The Gatekeeper, or a project gating system
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2501 lines
101 KiB

# 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
# 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
from contextlib import contextmanager
import copy
import itertools
import os
import logging
import textwrap
import io
import re
import subprocess
import voluptuous as vs
from zuul.driver.sql.sqlconnection import SQLConnection
from zuul import model
from zuul.lib import yamlutil as yaml
import zuul.manager.dependent
import zuul.manager.independent
import zuul.manager.supercedent
import zuul.manager.serial
from zuul.lib.logutil import get_annotated_logger
from zuul.lib.re2util import filter_allowed_disallowed
from zuul.lib.varnames import check_varnames
from zuul.zk.semaphore import SemaphoreHandler
# Several forms accept either a single item or a list, this makes
# specifying that in the schema easy (and explicit).
def to_list(x):
return vs.Any([x], x)
def as_list(item):
if not item:
return []
if isinstance(item, list):
return item
return [item]
class ConfigurationSyntaxError(Exception):
class NodeFromGroupNotFoundError(Exception):
def __init__(self, nodeset, node, group):
message = textwrap.dedent("""\
In nodeset "{nodeset}" the group "{group}" contains a
node named "{node}" which is not defined in the nodeset.""")
message = textwrap.fill(message.format(nodeset=nodeset,
node=node, group=group))
super(NodeFromGroupNotFoundError, self).__init__(message)
class DuplicateNodeError(Exception):
def __init__(self, nodeset, node):
message = textwrap.dedent("""\
In nodeset "{nodeset}" the node "{node}" appears multiple times.
Node names must be unique within a nodeset.""")
message = textwrap.fill(message.format(nodeset=nodeset,
super(DuplicateNodeError, self).__init__(message)
class UnknownConnection(Exception):
def __init__(self, connection_name):
message = textwrap.dedent("""\
Unknown connection named "{connection}".""")
message = textwrap.fill(message.format(connection=connection_name))
super(UnknownConnection, self).__init__(message)
class LabelForbiddenError(Exception):
def __init__(self, label, allowed_labels, disallowed_labels):
message = textwrap.dedent("""\
Label named "{label}" is not part of the allowed
labels ({allowed_labels}) for this tenant.""")
# Make a string that looks like "a, b and not c, d" if we have
# both allowed and disallowed labels.
labels = ", ".join(allowed_labels or [])
if allowed_labels and disallowed_labels:
labels += ' and '
if disallowed_labels:
labels += 'not '
labels += ", ".join(disallowed_labels)
message = textwrap.fill(message.format(
super(LabelForbiddenError, self).__init__(message)
class MaxTimeoutError(Exception):
def __init__(self, job, tenant):
message = textwrap.dedent("""\
The job "{job}" exceeds tenant max-job-timeout {maxtimeout}.""")
message = textwrap.fill(message.format(, maxtimeout=tenant.max_job_timeout))
super(MaxTimeoutError, self).__init__(message)
class DuplicateGroupError(Exception):
def __init__(self, nodeset, group):
message = textwrap.dedent("""\
In nodeset "{nodeset}" the group "{group}" appears multiple times.
Group names must be unique within a nodeset.""")
message = textwrap.fill(message.format(nodeset=nodeset,
super(DuplicateGroupError, self).__init__(message)
class ProjectNotFoundError(Exception):
def __init__(self, project):
message = textwrap.dedent("""\
The project "{project}" was not found. All projects
referenced within a Zuul configuration must first be
added to the main configuration file by the Zuul
message = textwrap.fill(message.format(project=project))
super(ProjectNotFoundError, self).__init__(message)
class TemplateNotFoundError(Exception):
def __init__(self, template):
message = textwrap.dedent("""\
The project template "{template}" was not found.
message = textwrap.fill(message.format(template=template))
super(TemplateNotFoundError, self).__init__(message)
class NodesetNotFoundError(Exception):
def __init__(self, nodeset):
message = textwrap.dedent("""\
The nodeset "{nodeset}" was not found.
message = textwrap.fill(message.format(nodeset=nodeset))
super(NodesetNotFoundError, self).__init__(message)
class PipelineNotPermittedError(Exception):
def __init__(self):
message = textwrap.dedent("""\
Pipelines may not be defined in untrusted repos,
they may only be defined in config repos.""")
message = textwrap.fill(message)
super(PipelineNotPermittedError, self).__init__(message)
class ProjectNotPermittedError(Exception):
def __init__(self):
message = textwrap.dedent("""\
Within an untrusted project, the only project definition
permitted is that of the project itself.""")
message = textwrap.fill(message)
super(ProjectNotPermittedError, self).__init__(message)
class YAMLDuplicateKeyError(ConfigurationSyntaxError):
def __init__(self, key, node, context, start_mark):
intro = textwrap.fill(textwrap.dedent("""\
Zuul encountered a syntax error while parsing its configuration in the
repo {repo} on branch {branch}. The error was:""".format(,
e = textwrap.fill(textwrap.dedent("""\
The key "{key}" appears more than once; duplicate keys are not
m = textwrap.dedent("""\
The error appears in the following stanza:
m = m.format(intro=intro,
super(YAMLDuplicateKeyError, self).__init__(m)
def indent(s):
return '\n'.join([' ' + x for x in s.split('\n')])
def project_configuration_exceptions(context, accumulator):
except ConfigurationSyntaxError:
except Exception as e:
intro = textwrap.fill(textwrap.dedent("""\
Zuul encountered an error while accessing the repo {repo}. The error
m = textwrap.dedent("""\
m = m.format(intro=intro,
accumulator.addError(context, None, m)
def early_configuration_exceptions(context):
except ConfigurationSyntaxError:
except Exception as e:
intro = textwrap.fill(textwrap.dedent("""\
Zuul encountered a syntax error while parsing its configuration in the
repo {repo} on branch {branch}. The error was:""".format(,
m = textwrap.dedent("""\
m = m.format(intro=intro,
raise ConfigurationSyntaxError(m)
def configuration_exceptions(stanza, conf, accumulator):
except ConfigurationSyntaxError:
except Exception as e:
conf = copy.deepcopy(conf)
context = conf.pop('_source_context')
start_mark = conf.pop('_start_mark')
intro = textwrap.fill(textwrap.dedent("""\
Zuul encountered a syntax error while parsing its configuration in the
repo {repo} on branch {branch}. The error was:""".format(,
m = textwrap.dedent("""\
The error appears in the following {stanza} stanza:
m = m.format(intro=intro,
accumulator.addError(context, start_mark, m, str(e))
def reference_exceptions(stanza, obj, accumulator):
except ConfigurationSyntaxError:
except Exception as e:
context = obj.source_context
start_mark = obj.start_mark
intro = textwrap.fill(textwrap.dedent("""\
Zuul encountered a syntax error while parsing its configuration in the
repo {repo} on branch {branch}. The error was:""".format(,
m = textwrap.dedent("""\
The error appears in the following {stanza} stanza:
m = m.format(intro=intro,
accumulator.addError(context, start_mark, m, str(e))
class ZuulMark(object):
# The yaml mark class differs between the C and python versions.
# The C version does not provide a snippet, and also appears to
# lose data under some circumstances.
def __init__(self, start_mark, end_mark, stream): =
self.index = start_mark.index
self.line = start_mark.line
self.end_line = end_mark.line
self.end_index = end_mark.index
self.column = start_mark.column
self.end_column = end_mark.column
self.snippet = stream[start_mark.index:end_mark.index]
def __str__(self):
return ' in "{name}", line {line}, column {column}'.format(,
line=self.line + 1,
column=self.column + 1,
def __eq__(self, other):
return (self.line == other.line and
self.snippet == other.snippet)
class ZuulSafeLoader(yaml.EncryptedLoader):
zuul_node_types = frozenset(('job', 'nodeset', 'secret', 'pipeline',
'project', 'project-template',
'semaphore', 'queue', 'pragma'))
def __init__(self, stream, context):
wrapped_stream = io.StringIO(stream) = str(context)
super(ZuulSafeLoader, self).__init__(wrapped_stream) = str(context)
self.zuul_context = context
self.zuul_stream = stream
def construct_mapping(self, node, deep=False):
keys = set()
for k, v in node.value:
# The key << needs to be treated special since that will merge
# the anchor into the mapping and not create a key on its own.
if k.value == '<<':
if k.value in keys:
mark = ZuulMark(node.start_mark, node.end_mark,
raise YAMLDuplicateKeyError(k.value, node, self.zuul_context,
r = super(ZuulSafeLoader, self).construct_mapping(node, deep)
keys = frozenset(r.keys())
if len(keys) == 1 and keys.intersection(self.zuul_node_types):
d = list(r.values())[0]
if isinstance(d, dict):
d['_start_mark'] = ZuulMark(node.start_mark, node.end_mark,
d['_source_context'] = self.zuul_context
return r
def safe_load_yaml(stream, context):
loader = ZuulSafeLoader(stream, context)
return loader.get_single_data()
except yaml.YAMLError as e:
m = """
Zuul encountered a syntax error while parsing its configuration in the
repo {repo} on branch {branch}. The error was:
m = m.format(,
raise ConfigurationSyntaxError(m)
def ansible_var_name(value):
if not re.fullmatch(r"[a-zA-Z][a-zA-Z0-9_]*", value):
raise vs.Invalid("Invalid Ansible variable name '{}'".format(value))
def ansible_vars_dict(value):
for key in value:
class PragmaParser(object):
pragma = {
'implied-branch-matchers': bool,
'implied-branches': to_list(str),
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
schema = vs.Schema(pragma)
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.PragmaParser")
self.pcontext = pcontext
def fromYaml(self, conf):
bm = conf.get('implied-branch-matchers')
source_context = conf['_source_context']
if bm is not None:
source_context.implied_branch_matchers = bm
branches = conf.get('implied-branches')
if branches is not None:
source_context.implied_branches = as_list(branches)
class NodeSetParser(object):
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.NodeSetParser")
self.pcontext = pcontext
self.schema = self.getSchema(False)
self.anon_schema = self.getSchema(True)
def getSchema(self, anonymous=False):
node = {vs.Required('name'): to_list(str),
vs.Required('label'): str,
group = {vs.Required('name'): str,
vs.Required('nodes'): to_list(str),
nodeset = {vs.Required('nodes'): to_list(node),
'groups': to_list(group),
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
if not anonymous:
nodeset[vs.Required('name')] = str
return vs.Schema(nodeset)
def fromYaml(self, conf, anonymous=False):
if anonymous:
ns = model.NodeSet(conf.get('name'))
ns.source_context = conf.get('_source_context')
ns.start_mark = conf.get('_start_mark')
node_names = set()
group_names = set()
allowed_labels = self.pcontext.tenant.allowed_labels
disallowed_labels = self.pcontext.tenant.disallowed_labels
requested_labels = [n['label'] for n in as_list(conf['nodes'])]
filtered_labels = filter_allowed_disallowed(
requested_labels, allowed_labels, disallowed_labels)
rejected_labels = set(requested_labels) - set(filtered_labels)
for name in rejected_labels:
raise LabelForbiddenError(
for conf_node in as_list(conf['nodes']):
if "localhost" in as_list(conf_node['name']):
raise Exception("Nodes named 'localhost' are not allowed.")
for name in as_list(conf_node['name']):
if name in node_names:
raise DuplicateNodeError(name, conf_node['name'])
node = model.Node(as_list(conf_node['name']), conf_node['label'])
for name in as_list(conf_node['name']):
for conf_group in as_list(conf.get('groups', [])):
if "localhost" in conf_group['name']:
raise Exception("Groups named 'localhost' are not allowed.")
for node_name in as_list(conf_group['nodes']):
if node_name not in node_names:
raise NodeFromGroupNotFoundError(conf['name'], node_name,
if conf_group['name'] in group_names:
raise DuplicateGroupError(conf['name'], conf_group['name'])
group = model.Group(conf_group['name'],
return ns
class SecretParser(object):
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.SecretParser")
self.pcontext = pcontext
self.schema = self.getSchema()
def getSchema(self):
secret = {vs.Required('name'): str,
vs.Required('data'): dict,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
return vs.Schema(secret)
def fromYaml(self, conf):
s = model.Secret(conf['name'], conf['_source_context'])
s.source_context = conf['_source_context']
s.start_mark = conf['_start_mark']
s.secret_data = conf['data']
return s
class JobParser(object):
ANSIBLE_ROLE_RE = re.compile(r'^(ansible[-_.+]*)*(role[-_.+]*)*')
zuul_role = {vs.Required('zuul'): str,
'name': str}
galaxy_role = {vs.Required('galaxy'): str,
'name': str}
role = vs.Any(zuul_role, galaxy_role)
job_project = {vs.Required('name'): str,
'override-branch': str,
'override-checkout': str}
job_dependency = {vs.Required('name'): str,
'soft': bool}
secret = {vs.Required('name'): ansible_var_name,
vs.Required('secret'): str,
'pass-to-parent': bool}
semaphore = {vs.Required('name'): str,
'resources-first': bool}
# Attributes of a job that can also be used in Project and ProjectTemplate
job_attributes = {'parent': vs.Any(str, None),
'final': bool,
'abstract': bool,
'protected': bool,
'intermediate': bool,
'requires': to_list(str),
'provides': to_list(str),
'failure-message': str,
'success-message': str,
# TODO: ignored, remove for v5
'failure-url': str,
# TODO: ignored, remove for v5
'success-url': str,
'hold-following-changes': bool,
'voting': bool,
'semaphore': vs.Any(semaphore, str),
'semaphores': to_list(vs.Any(semaphore, str)),
'tags': to_list(str),
'branches': to_list(str),
'files': to_list(str),
'secrets': to_list(vs.Any(secret, str)),
'irrelevant-files': to_list(str),
# validation happens in NodeSetParser
'nodeset': vs.Any(dict, str),
'timeout': int,
'post-timeout': int,
'attempts': int,
'pre-run': to_list(str),
'post-run': to_list(str),
'run': to_list(str),
'cleanup-run': to_list(str),
'ansible-version': vs.Any(str, float),
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
'roles': to_list(role),
'required-projects': to_list(vs.Any(job_project, str)),
'vars': ansible_vars_dict,
'extra-vars': ansible_vars_dict,
'host-vars': {str: ansible_vars_dict},
'group-vars': {str: ansible_vars_dict},
'dependencies': to_list(vs.Any(job_dependency, str)),
'allowed-projects': to_list(str),
'override-branch': str,
'override-checkout': str,
'description': str,
'variant-description': str,
'post-review': bool,
'match-on-config-updates': bool,
'workspace-scheme': vs.Any('golang', 'flat'),
job_name = {vs.Required('name'): str}
job = dict(collections.ChainMap(job_name, job_attributes))
schema = vs.Schema(job)
simple_attributes = [
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.JobParser")
self.pcontext = pcontext
def fromYaml(self, conf, project_pipeline=False, name=None,
if validate:
if name is None:
name = conf['name']
# NB: The default detection system in the Job class requires
# that we always assign values directly rather than modifying
# them (e.g., " = ..." rather than
# "").
job = model.Job(name)
job.description = conf.get('description')
job.source_context = conf['_source_context']
job.start_mark = conf['_start_mark']
job.variant_description = conf.get(
'variant-description', " ".join(as_list(conf.get('branches'))))
if project_pipeline and conf['_source_context'].trusted:
# A config project has attached this job to a
# project-pipeline. In this case, we can ignore
# allowed-projects -- the superuser has stated they want
# it to run. This can be useful to allow untrusted jobs
# with secrets to be run in other untrusted projects.
job.ignore_allowed_projects = True
if 'parent' in conf:
if conf['parent'] is not None:
# Parent job is explicitly specified, so inherit from it.
job.parent = conf['parent']
# Parent is explicitly set as None, so user intends
# this to be a base job. That's only okay if we're in
# a config project.
if not conf['_source_context'].trusted:
raise Exception(
"Base jobs must be defined in config projects")
job.parent = job.BASE_JOB_MARKER
# Secrets are part of the playbook context so we must establish
# them earlier than playbooks.
secrets = []
for secret_config in as_list(conf.get('secrets', [])):
if isinstance(secret_config, str):
secret_name = secret_config
secret_alias = secret_config
secret_ptp = False
secret_name = secret_config['secret']
secret_alias = secret_config['name']
secret_ptp = secret_config.get('pass-to-parent', False)
secret_use = model.SecretUse(secret_name, secret_alias)
secret_use.pass_to_parent = secret_ptp
job.secrets = tuple(secrets)
# A job in an untrusted repo that uses secrets requires
# special care. We must note this, and carry this flag
# through inheritance to ensure that we don't run this job in
# an unsafe check pipeline. We must also set allowed-projects
# to only the current project, as otherwise, other projects
# might be able to cause something to happen with the secret
# by using a depends-on header.
if secrets and not conf['_source_context'].trusted:
job.post_review = True
job.allowed_projects = frozenset((
if (conf.get('timeout') and
self.pcontext.tenant.max_job_timeout != -1 and
int(conf['timeout']) > self.pcontext.tenant.max_job_timeout):
raise MaxTimeoutError(job, self.pcontext.tenant)
if (conf.get('post-timeout') and
self.pcontext.tenant.max_job_timeout != -1 and
int(conf['post-timeout']) > self.pcontext.tenant.max_job_timeout):
raise MaxTimeoutError(job, self.pcontext.tenant)
if 'post-review' in conf:
if conf['post-review']:
job.post_review = True
raise Exception("Once set, the post-review attribute "
"may not be unset")
# Configure and validate ansible version
if 'ansible-version' in conf:
# The ansible-version can be treated by yaml as a float so convert
# it to a string.
ansible_version = str(conf['ansible-version'])
job.ansible_version = ansible_version
# Roles are part of the playbook context so we must establish
# them earlier than playbooks.
roles = []
if 'roles' in conf:
for role in conf.get('roles', []):
if 'zuul' in role:
r = self._makeZuulRole(job, role)
if r:
# A job's repo should be an implicit role source for that job,
# but not in a project-pipeline variant.
if not project_pipeline:
r = self._makeImplicitRole(job)
roles.insert(0, r)
for pre_run_name in as_list(conf.get('pre-run')):
pre_run = model.PlaybookContext(job.source_context,
pre_run_name, job.roles,
job.pre_run = job.pre_run + (pre_run,)
# NOTE(pabelanger): Reverse the order of our post-run list. We prepend
# post-runs for inherits however, we want to execute post-runs in the
# order they are listed within the job.
for post_run_name in reversed(as_list(conf.get('post-run'))):
post_run = model.PlaybookContext(job.source_context,
post_run_name, job.roles,
job.post_run = (post_run,) + job.post_run
for cleanup_run_name in reversed(as_list(conf.get('cleanup-run'))):
cleanup_run = model.PlaybookContext(job.source_context,
cleanup_run_name, job.roles,
job.cleanup_run = (cleanup_run,) + job.cleanup_run
if 'run' in conf:
for run_name in as_list(conf.get('run')):
run = model.PlaybookContext(job.source_context, run_name,
job.roles, secrets) = + (run,)
if conf.get('intermediate', False) and not conf.get('abstract', False):
raise Exception("An intermediate job must also be abstract")
for k in self.simple_attributes:
a = k.replace('-', '_')
if k in conf:
setattr(job, a, conf[k])
if 'nodeset' in conf:
conf_nodeset = conf['nodeset']
if isinstance(conf_nodeset, str):
# This references an existing named nodeset in the
# layout; it will be validated later.
ns = conf_nodeset
ns = self.pcontext.nodeset_parser.fromYaml(
conf_nodeset, anonymous=True)
job.nodeset = ns
if 'required-projects' in conf:
new_projects = {}
projects = as_list(conf.get('required-projects', []))
unknown_projects = []
for project in projects:
if isinstance(project, dict):
project_name = project['name']
project_override_branch = project.get('override-branch')
project_override_checkout = project.get(
project_name = project
project_override_branch = None
project_override_checkout = None
(trusted, project) = self.pcontext.tenant.getProject(
if project is None:
job_project = model.JobProject(project.canonical_name,
new_projects[project.canonical_name] = job_project
# NOTE(mnaser): We accumulate all unknown projects and throw an
# exception only once to capture all of them in the
# error message.
if unknown_projects:
names = ", ".join(unknown_projects)
raise Exception("Unknown projects: %s" % (names,))
job.required_projects = new_projects
if 'dependencies' in conf:
new_dependencies = []
dependencies = as_list(conf.get('dependencies', []))
for dep in dependencies:
if isinstance(dep, dict):
dep_name = dep['name']
dep_soft = dep.get('soft', False)
dep_name = dep
dep_soft = False
job_dependency = model.JobDependency(dep_name, dep_soft)
job.dependencies = new_dependencies
semaphores = as_list(conf.get('semaphores', conf.get('semaphore', [])))
job_semaphores = []
for semaphore in semaphores:
if isinstance(semaphore, str):
semaphore.get('resources-first', False)))
if job_semaphores:
# Sort the list of semaphores to avoid issues with
# contention (where two jobs try to start at the same time
# and fail due to acquiring the same semaphores but in
# reverse order.
job.semaphores = tuple(sorted(job_semaphores,
key=lambda x:
for k in ('tags', 'requires', 'provides'):
v = frozenset(as_list(conf.get(k)))
if v:
setattr(job, k, v)
variables = conf.get('vars', None)
if variables:
job.variables = variables
extra_variables = conf.get('extra-vars', None)
if extra_variables:
job.extra_variables = extra_variables
host_variables = conf.get('host-vars', None)
if host_variables:
for host, hvars in host_variables.items():
job.host_variables = host_variables
group_variables = conf.get('group-vars', None)
if group_variables:
for group, gvars in group_variables.items():
job.group_variables = group_variables
allowed_projects = conf.get('allowed-projects', None)
# See note above at "post-review".
if allowed_projects and not job.allowed_projects:
allowed = []
for p in as_list(allowed_projects):
(trusted, project) = self.pcontext.tenant.getProject(p)
if project is None:
raise Exception("Unknown project %s" % (p,))
job.allowed_projects = frozenset(allowed)
branches = None
implied = False
if 'branches' in conf:
branches = as_list(conf['branches'])
elif not project_pipeline:
branches = self.pcontext.getImpliedBranches(job.source_context)
implied = True
if branches:
job.setBranchMatcher(branches, implied=implied)
if 'files' in conf:
if 'irrelevant-files' in conf:
return job
def _makeZuulRole(self, job, role):
name = role['zuul'].split('/')[-1]
(trusted, project) = self.pcontext.tenant.getProject(role['zuul'])
if project is None:
return None
return model.ZuulRole(role.get('name', name),
def _makeImplicitRole(self, job):
project = job.source_context.project
name ='/')[-1]
name = JobParser.ANSIBLE_ROLE_RE.sub('', name) or name
return model.ZuulRole(name,
class ProjectTemplateParser(object):
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.ProjectTemplateParser")
self.pcontext = pcontext
self.schema = self.getSchema()
self.not_pipelines = ['name', 'description', 'templates',
'merge-mode', 'default-branch', 'vars',
'queue', '_source_context', '_start_mark']
def getSchema(self):
job = {str: vs.Any(str, JobParser.job_attributes)}
job_list = [vs.Any(str, job)]
pipeline_contents = {
# TODO(tobiash): Remove pipeline specific queue after deprecation
'queue': str,
'debug': bool,
'fail-fast': bool,
'jobs': job_list
project = {
'name': str,
'description': str,
'queue': str,
'vars': ansible_vars_dict,
str: pipeline_contents,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
return vs.Schema(project)
def fromYaml(self, conf, validate=True, freeze=True):
if validate:
source_context = conf['_source_context']
start_mark = conf['_start_mark']
project_template = model.ProjectConfig(conf.get('name'))
project_template.source_context = conf['_source_context']
project_template.start_mark = conf['_start_mark']
project_template.queue_name = conf.get('queue')
for pipeline_name, conf_pipeline in conf.items():
if pipeline_name in self.not_pipelines:
project_pipeline = model.ProjectPipelineConfig()
project_template.pipelines[pipeline_name] = project_pipeline
# TODO(tobiash): Remove pipeline specific queue after deprecation
project_pipeline.queue_name = conf_pipeline.get('queue')
project_pipeline.debug = conf_pipeline.get('debug')
project_pipeline.fail_fast = conf_pipeline.get(
conf_pipeline.get('jobs', []),
source_context, start_mark, project_pipeline.job_list)
# If this project definition is in a place where it
# should get implied branch matchers, set it.
branches = self.pcontext.getImpliedBranches(source_context)
if branches:
variables = conf.get('vars', {})
forbidden = {'zuul', 'nodepool', 'unsafe_vars'}
if variables:
if set(variables.keys()).intersection(forbidden):
raise Exception("Variables named 'zuul', 'nodepool', "
"or 'unsafe_vars' are not allowed.")
project_template.variables = variables
if freeze:
return project_template
def parseJobList(self, conf, source_context, start_mark, job_list):
for conf_job in conf:
if isinstance(conf_job, str):
jobname = conf_job
attrs = {}
elif isinstance(conf_job, dict):
# A dictionary in a job tree may override params
jobname, attrs = list(conf_job.items())[0]
raise Exception("Job must be a string or dictionary")
attrs['_source_context'] = source_context
attrs['_start_mark'] = start_mark
attrs, project_pipeline=True,
name=jobname, validate=False))
class ProjectParser(object):
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.ProjectParser")
self.pcontext = pcontext
self.schema = self.getSchema()
def getSchema(self):
job = {str: vs.Any(str, JobParser.job_attributes)}
job_list = [vs.Any(str, job)]
pipeline_contents = {
# TODO(tobiash): Remove pipeline specific queue after deprecation
'queue': str,
'debug': bool,
'fail-fast': bool,
'jobs': job_list
project = {
'name': str,
'description': str,
'vars': ansible_vars_dict,
'templates': [str],
'merge-mode': vs.Any('merge', 'merge-resolve',
'cherry-pick', 'squash-merge'),
'default-branch': str,
'queue': str,
str: pipeline_contents,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
return vs.Schema(project)
def fromYaml(self, conf):
project_name = conf.get('name')
if not project_name:
# There is no name defined so implicitly add the name
# of the project where it is defined.
project_name = (conf['_source_context'].project.canonical_name)
if project_name.startswith('^'):
# regex matching is designed to match other projects so disallow
# in untrusted contexts
if not conf['_source_context'].trusted:
raise ProjectNotPermittedError()
# Parse the project as a template since they're mostly the
# same.
project_config = self.pcontext.project_template_parser. \
fromYaml(conf, validate=False, freeze=False) = project_name
(trusted, project) = self.pcontext.tenant.getProject(project_name)
if project is None:
raise ProjectNotFoundError(project_name)
if not conf['_source_context'].trusted:
if project != conf['_source_context'].project:
raise ProjectNotPermittedError()
# Parse the project as a template since they're mostly the
# same.
project_config = self.pcontext.project_template_parser.\
fromYaml(conf, validate=False, freeze=False) = project.canonical_name
# Pragmas can cause templates to end up with implied
# branch matchers for arbitrary branches, but project
# stanzas should not. They should either have the current
# branch or no branch matcher.
if conf['_source_context'].trusted:
# Add templates
for name in conf.get('templates', []):
if name not in project_config.templates:
mode = conf.get('merge-mode', 'merge-resolve')
project_config.merge_mode = model.MERGER_MAP[mode]
default_branch = conf.get('default-branch', 'master')
project_config.default_branch = default_branch
project_config.queue_name = conf.get('queue', None)
variables = conf.get('vars', {})
forbidden = {'zuul', 'nodepool', 'unsafe_vars'}
if variables:
if set(variables.keys()).intersection(forbidden):
raise Exception("Variables named 'zuul', 'nodepool', "
"or 'unsafe_vars' are not allowed.")
project_config.variables = variables
return project_config
class PipelineParser(object):
# A set of reporter configuration keys to action mapping
reporter_actions = {
'enqueue': 'enqueue_actions',
'start': 'start_actions',
'success': 'success_actions',
'failure': 'failure_actions',
'merge-failure': 'merge_failure_actions',
'no-jobs': 'no_jobs_actions',
'disabled': 'disabled_actions',
'dequeue': 'dequeue_actions',
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.PipelineParser")
self.pcontext = pcontext
self.schema = self.getSchema()
def getDriverSchema(self, dtype):
methods = {
'trigger': 'getTriggerSchema',
'reporter': 'getReporterSchema',
'require': 'getRequireSchema',
'reject': 'getRejectSchema',
schema = {}
# Add the configured connections as available layout options
for connection_name, connection in \
method = getattr(connection.driver, methods[dtype], None)
if method:
schema[connection_name] = to_list(method())
return schema
def getSchema(self):
manager = vs.Any('independent',
precedence = vs.Any('normal', 'low', 'high')
window = vs.All(int, vs.Range(min=0))
window_floor = vs.All(int, vs.Range(min=1))
window_type = vs.Any('linear', 'exponential')
window_factor = vs.All(int, vs.Range(min=1))
pipeline = {vs.Required('name'): str,
vs.Required('manager'): manager,
'precedence': precedence,
'supercedes': to_list(str),
'description': str,
'success-message': str,
'failure-message': str,
'start-message': str,
'merge-failure-message': str,
'no-jobs-message': str,
'footer-message': str,
'dequeue-message': str,
'dequeue-on-new-patchset': bool,
'ignore-dependencies': bool,
'post-review': bool,
vs.All(int, vs.Range(min=1)),
'window': window,
'window-floor': window_floor,
'window-increase-type': window_type,
'window-increase-factor': window_factor,
'window-decrease-type': window_type,
'window-decrease-factor': window_factor,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
pipeline['require'] = self.getDriverSchema('require')
pipeline['reject'] = self.getDriverSchema('reject')
pipeline['trigger'] = vs.Required(self.getDriverSchema('trigger'))
for action in ['enqueue', 'start', 'success', 'failure',
'merge-failure', 'no-jobs', 'disabled', 'dequeue']:
pipeline[action] = self.getDriverSchema('reporter')
return vs.Schema(pipeline)
def fromYaml(self, conf):
pipeline = model.Pipeline(conf['name'], self.pcontext.tenant)
pipeline.source_context = conf['_source_context']
pipeline.start_mark = conf['_start_mark']
pipeline.description = conf.get('description')
pipeline.supercedes = as_list(conf.get('supercedes', []))
precedence = model.PRECEDENCE_MAP[conf.get('precedence')]
pipeline.precedence = precedence
pipeline.failure_message = conf.get('failure-message',
"Build failed.")
pipeline.merge_failure_message = conf.get(
'merge-failure-message', "Merge Failed.\n\nThis change or one "
"of its cross-repo dependencies was unable to be "
"automatically merged with the current state of its "
"repository. Please rebase the change and upload a new "
pipeline.success_message = conf.get('success-message',
"Build succeeded.")
pipeline.footer_message = conf.get('footer-message', "")
pipeline.start_message = conf.get('start-message',
"Starting {} jobs.")
pipeline.enqueue_message = conf.get('enqueue-message', "")
pipeline.no_jobs_message = conf.get('no-jobs-message', "")
pipeline.dequeue_message = conf.get(
"dequeue-message", "Build canceled."
pipeline.dequeue_on_new_patchset = conf.get(
'dequeue-on-new-patchset', True)
pipeline.ignore_dependencies = conf.get(
'ignore-dependencies', False)
pipeline.post_review = conf.get(
'post-review', False)
for conf_key, action in self.reporter_actions.items():
reporter_set = []
allowed_reporters = self.pcontext.tenant.allowed_reporters
if conf.get(conf_key):
for reporter_name, params \
in conf.get(conf_key).items():
if allowed_reporters is not None and \
reporter_name not in allowed_reporters:
raise UnknownConnection(reporter_name)
if type(self.pcontext.connections
.connections[reporter_name]) == SQLConnection:
# (1)... explicit SQL reporters are ignored)
self.log.warning("Ignoring SQL reporter configured in"
" pipeline %s" %
reporter = self.pcontext.connections.getReporter(
reporter_name, pipeline, params)
setattr(pipeline, action, reporter_set)
# If merge-failure actions aren't explicit, use the failure actions
if not pipeline.merge_failure_actions:
pipeline.merge_failure_actions = pipeline.failure_actions
pipeline.disable_at = conf.get(
'disable-after-consecutive-failures', None)
pipeline.window = conf.get('window', 20)
pipeline.window_floor = conf.get('window-floor', 3)
pipeline.window_increase_type = conf.get(
'window-increase-type', 'linear')
pipeline.window_increase_factor = conf.get(
'window-increase-factor', 1)
pipeline.window_decrease_type = conf.get(
'window-decrease-type', 'exponential')
pipeline.window_decrease_factor = conf.get(
'window-decrease-factor', 2)
manager_name = conf['manager']
if manager_name == 'dependent':
manager = zuul.manager.dependent.DependentPipelineManager(
self.pcontext.scheduler, pipeline)
elif manager_name == 'independent':
manager = zuul.manager.independent.IndependentPipelineManager(
self.pcontext.scheduler, pipeline)
elif manager_name == 'serial':
manager = zuul.manager.serial.SerialPipelineManager(
self.pcontext.scheduler, pipeline)
elif manager_name == 'supercedent':
manager = zuul.manager.supercedent.SupercedentPipelineManager(
self.pcontext.scheduler, pipeline)
for source_name, require_config in conf.get('require', {}).items():
source = self.pcontext.connections.getSource(source_name)
for source_name, reject_config in conf.get('reject', {}).items():
source = self.pcontext.connections.getSource(source_name)
for connection_name, trigger_config in conf.get('trigger').items():
if self.pcontext.tenant.allowed_triggers is not None and \
connection_name not in self.pcontext.tenant.allowed_triggers:
raise UnknownConnection(connection_name)
trigger = self.pcontext.connections.getTrigger(
connection_name, trigger_config)
# Pipelines don't get frozen
return pipeline
class SemaphoreParser(object):
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.SemaphoreParser")
self.pcontext = pcontext
self.schema = self.getSchema()
def getSchema(self):
semaphore = {vs.Required('name'): str,
'max': int,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
return vs.Schema(semaphore)
def fromYaml(self, conf):
semaphore = model.Semaphore(conf['name'], conf.get('max', 1))
semaphore.source_context = conf.get('_source_context')
semaphore.start_mark = conf.get('_start_mark')
return semaphore
class QueueParser:
def __init__(self, pcontext):
self.log = logging.getLogger("zuul.QueueParser")
self.pcontext = pcontext
self.schema = self.getSchema()
def getSchema(self):
queue = {vs.Required('name'): str,
'per-branch': bool,
'allow-circular-dependencies': bool,
'_source_context': model.SourceContext,
'_start_mark': ZuulMark,
return vs.Schema(queue)
def fromYaml(self, conf):
queue = model.Queue(
conf.get('per-branch', False),
conf.get('allow-circular-dependencies', False),
queue.source_context = conf.get('_source_context')
queue.start_mark = conf.get('_start_mark')
return queue
class AuthorizationRuleParser(object):
def __init__(self):
self.log = logging.getLogger("zuul.AuthorizationRuleParser")
self.schema = self.getSchema()
def getSchema(self):
authRule = {vs.Required('name'): str,
vs.Required('conditions'): to_list(dict)
return vs.Schema(authRule)
def fromYaml(self, conf):
a = model.AuthZRuleTree(conf['name'])
def parse_tree(node):
if isinstance(node, list):
return model.OrRule(parse_tree(x) for x in node)
elif isinstance(node, dict):
subrules = []
for claim, value in node.items():
if claim == 'zuul_uid':
claim = '__zuul_uid_claim'
subrules.append(model.ClaimRule(claim, value))
return model.AndRule(subrules)
raise Exception('Invalid claim declaration %r' % node)
a.ruletree = parse_tree(conf['conditions'])
return a
class ParseContext(object):
"""Hold information about a particular run of the parser"""
def __init__(self, connections, scheduler, tenant, ansible_manager):
self.connections = connections
self.scheduler = scheduler
self.tenant = tenant
self.ansible_manager = ansible_manager
self.pragma_parser = PragmaParser(self)
self.pipeline_parser = PipelineParser(self)
self.nodeset_parser = NodeSetParser(self)
self.secret_parser = SecretParser(self)
self.job_parser = JobParser(self)
self.semaphore_parser = SemaphoreParser(self)
self.queue_parser = QueueParser(self)
self.project_template_parser = ProjectTemplateParser(self)
self.project_parser = ProjectParser(self)
def getImpliedBranches(self, source_context):
# If the user has set a pragma directive for this, use the
# value (if unset, the value is None).
if source_context.implied_branch_matchers is True:
if source_context.implied_branches is not None:
return source_context.implied_branches
return [source_context.branch]
elif source_context.implied_branch_matchers is False:
return None
# If this is a trusted project, don't create implied branch
# matchers.
if source_context.trusted:
return None
# If this project only has one branch, don't create implied
# branch matchers. This way central job repos can work.
branches = self.tenant.getProjectBranches(
if len(branches) == 1:
return None
if source_context.implied_branches is not None:
return source_context.implied_branches
return [source_context.branch]
class TenantParser(object):
def __init__(self, connections, scheduler, merger, keystorage):
self.log = logging.getLogger("zuul.TenantParser")
self.connections = connections
self.scheduler = scheduler
self.merger = merger
self.keystorage = keystorage
self.unparsed_config_cache = self.scheduler.unparsed_config_cache
classes = vs.Any('pipeline', 'job', 'semaphore', 'project',
'project-template', 'nodeset', 'secret', 'queue')
project_dict = {str: {
'include': to_list(classes),
'exclude': to_list(classes),
'shadow': to_list(str),
'exclude-unprotected-branches': bool,
'extra-config-paths': to_list(str),
'load-branch': str,
'allow-circular-dependencies': bool,
project = vs.Any(str, project_dict)
group = {
'include': to_list(classes),
'exclude': to_list(classes),
vs.Required('projects'): to_list(project),
project_or_group = vs.Any(project, group)
tenant_source = vs.Schema({
'config-projects': to_list(project_or_group),
'untrusted-projects': to_list(project_or_group),
def validateTenantSources(self):
def v(value, path=[]):
if isinstance(value, dict):
for k, val in value.items():
self.validateTenantSource(val, path + [k])
raise vs.Invalid("Invalid tenant source", path)
return v
def validateTenantSource(self, value, path=[]):
def getSchema(self):
tenant = {vs.Required('name'): str,
'max-nodes-per-job': int,
'max-job-timeout': int,
'source': self.validateTenantSources(),
'exclude-unprotected-branches': bool,
'allowed-triggers': to_list(str),
'allowed-reporters': to_list(str),
'allowed-labels': to_list(str),
'disallowed-labels': to_list(str),
'allow-circular-dependencies': bool,
'default-parent': str,
'default-ansible-version': vs.Any(str, float),
'admin-rules': to_list(str),
# TODO: Ignored, allowed for backwards compat, remove for v5.
'report-build-page': bool,
'web-root': str,
return vs.Schema(tenant)
def fromYaml(self, abide, conf, ansible_manager, min_ltimes=None):
tenant = model.Tenant(conf['name'])
pcontext = ParseContext(self.connections, self.scheduler,
tenant, ansible_manager)
if conf.get('max-nodes-per-job') is not None:
tenant.max_nodes_per_job = conf['max-nodes-per-job']
if conf.get('max-job-timeout') is not None:
tenant.max_job_timeout = int(conf['max-job-timeout'])
if conf.get('exclude-unprotected-branches') is not None:
tenant.exclude_unprotected_branches = \
if conf.get('admin-rules') is not None:
tenant.authorization_rules = conf['admin-rules']
tenant.web_root = conf.get('web-root', self.scheduler.web_root)
if tenant.web_root and not tenant.web_root.endswith('/'):
tenant.web_root += '/'
tenant.allowed_triggers = conf.get('allowed-triggers')
tenant.allowed_reporters = conf.get('allowed-reporters')
tenant.allowed_labels = conf.get('allowed-labels')
tenant.disallowed_labels = conf.get('disallowed-labels')
tenant.default_base_job = conf.get('default-parent', 'base')
tenant.unparsed_config = conf
# tpcs is TenantProjectConfigs
config_tpcs = abide.getConfigTPCs(
for tpc in config_tpcs:
untrusted_tpcs = abide.getUntrustedTPCs(
for tpc in untrusted_tpcs:
# We prepare a stack to store config loading issues
loading_errors = model.LoadingErrors()
for tpc in config_tpcs + untrusted_tpcs:
source_context = model.ProjectContext(tpc.project)
with project_configuration_exceptions(source_context,
self._getProjectBranches(tenant, tpc)
self._resolveShadowProjects(tenant, tpc)
# Set default ansible version
default_ansible_version = conf.get('default-ansible-version')
if default_ansible_version is not None:
# The ansible version can be interpreted as float by yaml so make
# sure it's a string.
default_ansible_version = str(default_ansible_version)
default_ansible_version = ansible_manager.default_version
tenant.default_ansible_version = default_ansible_version
# Start by fetching any YAML needed by this tenant which isn't
# already cached. Full reconfigurations start with an empty
# cache.
self._cacheTenantYAML(abide, tenant, loading_errors, min_ltimes)
# Then collect the appropriate YAML based on this tenant
# config.
config_projects_config, untrusted_projects_config = \
self._loadTenantYAML(abide, tenant, loading_errors)
# Then convert the YAML to configuration objects which we
# cache on the tenant.
tenant.config_projects_config = self.parseConfig(
tenant, config_projects_config, loading_errors, pcontext)
tenant.untrusted_projects_config = self.parseConfig(
tenant, untrusted_projects_config, loading_errors, pcontext)
# Combine the trusted and untrusted config objects
parsed_config = model.ParsedConfig()
# Cache all of the objects on the individual project-branches
# for later use during dynamic reconfigurations.
self.cacheConfig(tenant, parsed_config)
tenant.layout = self._parseLayout(
tenant, parsed_config, loading_errors)
tenant.semaphore_handler = SemaphoreHandler(
self.scheduler.zk_client, self.scheduler.statsd,, tenant.layout
return tenant
def _resolveShadowProjects(self, tenant, tpc):
shadow_projects = []
for sp in tpc.shadow_projects:
_, project = tenant.getProject(sp)
if project is None:
raise ProjectNotFoundError(sp)
tpc.shadow_projects = frozenset(shadow_projects)
def _getProjectBranches(self, tenant, tpc):
branches = sorted(tpc.project.source.getProjectBranches(
tpc.project, tenant))
if 'master' in branches:
branches = ['master'] + branches
tpc.branches = branches
def _loadProjectKeys(self, connection_name, project):
project.private_secrets_key, project.public_secrets_key = (
project.private_ssh_key, project.public_ssh_key = (
def _getProject(source, conf, current_include):
extra_config_files = ()
extra_config_dirs = ()
if isinstance(conf, str):
# Return a project object whether conf is a dict or a str
project = source.getProject(conf)
project_include = current_include
shadow_projects = []
project_exclude_unprotected_branches = None
project_load_branch = None
project_name = list(conf.keys())[0]
project = source.getProject(project_name)
shadow_projects = as_list(conf[project_name].get('shadow', []))
# We check for None since the user may set include to an empty list
if conf[project_name].get("include") is None:
project_include = current_include
project_include = frozenset(
project_exclude = frozenset(
as_list(conf[project_name].get('exclude', [])))
if project_exclude:
project_include = frozenset(project_include - project_exclude)
project_exclude_unprotected_branches = conf[project_name].get(
'exclude-unprotected-branches', None)
if conf[project_name].get('extra-config-paths') is not None:
extra_config_paths = as_list(
extra_config_files = tuple([x for x in extra_config_paths
if not x.endswith('/')])
extra_config_dirs = tuple([x[:-1] for x in extra_config_paths
if x.endswith('/')])
project_load_branch = conf[project_name].get(
'load-branch', None)
tenant_project_config = model.TenantProjectConfig(project)
tenant_project_config.load_classes = frozenset(project_include)
tenant_project_config.shadow_projects = shadow_projects
tenant_project_config.exclude_unprotected_branches = \
tenant_project_config.extra_config_files = extra_config_files
tenant_project_config.extra_config_dirs = extra_config_dirs
tenant_project_config.load_branch = project_load_branch
return tenant_project_config
def _getProjects(self, source, conf, current_include):
# Return a project object whether conf is a dict or a str
projects = []
if isinstance(conf, str):
# A simple project name string
projects.append(self._getProject(source, conf, current_include))
elif len(conf.keys()) > 1 and 'projects' in conf:
# This is a project group
if 'include' in conf:
current_include = set(as_list(conf['include']))
current_include = current_include.copy()
if 'exclude' in conf:
exclude = set(as_list(conf['exclude']))
current_include = current_include - exclude
for project in conf['projects']:
sub_projects = self._getProjects(
source, project, current_include)
elif len(conf.keys()) == 1:
# A project with overrides
source, conf, current_include))
raise Exception("Unable to parse project %s", conf)
return projects
def loadTenantProjects(self, conf_tenant):
config_projects = []
untrusted_projects = []
default_include = frozenset(['pipeline', 'job', 'semaphore', 'project',
'secret', 'project-template', 'nodeset',
for source_name, conf_source in conf_tenant.get('source', {}).items():
source = self.connections.getSource(source_name)
current_include = default_include
for conf_repo in conf_source.get('config-projects', []):
# tpcs = TenantProjectConfigs
tpcs = self._getProjects(source, conf_repo, current_include)
for tpc in tpcs:
self._loadProjectKeys(source_name, tpc.project)
current_include = frozenset(default_include - set(['pipeline']))
for conf_repo in conf_source.get('untrusted-projects', []):
tpcs = self._getProjects(source, conf_repo,
for tpc in tpcs:
self._loadProjectKeys(source_name, tpc.project)
return config_projects, untrusted_projects
def _cacheTenantYAML(self, abide, tenant, loading_errors, min_ltimes):
jobs = []
for project in itertools.chain(
tenant.config_projects, tenant.untrusted_projects):
tpc = tenant.project_configs[project.canonical_name]
# For each branch in the repo, get the zuul.yaml for that
# branch. Remember the branch and then implicitly add a
# branch selector to each job there. This makes the
# in-repo configuration apply only to that branch.
branches = tenant.getProjectBranches(project)
for branch in branches:
if not tpc.load_classes:
# If all config classes are excluded then do not
# request any getFiles jobs.
source_context = model.SourceContext(
project, branch, '', False)
if min_ltimes is not None:
files_cache = self.unparsed_config_cache.getFilesCache(
project.canonical_name, branch)
with self.unparsed_config_cache.readLock(
pb_ltime = min_ltimes[project.canonical_name][branch]
if files_cache.isValidFor(tpc, pb_ltime):
"Using files from cache for project %s @%s",
project.canonical_name, branch)
branch_cache = abide.getUnparsedBranchCache(
project.canonical_name, branch)
if branch_cache.isValidFor(tpc, files_cache.ltime):
# Unparsed branch cache is already up-to-date
abide, tenant, source_context, files_cache,
loading_errors, files_cache.ltime)
extra_config_files = abide.getExtraConfigFiles(
extra_config_dirs = abide.getExtraConfigDirs(
ltime = self.scheduler.zk_client.getCurrentLtime()
job = self.merger.getFiles(
project.source.connection.connection_name,, branch,
files=(['zuul.yaml', '.zuul.yaml'] +
dirs=['zuul.d', '.zuul.d'] + list(extra_config_dirs))
self.log.debug("Submitting cat job %s for %s %s %s" % (
job, project.source.connection.connection_name,, branch))
job.extra_config_files = extra_config_files
job.extra_config_dirs = extra_config_dirs
job.ltime = ltime
job.source_context = source_context
for job in jobs:
self.log.debug("Waiting for cat job %s" % (job,))
if not hasattr(job, 'updated'):
# We timed out
raise Exception("Cat job %s timed out; consider setting "
"merger.git_timeout in zuul.conf" % (job,))
if not job.updated:
raise Exception("Cat job %s failed" % (job,))
self.log.debug("Cat job %s got files %s" %
(job, job.files.keys()))
self._updateUnparsedBranchCache(abide, tenant, job.source_context,
job.files, loading_errors,
# Save all config files in Zookeeper (not just for the current tpc)
files_cache = self.unparsed_config_cache.getFilesCache(
with self.unparsed_config_cache.writeLock(project.canonical_name):
for fn, content in job.files.items():
# Cache file in Zookeeper
if content is not None:
files_cache[fn] = content
def _updateUnparsedBranchCache(self, abide, tenant, source_context, files,
loading_errors, ltime):
loaded = False
tpc = tenant.project_configs[source_context.project.canonical_name]
# Make sure we are clearing the local cache before updating it.
branch_cache = abide.getUnparsedBranchCache(
for conf_root in (
('zuul.yaml', 'zuul.d', '.zuul.yaml', '.zuul.d') +
tpc.extra_config_files + tpc.extra_config_dirs):
for fn in sorted(files.keys()):
fn_root = fn.split('/')[0]
if fn_root != conf_root or not files.get(fn):
# Don't load from more than one configuration in a
# project-branch (unless an "extra" file/dir).
if (conf_root not in tpc.extra_config_files and
conf_root not in tpc.extra_config_dirs):
if (loaded and loaded != conf_root):
self.log.warning("Multiple configuration files in %s",
loaded = conf_root
# Create a new source_context so we have unique filenames.
source_context = source_context.copy()
source_context.path = fn
"Loading configuration from %s" %
incdata = self.loadProjectYAML(
files[fn], source_context, loading_errors)
branch_cache.put(source_context.path, incdata)
branch_cache.setValidFor(tpc, ltime)
def _loadTenantYAML(self, abide, tenant, loading_errors):
config_projects_config = model.UnparsedConfig()
untrusted_projects_config = model.UnparsedConfig()
for project in tenant.config_projects:
tpc = tenant.project_configs.get(project.canonical_name)
branch = tpc.load_branch if tpc.load_branch else 'master'
branch_cache = abide.getUnparsedBranchCache(
project.canonical_name, branch)
tpc = tenant.project_configs[project.canonical_name]
unparsed_branch_config = branch_cache.get(tpc)
if unparsed_branch_config:
unparsed_branch_config = self.filterConfigProjectYAML(
for project in tenant.untrusted_projects:
branches = tenant.getProjectBranches(project)
for branch in branches:
branch_cache = abide.getUnparsedBranchCache(
project.canonical_name, branch)
tpc = tenant.project_configs[project.canonical_name]
unparsed_branch_config = branch_cache.get(tpc)
if unparsed_branch_config:
unparsed_branch_config = self.filterUntrustedProjectYAML(
unparsed_branch_config, loading_errors)
return config_projects_config, untrusted_projects_config
def loadProjectYAML(self, data, source_context, loading_errors):
config = model.UnparsedConfig()
with early_configuration_exceptions(source_context):
r = safe_load_yaml(data, source_context)
except ConfigurationSyntaxError as e:
loading_errors.addError(source_context, None, e)
return config
def filterConfigProjectYAML(self, data):
# Any config object may appear in a config project.
return data.copy(trusted=True)
def filterUntrustedProjectYAML(self, data, loading_errors):
if data and data.pipelines:
with configuration_exceptions(
'pipeline', data.pipelines[0], loading_errors):
raise PipelineNotPermittedError()
return data.copy(trusted=False)
def _getLoadClasses(self, tenant, conf_object):
project = conf_object.get('_source_context').project
tpc = tenant.project_configs[project.canonical_name]
return tpc.load_classes
def parseConfig(self, tenant, unparsed_config, loading_errors, pcontext):
parsed_config = model.ParsedConfig()
# Handle pragma items first since they modify the source context
# used by other classes.
for config_pragma in unparsed_config.pragmas:
except ConfigurationSyntaxError as e:
config_pragma['_start_mark'], e)
for config_pipeline in unparsed_config.pipelines:
classes = self._getLoadClasses(tenant, config_pipeline)
if 'pipeline' not in classes:
with configuration_exceptions('pipeline',
config_pipeline, loading_errors):
for config_nodeset in unparsed_config.nodesets:
classes = self._getLoadClasses(tenant, config_nodeset)
if 'nodeset' not in classes:
with configuration_exceptions('nodeset',
config_nodeset, loading_errors):
for config_secret in unparsed_config.secrets:
classes = self._getLoadClasses(tenant, config_secret)
if 'secret' not in classes:
with configuration_exceptions('secret',
config_secret, loading_errors):
for config_job in
classes = self._getLoadClasses(tenant, config_job)
if 'job' not in classes:
with configuration_exceptions('job',
config_job, loading_errors):
for config_semaphore in unparsed_config.semaphores:
classes = self._getLoadClasses(tenant, config_semaphore)
if 'semaphore' not in classes:
with configuration_exceptions('semaphore',
config_semaphore, loading_errors):
for config_queue in unparsed_config.queues:
classes = self._getLoadClasses(tenant, config_queue)
if 'queue' not in classes:
with configuration_exceptions('queue',
config_queue, loading_errors):
for config_template in unparsed_config.project_templates:
classes = self._getLoadClasses(tenant, config_template)
if 'project-template' not in classes:
with configuration_exceptions(
'project-template', config_template, loading_errors):
for config_project in unparsed_config.projects:
classes = self._getLoadClasses(tenant, config_project)
if 'project' not in classes:
with configuration_exceptions('project', config_project,
# we need to separate the regex projects as they are
# processed differently later
name = config_project.get('name')
parsed_project = pcontext.project_parser.fromYaml(
if name and name.startswith('^'):
name, []).append(parsed_project)
return parsed_config
def cacheConfig(self, tenant, parsed_config):
def _cache(attr, obj):
tpc = tenant.project_configs[
branch_cache = tpc.parsed_branch_config.get(
if branch_cache is None:
branch_cache = tpc.parsed_branch_config.setdefault(
lst = getattr(branch_cache, attr)
# We don't cache pragma objects as they are acted on when
# parsed.
for pipeline in parsed_config.pipelines:
_cache('pipelines', pipeline)
for nodeset in parsed_config.nodesets:
_cache('nodesets', nodeset)
for secret in parsed_config.secrets:
_cache('secrets', secret)
for job in
_cache('jobs', job)
for queue in parsed_config.queues:
_cache('queues', queue)
for semaphore in parsed_config.semaphores:
_cache('semaphores', semaphore)
for template in parsed_config.project_templates:
_cache('project_templates', template)
for project_config in parsed_config.projects:
_cache('projects', project_config)
def _addLayoutItems(self, layout, tenant, parsed_config,
skip_pipelines=False, skip_semaphores=False):
# TODO(jeblair): make sure everything needing
# reference_exceptions has it; add tests if needed.
if not skip_pipelines:
for pipeline in parsed_config.pipelines:
with reference_exceptions(
'pipeline', pipeline, layout.loading_errors):
for nodeset in parsed_config.nodesets:
with reference_exceptions(
'nodeset', nodeset, layout.loading_errors):
for secret in parsed_config.secrets:
with reference_exceptions('secret', secret, layout.loading_errors):
for job in
with reference_exceptions('job', job, layout.loading_errors):
added = layout.addJob(job)
if not added:
"Skipped adding job %s which shadows an existing job" %
# Now that all the jobs are loaded, verify references to other
# config objects.
for jobs in
for job in jobs:
with reference_exceptions('job', job, layout.loading_errors):
for pipeline in layout.pipelines.values():
with reference_exceptions(
'pipeline', pipeline, layout.loading_errors):
if skip_semaphores:
# We should not actually update the layout with new
# semaphores, but so that we can validate that the config
# is correct, create a shadow layout here to which we add
# new semaphores so validation is complete.
semaphore_layout = model.Layout(tenant)
semaphore_layout = layout
for semaphore in parsed_config.semaphores:
with reference_exceptions(
'semaphore', semaphore, layout.loading_errors):
for queue in parsed_config.queues:
with reference_exceptions('queue', queue, layout.loading_errors):
for template in parsed_config.project_templates:
with reference_exceptions(
'project-template', template, layout.loading_errors):
# The project stanzas containing a regex are separated from the normal
# project stanzas and organized by regex. We need to loop over each
# regex and copy each stanza below the regex for each matching project.
for regex, config_projects in parsed_config.projects_by_regex.items():
projects_matching_regex = tenant.getProjectsByRegex(regex)
for trusted, project in projects_matching_regex:
for config_project in config_projects:
# we just override the project name here so a simple copy
# should be enough
conf = config_project.copy()
name = project.canonical_name = name
for project in parsed_config.projects:
# Now that all the project pipelines are loaded, fixup and
# verify references to other config objects.
def _validateProjectPipelineConfigs(self, layout):
# Validate references to other config objects
def inner_validate_ppcs(ppc):
for jobs in
for job in jobs:
# validate that the job exists on its own (an
# additional requirement for project-pipeline
# jobs)
for project_name in layout.project_configs:
for project_config in layout.project_configs[project_name]:
with reference_exceptions(
'project', project_config, layout.loading_errors):
for template_name in project_config.templates:
if template_name not in layout.project_templates:
raise TemplateNotFoundError(template_name)
project_templates = layout.getProjectTemplates(
for p_tmpl in project_templates:
with reference_exceptions(
'project-template', p_tmpl,
for ppc in p_tmpl.pipelines.values():
for ppc in project_config.pipelines.values():
def _parseLayout(self, tenant, data, loading_errors):
# Don't call this method from dynamic reconfiguration because
# it interacts with drivers and connections.
layout = model.Layout(tenant)
layout.loading_errors = loading_errors
self.log.debug("Created layout id %s", layout.uuid)
self._addLayoutItems(layout, tenant, data)
for pipeline in layout.pipelines.values():
return layout
class ConfigLoader(object):
log = logging.getLogger("zuul.ConfigLoader")
def __init__(self, connections, scheduler, merger, keystorage):
self.connections = connections
self.scheduler = scheduler
self.merger = merger
self.keystorage = keystorage
self.tenant_parser = TenantParser(connections, scheduler,
merger, self.keystorage)
self.admin_rule_parser = AuthorizationRuleParser()
def expandConfigPath(self, config_path):
if config_path:
config_path = os.path.expanduser(config_path)
if not os.path.exists(config_path):
raise Exception("Unable to read tenant config file at %s" %
return config_path
def readConfig(self, config_path, from_script=False):
config_path = self.expandConfigPath(config_path)
if not from_script:
with open(config_path) as config_file:"Loading configuration from %s" % (config_path,))
data = yaml.safe_load(config_file)
if not os.access(config_path, os.X_OK):
"Unable to read tenant configuration from a non "
"executable script (%s)" % config_path)
data = []
"Loading configuration from script %s" % config_path)
ret =
[config_path], stdout=subprocess.PIPE,
data = yaml.safe_load(ret.stdout)
except subprocess.CalledProcessError as error:
"Tenant config script exec failed: %s (%s)" % (
str(error), str(ret.stderr)))
data = []
unparsed_abide = model.UnparsedAbideConfig()
return unparsed_abide
def loadAdminRules(self, abide, unparsed_abide):
for conf_admin_rule in unparsed_abide.admin_rules:
admin_rule = self.admin_rule_parser.fromYaml(conf_admin_rule)
abide.admin_rules[] = admin_rule
def loadTPCs(self, abide, unparsed_abide, tenants=None):
if tenants:
tenants_to_load = {t: unparsed_abide.tenants[t] for t in tenants}
tenants_to_load = unparsed_abide.tenants
# Pre-load TenantProjectConfigs so we can get and cache all of a
# project's config files (incl. tenant specific extra config) at once.
for tenant_name, unparsed_config in tenants_to_load.items():
config_tpcs, untrusted_tpcs = (
for tpc in config_tpcs:
abide.addConfigTPC(tenant_name, tpc)
for tpc in untrusted_tpcs:
abide.addUntrustedTPC(tenant_name, tpc)
def loadTenant(self, abide, tenant_name, ansible_manager, unparsed_abide,
"""(Re-)load a single tenant.
Description of cache stages:
We have a local unparsed branch cache on each scheduler and the
global config cache in Zookeeper. Depending on the event that
triggers (re-)loading of a tenant we must make sure that those
caches are considered valid or invalid correctly.
If provided, the ``min_ltimes`` argument is expected to be a
nested dictionary with the project-branches. The value defines
the minimum logical time that is required for a cached config to
be considered valid::
"": {
"master": 12234,
"stable": -1,
"": {
"master": -1,
There are four scenarios to consider when loading a tenant.
1. Processing a tenant reconfig event:
- The min. ltime for the changed project(-branches) will be
set to the event's ``zuul_event_ltime`` (to establish a
happened-before relation in respect to the config change).
The min. ltime for all other project-branches will be -1.
- Config for needed project-branch(es) is updated via cat job
if the cache is not valid (cache ltime < min. ltime).
- Cache in Zookeeper and local unparsed branch cache is
updated. The ltime of the cache will be the timestamp
created shortly before requesting the config via the
mergers (only for outdated items).
2. Processing a FULL reconfiguration event:
- The min. ltime for all project-branches is given as the
``zuul_event_ltime`` of the reconfiguration event.
- Config for needed project-branch(es) is updated via cat job
if the cache is not valid (cache ltime < min. ltime).
Otherwise the local unparsed branch cache or the global
config cache in Zookeeper is used.
- Cache in Zookeeper and local unparsed branch cache is
updated, with the ltime shortly before requesting the
config via the mergers (only for outdated items).
3. Processing a SMART reconfiguration event:
- The min. ltime for all project-branches is given as -1 in
order to use cached data wherever possible.
- Config for new project-branch(es) is updated via cat job if
the project is not yet cached. Otherwise the local unparsed
branch cache or the global config cache in Zookeper is
- Cache in Zookeeper and local unparsed branch cache is
updated, with the ltime shortly before requesting the
config via the mergers (only for new items).
4. (Re-)loading a tenant due to a changed layout (happens after
an event according to one of the other scenarios was
processed on another scheduler):
- The min. ltime for all project-branches is given as -1 in
order to only use cached config.
- Local unparsed branch cache is updated if needed.
if tenant_name not in unparsed_abide.tenants:
# Copy tenants dictionary to not break concurrent iterations.
tenants = abide.tenants.copy()
del tenants[tenant_name]
abide.tenants = tenants
return None
unparsed_config = unparsed_abide.tenants[tenant_name]
new_tenant = self.tenant_parser.fromYaml(
abide, unparsed_config, ansible_manager, min_ltimes)
# Copy tenants dictionary to not break concurrent iterations.
tenants = abide.tenants.copy()
tenants[tenant_name] = new_tenant
abide.tenants = tenants
if len(new_tenant.layout.loading_errors):
"%s errors detected during %s tenant configuration loading",
len(new_tenant.layout.loading_errors), tenant_name)
# Log accumulated errors
for err in new_tenant.layout.loading_errors.errors[:10]:
return new_tenant