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.

1452 lines
62 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
import logging
import textwrap
import urllib
from abc import ABCMeta, abstractmethod
from zuul import exceptions
from zuul import model
from zuul.lib.dependson import find_dependency_headers
from zuul.lib.logutil import get_annotated_logger
from zuul.lib.tarjan import strongly_connected_components
class DynamicChangeQueueContextManager(object):
def __init__(self, change_queue):
self.change_queue = change_queue
def __enter__(self):
return self.change_queue
def __exit__(self, etype, value, tb):
if self.change_queue and not self.change_queue.queue:
class StaticChangeQueueContextManager(object):
def __init__(self, change_queue):
self.change_queue = change_queue
def __enter__(self):
return self.change_queue
def __exit__(self, etype, value, tb):
class PipelineManager(metaclass=ABCMeta):
"""Abstract Base Class for enqueing and processing Changes in a Pipeline"""
def __init__(self, sched, pipeline):
self.log = logging.getLogger("zuul.Pipeline.%s.%s" %
self.sched = sched
self.pipeline = pipeline
self.event_filters = []
self.ref_filters = []
def __str__(self):
return "<%s %s>" % (self.__class__.__name__,
def _postConfig(self, layout):
# All pipelines support shared queues for setting
# relative_priority; only the dependent pipeline uses them for
# pipeline queing.
def buildChangeQueues(self, layout):
self.log.debug("Building relative_priority queues")
change_queues = self.pipeline.relative_priority_queues
tenant = self.pipeline.tenant
layout_project_configs = layout.project_configs
for project_name, project_configs in layout_project_configs.items():
(trusted, project) = tenant.getProject(project_name)
project_queue_name = None
pipeline_queue_name = None
project_in_pipeline = False
for project_config in layout.getAllProjectConfigs(project_name):
project_pipeline_config = project_config.pipelines.get(
if project_pipeline_config is None:
project_in_pipeline = True
if not pipeline_queue_name:
pipeline_queue_name = project_pipeline_config.queue_name
if not project_queue_name:
project_queue_name = project_config.queue_name
if not project_in_pipeline:
# Note: we currently support queue name per pipeline and per
# project while project has precedence.
queue_name = project_queue_name or pipeline_queue_name
if not queue_name:
if queue_name in change_queues:
change_queue = change_queues[queue_name]
change_queue = []
change_queues[queue_name] = change_queue
self.log.debug("Created queue: %s" % queue_name)
self.log.debug("Added project %s to queue: %s" %
(project, queue_name))
def getSubmitAllowNeeds(self):
# Get a list of code review labels that are allowed to be
# "needed" in the submit records for a change, with respect
# to this queue. In other words, the list of review labels
# this queue itself is likely to set before submitting.
allow_needs = set()
for action_reporter in self.pipeline.success_actions:
return allow_needs
def eventMatches(self, event, change):
log = get_annotated_logger(self.log, event)
if event.forced_pipeline:
if event.forced_pipeline ==
log.debug("Event %s for change %s was directly assigned "
"to pipeline %s" % (event, change, self))
return True
return False
for ef in self.event_filters:
match_result = ef.matches(event, change)
if match_result:
log.debug("Event %s for change %s matched %s "
"in pipeline %s" % (event, change, ef, self))
return True
log.debug("Event %s for change %s does not match %s "
"in pipeline %s because %s" % (
event, change, ef, self, str(match_result)))
return False
def getNodePriority(self, item):
queue = self.pipeline.getRelativePriorityQueue(item.change.project)
items = self.pipeline.getAllItems()
items = [i for i in items
if i.change.project in queue and]
return items.index(item)
def isChangeAlreadyInPipeline(self, change):
# Checks live items in the pipeline
for item in self.pipeline.getAllItems():
if and change.equals(item.change):
return True
return False
def isChangeAlreadyInQueue(self, change, change_queue):
# Checks any item in the specified change queue
for item in change_queue.queue:
if change.equals(item.change):
return True
return False
def reportEnqueue(self, item):
if not self.pipeline._disabled:"Reporting enqueue, action %s item %s" %
(self.pipeline.enqueue_actions, item))
ret = self.sendReport(self.pipeline.enqueue_actions, item)
if ret:
self.log.error("Reporting item enqueued %s received: %s" %
(item, ret))
def reportStart(self, item):
if not self.pipeline._disabled:"Reporting start, action %s item %s" %
(self.pipeline.start_actions, item))
ret = self.sendReport(self.pipeline.start_actions, item)
if ret:
self.log.error("Reporting item start %s received: %s" %
(item, ret))
def reportDequeue(self, item):
if not self.pipeline._disabled:
"Reporting dequeue, action %s item%s",
ret = self.sendReport(self.pipeline.dequeue_actions, item)
if ret:
"Reporting item dequeue %s received: %s", item, ret
def sendReport(self, action_reporters, item, message=None):
"""Sends the built message off to configured reporters.
Takes the action_reporters, item, message and extra options and
sends them to the pluggable reporters.
log = get_annotated_logger(self.log, item.event)
report_errors = []
if len(action_reporters) > 0:
for reporter in action_reporters:
ret =
if ret:
except Exception as e:
log.exception("Exception while reporting")
return report_errors
def isChangeReadyToBeEnqueued(self, change, event):
return True
def enqueueChangesAhead(self, change, event, quiet, ignore_requirements,
change_queue, history=None, dependency_graph=None):
return True
def enqueueChangesBehind(self, change, event, quiet, ignore_requirements,
change_queue, history=None,
return True
def checkForChangesNeededBy(self, change, change_queue, event,
return True
def getFailingDependentItems(self, item):
return None
def getItemForChange(self, change, change_queue=None):
if change_queue is not None:
items = change_queue.queue
items = self.pipeline.getAllItems()
for item in items:
if item.change.equals(change):
return item
return None
def findOldVersionOfChangeAlreadyInQueue(self, change):
for item in self.pipeline.getAllItems():
if not
if change.isUpdateOf(item.change):
return item
return None
def removeOldVersionsOfChange(self, change, event):
if not self.pipeline.dequeue_on_new_patchset:
old_item = self.findOldVersionOfChangeAlreadyInQueue(change)
if old_item:
log = get_annotated_logger(self.log, event)
log.debug("Change %s is a new version of %s, removing %s",
change, old_item.change, old_item)
def removeAbandonedChange(self, change, event):
log = get_annotated_logger(self.log, event)
log.debug("Change %s abandoned, removing." % change)
for item in self.pipeline.getAllItems():
if not
if item.change.equals(change):
def getChangeQueue(self, change, event, existing=None):
def reEnqueueItem(self, item, last_head, old_item_ahead, item_ahead_valid):
log = get_annotated_logger(self.log, item.event)
with self.getChangeQueue(item.change, item.event,
last_head.queue) as change_queue:
if change_queue:
log.debug("Re-enqueing change %s in queue %s",
item.change, change_queue)
# If the old item ahead was re-enqued, this value will
# be true, so we should attempt to move the item back
# to where it was in case an item ahead is already
# failing.
if item_ahead_valid:
change_queue.moveItem(item, old_item_ahead)
# Get an updated copy of the layout and update the job
# graph if necessary. This resumes the buildset merge
# state machine. If we have an up-to-date layout, it
# will go ahead and refresh the job graph if needed;
# or it will send a new merge job if necessary, or it
# will do nothing if we're waiting on a merge job.
has_job_graph = bool(item.job_graph)
item.job_graph = None
item.layout = None
# If the item is no longer active, but has a job graph we
# will make sure to update it.
if or has_job_graph:
# Re-set build results in case any new jobs have been
# added to the tree.
for build in item.current_build_set.getBuilds():
if build.result:
# Similarly, reset the item state.
if item.current_build_set.unable_to_merge:
if item.current_build_set.config_errors:
if item.dequeued_needing_change:
# It can happen that all in-flight builds have been removed
# which would lead to paused parent jobs not being resumed.
# To prevent that resume parent jobs if necessary.
return True
log.error("Unable to find change queue for project %s",
return False
def addChange(self, change, event, quiet=False, enqueue_time=None,
ignore_requirements=False, live=True,
change_queue=None, history=None, dependency_graph=None):
log = get_annotated_logger(self.log, event)
log.debug("Considering adding change %s" % change)
history = history if history is not None else []
log.debug("History: %s", history)
# Ensure the dependency graph is created when the first change is
# processed to allow cycle detection with the Tarjan algorithm
dependency_graph = dependency_graph or collections.OrderedDict()
log.debug("Dependency graph: %s", dependency_graph)
# If we are adding a live change, check if it's a live item
# anywhere in the pipeline. Otherwise, we will perform the
# duplicate check below on the specific change_queue.
if live and self.isChangeAlreadyInPipeline(change):
log.debug("Change %s is already in pipeline, ignoring" % change)
return True
if not ignore_requirements:
for f in self.ref_filters:
if f.connection_name != change.project.connection_name:
log.debug("Filter %s skipped for change %s due "
"to mismatched connections" % (f, change))
match_result = f.matches(change)
if not match_result:
log.debug("Change %s does not match pipeline "
"requirement %s because %s" % (
change, f, str(match_result)))
return False
if not self.isChangeReadyToBeEnqueued(change, event):
log.debug("Change %s is not ready to be enqueued, ignoring" %
return False
with self.getChangeQueue(change, event, change_queue) as change_queue:
if not change_queue:
log.debug("Unable to find change queue for "
"change %s in project %s" %
(change, change.project))
return False
if not self.enqueueChangesAhead(change, event, quiet,
change_queue, history=history,
self.dequeueIncompleteCycle(change, dependency_graph, event,
log.debug("Failed to enqueue changes ahead of %s" % change)
return False
log.debug("History after enqueuing changes ahead: %s", history)
if self.isChangeAlreadyInQueue(change, change_queue):
log.debug("Change %s is already in queue, ignoring" % change)
return True
cycle = None
if hasattr(change, "needs_changes"):
cycle = self.cycleForChange(change, dependency_graph, event)
if cycle and not self.canProcessCycle(change.project):"Dequeing change %s since at least one project "
"does not allow circular dependencies", change)
actions = self.pipeline.failure_actions
ci = model.QueueItem(self, cycle[-1], event)
ci.warning("Dependency cycle detected")
# Only report the cycle if the project is in the current
# pipeline. Otherwise the change could be spammed by
# reports from unrelated pipelines.
if self.pipeline.tenant.layout.getProjectPipelineConfig(
self.sendReport(actions, ci)
return False"Adding change %s to queue %s in %s" %
(change, change_queue, self.pipeline))
item = change_queue.enqueueChange(change, event)
self.updateBundle(item, change_queue, cycle)
if enqueue_time:
item.enqueue_time = enqueue_time = live
item.quiet = quiet
if and not item.reported_enqueue:
item.reported_enqueue = True
# Items in a dependency cycle are expected to be enqueued after
# each other. To prevent non-cycle items from being enqueued
# between items of the same cycle, we skip that step when a cycle
# was found.
if not cycle:
self.enqueueChangesBehind(change, event, quiet,
ignore_requirements, change_queue,
history, dependency_graph)
self.log.debug("Skip enqueueing changes behind because of "
"dependency cycle")
zuul_driver = self.sched.connections.drivers['zuul']
tenant = self.pipeline.tenant
tenant, item.change, self.pipeline, event)
return True
def cycleForChange(self, change, dependency_graph, event):
log = get_annotated_logger(self.log, event)
log.debug("Running Tarjan's algorithm on current dependencies: %s",
sccs = [s for s in strongly_connected_components(dependency_graph)
if len(s) > 1]
log.debug("Strongly connected components (cyles): %s", sccs)
for scc in sccs:
if change in scc:
log.debug("Dependency cycle detected for "
"change %s in project %s",
change, change.project)
# Change can not be part of multiple cycles, so we can return
return scc
def canProcessCycle(self, project):
layout = self.pipeline.tenant.layout
pipeline_queue_name = None
project_queue_name = None
for project_config in layout.getAllProjectConfigs(
if not project_queue_name:
project_queue_name = project_config.queue_name
project_pipeline_config = project_config.pipelines.get(
if project_pipeline_config is None:
# TODO(simonw): Remove pipeline_queue_name after deprecation
if not pipeline_queue_name:
pipeline_queue_name = project_pipeline_config.queue_name
# Note: we currently support queue name per pipeline and per
# project while project has precedence.
queue_name = project_queue_name or pipeline_queue_name
if queue_name is None:
return False
queue_config = layout.queues.get(queue_name)
return (
queue_config is not None and
def updateBundle(self, item, change_queue, cycle):
if not cycle:
log = get_annotated_logger(self.log, item.event)
item.bundle = model.Bundle()
# Try to find already enqueued items of this cycle, so we use
# the same bundle
for needed_change in (c for c in cycle if c is not item.change):
needed_item = self.getItemForChange(needed_change, change_queue)
if not needed_item:
# Use a common bundle for the cycle
item.bundle = needed_item.bundle
break"Adding cycle item %s to bundle %s", item, item.bundle)
def dequeueIncompleteCycle(self, change, dependency_graph, event,
log = get_annotated_logger(self.log, event)
cycle = self.cycleForChange(change, dependency_graph, event) or []
enqueued_cycle_items = [i for i in (self.getItemForChange(c,
for c in cycle) if i is not None]
if enqueued_cycle_items:"Dequeuing incomplete cycle items: %s",
for cycle_item in enqueued_cycle_items:
def dequeueItem(self, item):
log = get_annotated_logger(self.log, item.event)
log.debug("Removing change %s from queue", item.change)
# In case a item is dequeued that doesn't have a result yet
# (success/failed/...) we report it as dequeued.
# Without this check, all items with a valid result would be reported
# twice.
if not item.current_build_set.result and
def removeItem(self, item):
log = get_annotated_logger(self.log, item.event)
# Remove an item from the queue, probably because it has been
# superseded by another change.
log.debug("Canceling builds behind change: %s "
"because it is being removed.", item.change)
if item.bundle is None:
log.debug("Dequeueing items in bundle %s", item.bundle)
bundle_iter = (i for i in item.bundle.items if i is not item)
for bundle_item in bundle_iter:
def dequeueSupercededItems(self, item):
for other_name in self.pipeline.supercedes:
other_pipeline = self.pipeline.tenant.layout.pipelines.get(
if not other_pipeline:
found = None
for other_item in other_pipeline.getAllItems():
if and other_item.change.equals(item.change):
found = other_item
if found:"Item %s is superceded by %s, removing" %
(found, item))
def updateCommitDependencies(self, change, change_queue, event):
log = get_annotated_logger(self.log, event)
# Search for Depends-On headers and find appropriate changes
log.debug(" Updating commit dependencies for %s", change)
change.refresh_deps = False
dependencies = []
seen = set()
for match in find_dependency_headers(change.message):
log.debug(" Found Depends-On header: %s", match)
if match in seen:
url = urllib.parse.urlparse(match)
except ValueError:
source = self.sched.connections.getSourceByHostname(
if not source:
log.debug(" Found source: %s", source)
dep = source.getChangeByURL(match, event)
if dep and (not dep.is_merged) and dep not in dependencies:
log.debug(" Adding dependency: %s", dep)
change.commit_needs_changes = dependencies
def provisionNodes(self, item):
log = item.annotateLogger(self.log)
jobs = item.findJobsToRequest(item.pipeline.tenant.semaphore_handler)
if not jobs:
return False
build_set = item.current_build_set
log.debug("Requesting nodes for change %s", item.change)
if self.sched.use_relative_priority:
priority = item.getNodePriority()
priority = 0
for job in jobs:
req = self.sched.nodepool.requestNodes(
build_set, job, priority, event=item.event)
log.debug("Adding node request %s for job %s to item %s",
req, job, item)
build_set.setJobNodeRequest(, req)
return True
def _executeJobs(self, item, jobs):
log = get_annotated_logger(self.log, item.event)
log.debug("Executing jobs for change %s", item.change)
build_set = item.current_build_set
for job in jobs:
log.debug("Found job %s for change %s", job, item.change)
nodeset = item.current_build_set.getJobNodeSet(
nodeset, build_set=item.current_build_set,
job, item, self.pipeline,
except Exception:
log.exception("Exception while executing job %s "
"for change %s:", job, item.change)
# If we hit an exception we don't have a build in the
# current item so a potentially aquired semaphore must be
# released as it won't be released on dequeue of the item.
tenant = item.pipeline.tenant
tenant.semaphore_handler.release(item, job)
except Exception:
log.exception("Exception while releasing semaphore")
def executeJobs(self, item):
# TODO(jeblair): This should return a value indicating a job
# was executed. Appears to be a longstanding bug.
if not item.layout:
return False
jobs = item.findJobsToRun(
if jobs:
self._executeJobs(item, jobs)
def cancelJobs(self, item, prime=True):
log = get_annotated_logger(self.log, item.event)
log.debug("Cancel jobs for change %s", item.change)
canceled = False
old_build_set = item.current_build_set
jobs_to_cancel = item.getJobs()
# Don't reset builds for a failing bundle when it has already started
# reporting, to keep available build results. Those items will be
# reported immediately afterwards during queue processing.
if (prime and item.current_build_set.ref and not
for job in jobs_to_cancel:
self.sched.cancelJob(old_build_set, job)
for item_behind in item.items_behind:
log.debug("Canceling jobs for change %s, behind change %s",
item_behind.change, item.change)
if self.cancelJobs(item_behind, prime=prime):
canceled = True
return canceled
def _findRelevantErrors(self, item, layout):
if item.item_ahead:
parent_layout = item.item_ahead.layout
parent_layout = item.pipeline.tenant.layout
relevant_errors = []
for err in layout.loading_errors.errors:
econtext = err.key.context
if ((err.key not in
parent_layout.loading_errors.error_keys) or
(econtext.project == and
econtext.branch == item.change.branch)):
return relevant_errors
def _loadDynamicLayout(self, item):
log = get_annotated_logger(self.log, item.event)
# Load layout
# Late import to break an import loop
import zuul.configloader
loader = zuul.configloader.ConfigLoader(
self.sched.connections, self.sched, None, None)
log.debug("Loading dynamic layout")
(trusted_updates, untrusted_updates) = item.includesConfigUpdates()
build_set = item.current_build_set
trusted_layout = None
trusted_errors = False
untrusted_layout = None
untrusted_errors = False
# First parse the config as it will land with the
# full set of config and project repos. This lets us
# catch syntax errors in config repos even though we won't
# actually run with that config.
if trusted_updates:
log.debug("Loading dynamic layout (phase 1)")
trusted_layout = loader.createDynamicLayout(
trusted_errors = len(trusted_layout.loading_errors) > 0
# Then create the config a second time but without changes
# to config repos so that we actually use this config.
if untrusted_updates:
log.debug("Loading dynamic layout (phase 2)")
untrusted_layout = loader.createDynamicLayout(
untrusted_errors = len(untrusted_layout.loading_errors) > 0
# Configuration state handling switchboard. Intentionally verbose
# and repetetive to be exceptionally clear that we handle all
# possible cases correctly. Note we never return trusted_layout
# from a dynamic update.
# No errors found at all use dynamic untrusted layout
if (trusted_layout and not trusted_errors and
untrusted_layout and not untrusted_errors):
log.debug("Loading dynamic layout complete")
return untrusted_layout
# No errors in untrusted only layout update
elif (not trusted_layout and
untrusted_layout and not untrusted_errors):
log.debug("Loading dynamic layout complete")
return untrusted_layout
# No errors in trusted only layout update
elif (not untrusted_layout and
trusted_layout and not trusted_errors):
# We're a change to a config repo (with no untrusted
# config items ahead), so just use the current pipeline
# layout.
log.debug("Loading dynamic layout complete")
return item.queue.pipeline.tenant.layout
# Untrusted layout only works with trusted updates
elif (trusted_layout and not trusted_errors and
untrusted_layout and untrusted_errors):"Configuration syntax error in dynamic layout")
# The config is good if we include config-projects,
# but is currently invalid if we omit them. Instead
# of returning the whole error message, just leave a
# note that the config will work once the dependent
# changes land.
msg = "This change depends on a change "\
"to a config project.\n\n"
msg += textwrap.fill(textwrap.dedent("""\
The syntax of the configuration in this change has
been verified to be correct once the config project
change upon which it depends is merged, but it can not
be used until that occurs."""))
return None
# Untrusted layout is broken and trusted is broken or not set
elif untrusted_layout and untrusted_errors:
# Find a layout loading error that match
# the current item.change and only report
# if one is found.
relevant_errors = self._findRelevantErrors(item,
if relevant_errors:
return None
"Configuration syntax error not related to "
"change context. Error won't be reported.")
return untrusted_layout
# Trusted layout is broken
elif trusted_layout and trusted_errors:
# Find a layout loading error that match
# the current item.change and only report
# if one is found.
relevant_errors = self._findRelevantErrors(item,
if relevant_errors:
return None
"Configuration syntax error not related to "
"change context. Error won't be reported.")
# We're a change to a config repo with errors not relevant
# to this repo. We use the pipeline layout.
return item.queue.pipeline.tenant.layout
raise Exception("We have reached a configuration error that is"
"not accounted for.")
except Exception:
log.exception("Error in dynamic layout")
item.setConfigError("Unknown configuration error")
return None
def getLayout(self, item):
if item.item_ahead:
fallback_layout = item.item_ahead.layout
if fallback_layout is None:
# We're probably waiting on a merge job for the item ahead.
return None
fallback_layout = item.pipeline.tenant.layout
# If the current change does not update the layout, use its parent.
# If the bundle doesn't update the config or the bundle updates the
# config but the current change's project is not part of the tenant
# (e.g. when dealing w/ cross-tenant cycles), use the parent layout.
if not (
item.change.updatesConfig(item.pipeline.tenant) or
and item.bundle.updatesConfig(item.pipeline.tenant)
and item.pipeline.tenant.getProject(
)[1] is not None
return fallback_layout
# Else this item updates the config,
# ask the merger for the result.
build_set = item.current_build_set
if build_set.merge_state != build_set.COMPLETE:
return None
if build_set.unable_to_merge:
return fallback_layout
self.log.debug("Preparing dynamic layout for: %s" % item.change)
return self._loadDynamicLayout(item)
def scheduleMerge(self, item, files=None, dirs=None):
log = item.annotateLogger(self.log)
log.debug("Scheduling merge for item %s (files: %s, dirs: %s)" %
(item, files, dirs))
build_set = item.current_build_set
build_set.merge_state = build_set.PENDING
# If the involved projects exclude unprotected branches we should also
# exclude them from the merge and repo state except the branch of the
# change that is tested.
tenant = item.pipeline.tenant
items = list(item.items_ahead) + [item]
if item.bundle:
projects = {
item.change.project for item in items
if tenant.getProject(item.change.project.canonical_name)[1]
if all(tenant.getExcludeUnprotectedBranches(project)
for project in projects):
branches = set()
# Add all protected branches of all involved projects
for project in projects:
# Additionally add all target branches of all involved items.
branches.update(item.change.branch for item in items
if hasattr(item.change, 'branch'))
branches = list(branches)
branches = None
if isinstance(item.change, model.Change):
item.current_build_set, files, dirs,
return False
def scheduleFilesChanges(self, item):
log = item.annotateLogger(self.log)
log.debug("Scheduling fileschanged for item %s", item)
build_set = item.current_build_set
build_set.files_state = build_set.PENDING
item.change.ref, item.change.branch, build_set=build_set,
return False
def prepareItem(self, item):
build_set = item.current_build_set
tenant = item.pipeline.tenant
# We always need to set the configuration of the item if it
# isn't already set.
tpc = tenant.project_configs.get(item.change.project.canonical_name)
if not build_set.ref:
# Next, if a change ahead has a broken config, then so does
# this one. Record that and don't do anything else.
if (item.item_ahead and item.item_ahead.current_build_set and
msg = "This change depends on a change "\
"with an invalid configuration.\n"
return False
# The next section starts between 0 and 2 remote merger
# operations in parallel as needed.
ready = True
# If the project is in this tenant, fetch missing files so we
# know if it updates the config.
if tpc:
if build_set.files_state == build_set.NEW:
ready = self.scheduleFilesChanges(item)
if build_set.files_state == build_set.PENDING:
ready = False
# If this change alters config or is live, schedule merge and
# build a layout.
# If we are dealing w/ a bundle and the bundle updates config we also
# have to merge since a config change in any of the bundle's items
# applies to all items. This is, unless the current item is not part
# of this tenant (e.g. cross-tenant cycle).
if build_set.merge_state == build_set.NEW:
if or item.change.updatesConfig(tenant) or (
item.bundle and
item.bundle.updatesConfig(tenant) and tpc is not None
ready = self.scheduleMerge(
files=(['zuul.yaml', '.zuul.yaml'] +
dirs=(['zuul.d', '.zuul.d'] +
if build_set.merge_state == build_set.PENDING:
ready = False
# If a merger op is outstanding, we're not ready.
if not ready:
return False
# With the merges done, we have the info needed to get a
# layout. This may return the pipeline layout, a layout from
# a change ahead, a newly generated layout for this change, or
# None if there was an error that makes the layout unusable.
# In the last case, it will have set the config_errors on this
# item, which may be picked up by the next itme.
if not item.layout:
item.layout = self.getLayout(item)
if not item.layout:
return False
# If the change can not be merged or has config errors, don't
# run jobs.
if build_set.unable_to_merge:
return False
if build_set.config_errors:
return False
# We don't need to build a job graph for a non-live item, we
# just need the layout.
if not
return False
# At this point we have a layout for the item, and it's live,
# so freeze the job graph.
log = item.annotateLogger(self.log)
if not item.job_graph:
log.debug("Freezing job graph for %s" % (item,))
except Exception as e:
# TODOv3(jeblair): nicify this exception as it will be reported
log.exception("Error freezing job graph for %s" % (item,))
item.setConfigError("Unable to freeze job graph: %s" %
return False
return True
def _processOneItem(self, item, nnfi):
log = item.annotateLogger(self.log)
changed = False
ready = False
dequeued = False
failing_reasons = [] # Reasons this item is failing
item_ahead = item.item_ahead
if item_ahead and (not
item_ahead = None
change_queue = item.queue
if self.checkForChangesNeededBy(item.change, change_queue,
item.event) is not True:
# It's not okay to enqueue this change, we should remove it."Dequeuing change %s because "
"it can no longer merge" % item.change)
if item.isBundleFailing():
except exceptions.MergeFailure:
return (True, nnfi)
actionable = change_queue.isActionable(item) = actionable
dep_items = self.getFailingDependentItems(item)
if dep_items:
failing_reasons.append('a needed change is failing')
self.cancelJobs(item, prime=False)
item_ahead_merged = False
if (item_ahead and
hasattr(item_ahead.change, 'is_merged') and
item_ahead_merged = True
if (item_ahead != nnfi and not item_ahead_merged):
# Our current base is different than what we expected,
# and it's not because our current base merged. Something
# ahead must have failed."Resetting builds for change %s because the "
"item ahead, %s, is not the nearest non-failing "
"item, %s" % (item.change, item_ahead, nnfi))
change_queue.moveItem(item, nnfi)
changed = True
if actionable:
ready = self.prepareItem(item)
# Starting jobs reporting should only be done once if there are
# jobs to run for this item.
if ready and len(self.pipeline.start_actions) > 0 \
and len( > 0 \
and not item.reported_start \
and not item.quiet:
item.reported_start = True
if item.current_build_set.unable_to_merge:
failing_reasons.append("it has a merge conflict")
if item.current_build_set.config_errors:
failing_reasons.append("it has an invalid configuration")
if ready and self.provisionNodes(item):
changed = True
if ready and item.bundle and item.didBundleFinish():
# Since the bundle finished we need to check if any item
# can report. If that's the case we need to process the
# queue again.
changed = changed or any(
i.item_ahead is None for i in item.bundle.items)
if ready and self.executeJobs(item):
changed = True
if item.hasAnyJobFailed():
failing_reasons.append("at least one job failed")
if (not and (not item.items_behind) and (not dequeued):
failing_reasons.append("is a non-live item with no items behind")
changed = dequeued = True
can_report = not item_ahead and item.areAllJobsComplete() and
if can_report and item.bundle:
can_report = can_report and (
item.isBundleFailing() or item.didBundleFinish()
item.bundle.started_reporting = can_report
if can_report:
except exceptions.MergeFailure:
failing_reasons.append("it did not merge")
for item_behind in item.items_behind:"Resetting builds for change %s because the "
"item ahead, %s, failed to merge" %
(item_behind.change, item))
if item.bundle and not item.isBundleFailing():
item.bundle.failed_reporting = True
changed = dequeued = True
elif not failing_reasons and
nnfi = item
item.current_build_set.failing_reasons = failing_reasons
if failing_reasons:
log.debug("%s is a failing item because %s" %
(item, failing_reasons))
if and not dequeued and self.sched.use_relative_priority:
priority = item.getNodePriority()
for node_request in item.current_build_set.node_requests.values():
if node_request.relative_priority != priority:
node_request, priority)
return (changed, nnfi)
def reportProcessedBundleItems(self, item):
"""Report failure to already reported bundle items.
In case we encounter e.g. a merge failure when we already successfully
reported some items, we need to go back and report again.
reported_items = [i for i in item.bundle.items if i.reported]
actions = self.pipeline.failure_actions
for ri in reported_items:
self.sendReport(actions, ri)
def processQueue(self):
# Do whatever needs to be done for each change in the queue
self.log.debug("Starting queue processor: %s" %
changed = False
for queue in self.pipeline.queues:
queue_changed = False
nnfi = None # Nearest non-failing item
for item in queue.queue[:]:
item_changed, nnfi = self._processOneItem(
item, nnfi)
if item_changed:
queue_changed = True
if queue_changed:
changed = True
status = ''
for item in queue.queue:
status += item.formatStatus()
if status:
self.log.debug("Queue %s status is now:\n %s" %
(, status))
self.log.debug("Finished queue processor: %s (changed: %s)" %
(, changed))
return changed
def onBuildStarted(self, build):
log = get_annotated_logger(self.log, build.zuul_event_id)
log.debug("Build %s started", build)
return True
def onBuildPaused(self, build):
log = get_annotated_logger(self.log, build.zuul_event_id)
item = build.build_set.item
log.debug("Build %s of %s paused", build, item.change)
# We need to resume builds because we could either have no children
# or have children that are already skipped.
return True
def _resumeBuilds(self, build_set):
Resumes all paused builds of a buildset that may be resumed.
jobgraph = build_set.item.job_graph
for build in build_set.builds.values():
if not build.paused:
# check if all child jobs are finished
child_builds = [build_set.builds.get( for x in
all_completed = True
for child_build in child_builds:
if not child_build or not child_build.result:
all_completed = False
if all_completed:
build.paused = False
def _resetDependentBuilds(self, build_set, build):
jobgraph = build_set.item.job_graph
for job in jobgraph.getDependentJobsRecursively(
self.sched.cancelJob(build_set, job)
build = build_set.getBuild(
if build:
# Re-set build results in case we resetted builds that were skipped
# not by this build/
for build in build_set.getBuilds():
if build.result:
def _cancelRunningBuilds(self, build_set):
item = build_set.item
for job in item.getJobs():
build = build_set.getBuild(
if not build or not build.result:
self.sched.cancelJob(build_set, job, final=True)
def onBuildCompleted(self, build):
log = get_annotated_logger(self.log, build.zuul_event_id)
item = build.build_set.item
log.debug("Build %s of %s completed" % (build, item.change))
item.pipeline.tenant.semaphore_handler.release(item, build.job)
if item.getJob( is None:"Build %s no longer in job graph for item %s",
build, item)
log.debug("Item %s status is now:\n %s", item, item.formatStatus())
if build.retry:
if build.build_set.getJobNodeSet(
# in case this was a paused build we need to retry all child jobs
self._resetDependentBuilds(build.build_set, build)
if (item.project_pipeline_config.fail_fast and
build.failed and and not build.retry):
# If fail-fast is set and the build is not successful
# cancel all remaining jobs.
log.debug("Build %s failed and fail-fast enabled, canceling "
"running builds", build)
return True
def onFilesChangesCompleted(self, event):
build_set = event.build_set
item = build_set.item
item.change.files = event.files
build_set.files_state = build_set.COMPLETE
def onMergeCompleted(self, event):
build_set = event.build_set
item = build_set.item
item.change.containing_branches = event.item_in_branches
build_set.merge_state = build_set.COMPLETE
build_set.repo_state = event.repo_state
if event.merged:
build_set.commit = event.commit
items_ahead = item.getNonLiveItemsAhead()
for index, item in enumerate(items_ahead):
files = item.current_build_set.files
files.setFiles(event.files[:index + 1])
elif event.updated:
build_set.commit = (item.change.newrev or
if not build_set.commit:"Unable to merge change %s" % item.change)
def onNodesProvisioned(self, event):
# TODOv3(jeblair): handle provisioning failure here
request = event.request
log = get_annotated_logger(self.log, request.event_id)
build_set = request.build_set
if request.failed or not request.fulfilled:"Node request %s: failure for %s",
tenant = build_set.item.pipeline.tenant
tenant.semaphore_handler.release(build_set.item, request.job)"Completed node request %s for job %s of item %s "
"with nodes %s",
request, request.job, build_set.item, request.nodeset)
def reportItem(self, item):
log = get_annotated_logger(self.log, item.event)
if not item.reported:
# _reportItem() returns True if it failed to report.
item.reported = not self._reportItem(item)
if self.changes_merge:
succeeded = item.didAllJobsSucceed() and not item.isBundleFailing()
merged = item.reported
source = item.change.project.source
if merged:
merged = source.isMerged(item.change, item.change.branch)
change_queue = item.queue
if not (succeeded and merged):
if not item.job_graph or not
error_reason = "did not have any jobs configured"
elif not succeeded:
error_reason = "failed tests"
error_reason = "failed to merge""Reported change %s did not merge because it %s,"
"status: all-succeeded: %s, merged: %s",
item.change, error_reason, succeeded, merged)
if not succeeded:
log.debug("%s window size decreased to %s",
change_queue, change_queue.window)
raise exceptions.MergeFailure(
"Change %s failed to merge" % item.change)
else:"Reported change %s status: all-succeeded: %s, "
"merged: %s", item.change, succeeded, merged)
log.debug("%s window size increased to %s",
change_queue, change_queue.window)
zuul_driver = self.sched.connections.drivers['zuul']
tenant = self.pipeline.tenant
zuul_driver.onChangeMerged(tenant, item.change, source)
def _reportItem(self, item):
log = get_annotated_logger(self.log, item.event)
log.debug("Reporting change %s", item.change)
ret = True # Means error as returned by
# In the case of failure, we may not hove completed an initial
# merge which would get the layout for this item, so in order
# to determine whether this item's project is in this
# pipeline, use the dynamic layout if available, otherwise,
# fall back to the current static layout as a best
# approximation.
layout = (item.layout or self.pipeline.tenant.layout)
project_in_pipeline = True
ppc = None
ppc = layout.getProjectPipelineConfig(item)
except Exception:
log.exception("Invalid config for change %s", item.change)
if not ppc:
log.debug("Project %s not in pipeline %s for change %s",
item.change.project, self.pipeline, item.change)
project_in_pipeline = False
actions = self.pipeline.no_jobs_actions
elif item.getConfigErrors():
log.debug("Invalid config for change %s", item.change)
# TODOv3(jeblair): consider a new reporter action for this
actions = self.pipeline.merge_failure_actions
elif item.didMergerFail():
log.debug("Merger failure")
actions = self.pipeline.merge_failure_actions
elif item.wasDequeuedNeedingChange():
log.debug("Dequeued needing change")
actions = self.pipeline.failure_actions
elif not item.getJobs():
# We don't send empty reports with +1
log.debug("No jobs for change %s", item.change)
actions = self.pipeline.no_jobs_actions
elif item.isBundleFailing():
log.debug("Bundle is failing")
actions = self.pipeline.failure_actions
if not item.didAllJobsSucceed():
self.pipeline._consecutive_failures += 1
elif item.didAllJobsSucceed() and not item.isBundleFailing():
log.debug("success %s", self.pipeline.success_actions)
actions = self.pipeline.success_actions
self.pipeline._consecutive_failures = 0
actions = self.pipeline.failure_actions
self.pipeline._consecutive_failures += 1
if project_in_pipeline and self.pipeline._disabled:
actions = self.pipeline.disabled_actions
# Check here if we should disable so that we only use the disabled
# reporters /after/ the last disable_at failure is still reported as
# normal.
if (self.pipeline.disable_at and not self.pipeline._disabled and
self.pipeline._consecutive_failures >= self.pipeline.disable_at):
self.pipeline._disabled = True
if actions:"Reporting item %s, actions: %s", item, actions)
ret = self.sendReport(actions, item)
if ret:
log.error("Reporting item %s received: %s", item, ret)
return ret
def reportStats(self, item):
if not self.sched.statsd:
# Update the gauge on enqueue and dequeue, but timers only
# when dequeing.
if item.dequeue_time:
dt = int((item.dequeue_time - item.enqueue_time) * 1000)
dt = None
items = len(self.pipeline.getAllItems())
tenant = self.pipeline.tenant
basekey = 'zuul.tenant.%s' %
key = '%s.pipeline.%s' % (basekey,
# stats.timers.zuul.tenant.<tenant>.pipeline.<pipeline>.resident_time
# stats_counts.zuul.tenant.<tenant>.pipeline.<pipeline>.total_changes
# stats.gauges.zuul.tenant.<tenant>.pipeline.<pipeline>.current_changes
self.sched.statsd.gauge(key + '.current_changes', items)
if dt:
self.sched.statsd.timing(key + '.resident_time', dt)
self.sched.statsd.incr(key + '.total_changes')
if hasattr(item.change, 'branch'):
hostname = (item.change.project.canonical_hostname.
replace('.', '_'))
projectname = (
replace('.', '_').replace('/', '.'))
projectname = projectname.replace('.', '_').replace('/', '.')
branchname = item.change.branch.replace('.', '_').replace(
'/', '.')
# stats.timers.zuul.tenant.<tenant>.pipeline.<pipeline>.
# project.<host>.<project>.<branch>.resident_time
# stats_counts.zuul.tenant.<tenant>.pipeline.<pipeline>.
# project.<host>.<project>.<branch>.total_changes
key += '.project.%s.%s.%s' % (hostname, projectname,
if dt:
self.sched.statsd.timing(key + '.resident_time', dt)
self.sched.statsd.incr(key + '.total_changes')
except Exception:
self.log.exception("Exception reporting pipeline stats")