%s@%s:%s/%s' % (self.user, self.ssh_server, self.port, project.name) return url def _getWebUrl(self, project: Project, sha: str=None) -> str: return self.gitweb_url_template.format( baseurl=self.baseurl, project=project.getSafeAttributes(), sha=sha) def _getRemoteVersion(self): version = self.get('config/server/version') base = version.split('-')[0] parts = base.split('.') major = minor = micro = 0 if len(parts) > 0: major = int(parts[0]) if len(parts) > 1: minor = int(parts[1]) if len(parts) > 2: micro = int(parts[2]) self.version = (major, minor, micro) self.log.info("Remote version is: %s (parsed as %s)" % (version, self.version)) def _getRemoteInfo(self): info = self.get('config/server/info') change_info = info.get('change', {}) self.submit_whole_topic = change_info.get('submit_whole_topic', False) self.log.info("Remote submitWholeTopic: %s", self.submit_whole_topic) def refWatcherCallback(self, data): event = { 'type': 'ref-updated', 'refUpdate': { 'project': data['project'], 'refName': data['ref'], 'oldRev': data['oldrev'], 'newRev': data['newrev'], } } self.addEvent(event) def onLoad(self, zk_client, component_registry): self.log.debug("Starting Gerrit Connection/Watchers") if self.session: try: self._getRemoteVersion() except Exception: self.log.exception("Unable to determine remote Gerrit version") try: self._getRemoteInfo() except Exception: self.log.exception("Unable to fetch remote Gerrit info") # Set the project branch cache to read only if no scheduler is # provided to prevent fetching the branches from the connection. self.read_only = not self.sched self.log.debug('Creating Zookeeper branch cache') self._branch_cache = BranchCache(zk_client, self, component_registry) self.log.info("Creating Zookeeper event queue") if self.sched: component_info = self.sched.component_info else: component_info = None self.event_queue = ConnectionEventQueue( zk_client, self.connection_name, component_info) # If the connection was not loaded by a scheduler, but by e.g. # zuul-web, we want to stop here. if not self.sched: return self.log.debug('Creating Zookeeper change cache') self._change_cache = GerritChangeCache(zk_client, self) self.startEventSourceThread() # TODO: This is only for the checks plugin and can be removed # when checks support is removed. Until then, we always start # this thread, but if no checks are configured, it remains # idle. self.startPollerThread() self.startEventConnector() def onStop(self): self.log.debug("Stopping Gerrit Connection/Watchers") self.stopEventSourceThread() self.stopPollerThread() self.stopRefWatcherThread() self.stopEventConnector() if self._change_cache: self._change_cache.stop() def getEventQueue(self): return getattr(self, "event_queue", None) def stopEventSourceThread(self): if self.event_thread: self.event_thread.stop() self.event_thread.join() def startEventSourceThread(self): if self.event_source == self.EVENT_SOURCE_STREAM_EVENTS: self.startSSHListener() elif self.event_source == self.EVENT_SOURCE_KAFKA: self.startKafkaListener() elif self.event_source == self.EVENT_SOURCE_KINESIS: self.startAWSKinesisListener() elif self.event_source == self.EVENT_SOURCE_GCLOUD_PUBSUB: self.startGcloudPubsubListener() else: self.log.warning("No gerrit event source configured") self.startRefWatcherThread() if self.event_thread: self.event_thread.start() def startSSHListener(self): self.log.info("Starting SSH event stream client") self.event_thread = GerritSSHEventListener( self, self.connection_config) def startKafkaListener(self): self.log.info("Starting Kafka consumer") self.event_thread = GerritKafkaEventListener( self, self.connection_config) def startAWSKinesisListener(self): self.log.info("Starting AWS Kinesis consumer") self.event_thread = GerritAWSKinesisEventListener( self, self.connection_config) def startGcloudPubsubListener(self): self.log.info("Starting gcloud pubsub consumer") self.event_thread = GerritGcloudPubsubEventListener( self, self.connection_config) def startPollerThread(self): if self.session is not None: self.poller_thread = self._poller_class(self) self.poller_thread.start() else: self.log.info( "%s: Gerrit Poller is disabled because no " "HTTP authentication is defined", self.connection_name) def stopPollerThread(self): if self.poller_thread: self.poller_thread.stop() self.poller_thread.join() def stopRefWatcherThread(self): if self.ref_watcher_thread: self.ref_watcher_thread.stop() self.ref_watcher_thread.join() def startRefWatcherThread(self): self.ref_watcher_thread = self._ref_watcher_class( self, self.baseurl, self.ref_watcher_poll_interval, self.refWatcherCallback, election_name="ref-watcher") self.ref_watcher_thread.start() def startEventConnector(self): self.gerrit_event_connector = GerritEventConnector( self, self.replication_timeout) self.gerrit_event_connector.start() def stopEventConnector(self): if self.gerrit_event_connector: self.gerrit_event_connector.stop()