# Copyright 2021-2024 Acme Gating, LLC # # Licensed under the Apache License, Version 2.0 (the "License"); you may # not use this file except in compliance with the License. You may obtain # a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the # License for the specific language governing permissions and limitations # under the License. import collections import contextlib import itertools import logging import textwrap import time 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 import zuul.lib.tracing as tracing from zuul.model import ( Change, PipelineState, PipelineChangeList, QueueItem, filter_severity, EnqueueEvent ) from zuul.zk.change_cache import ChangeKey from zuul.zk.exceptions import LockException from zuul.zk.locks import pipeline_lock from opentelemetry import trace 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: self.change_queue.pipeline.removeQueue(self.change_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): pass 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" % (pipeline.tenant.name, pipeline.name,)) self.sched = sched self.pipeline = pipeline self.event_filters = [] self.ref_filters = [] # Cached dynamic layouts (layout uuid -> layout) self._layout_cache = {} # A small local cache to avoid hitting the ZK-based connection # change cache for multiple hits in the same pipeline run. self._change_cache = {} # Current ZK context when the pipeline is locked self.current_context = None # The pipeline summary used by zuul-web that is updated by the # schedulers after processing a pipeline. self.pipeline.summary = model.PipelineSummary() self.pipeline.summary._set(pipeline=self.pipeline) if sched: self.sql = sched.sql def __str__(self): return "<%s %s>" % (self.__class__.__name__, self.pipeline.name) @contextlib.contextmanager def currentContext(self, ctx): try: self.current_context = ctx yield finally: self.current_context = None def _postConfig(self): layout = self.pipeline.tenant.layout self.buildChangeQueues(layout) # Make sure we have state and change list objects. We # don't actually ensure they exist in ZK here; these are # just local objects until they are serialized the first # time. Since we don't hold the pipeline lock, we can't # reliably perform any read or write operations; we just # need to ensure we have in-memory objects to work with # and they will be initialized or loaded on the next # refresh. # These will be out of date until they are refreshed later. self.pipeline.state = PipelineState.create( self.pipeline, self.pipeline.state) self.pipeline.change_list = PipelineChangeList.create( self.pipeline) # Now, try to acquire a non-blocking pipeline lock and refresh # them for the side effect of initializing them if necessary. # In the case of a new pipeline, no one else should have a # lock anyway, and this helps us avoid emitting a whole bunch # of errors elsewhere on startup when these objects don't # exist. If the pipeline already exists and we can't acquire # the lock, that's fine, we're much less likely to encounter # read errors elsewhere in that case anyway. try: with (pipeline_lock( self.sched.zk_client, self.pipeline.tenant.name, self.pipeline.name, blocking=False) as lock, self.sched.createZKContext(lock, self.log) as ctx, self.currentContext(ctx)): if not self.pipeline.state.exists(ctx): # We only do this if the pipeline doesn't exist in # ZK because in that case, this process should be # fast since it's empty. If it does exist, # refreshing it may be slow and since other actors # won't encounter errors due to its absence, we # would rather defer the work to later. self.pipeline.state.refresh(ctx) self.pipeline.change_list.refresh(ctx) except LockException: pass 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) queue_name = None project_in_pipeline = False for project_config in layout.getAllProjectConfigs(project_name): project_pipeline_config = project_config.pipelines.get( self.pipeline.name) if not queue_name: queue_name = project_config.queue_name if project_pipeline_config is None: continue project_in_pipeline = True if not project_in_pipeline: continue if not queue_name: continue if queue_name in change_queues: change_queue = change_queues[queue_name] else: change_queue = [] change_queues[queue_name] = change_queue self.log.debug("Created queue: %s" % queue_name) change_queue.append(project) 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: allow_needs.update(action_reporter.getSubmitAllowNeeds()) return allow_needs def eventMatches(self, event, change): log = get_annotated_logger(self.log, event) if event.forced_pipeline: if event.forced_pipeline == self.pipeline.name: log.debug("Event %s for change %s was directly assigned " "to pipeline %s" % (event, change, self)) return True else: 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 else: 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, change): queue_projects = set(self.pipeline.getRelativePriorityQueue( change.project)) items = [] for i in self.pipeline.getAllItems(): if not i.live: continue item_projects = set([ c.project for c in i.changes]) if item_projects.intersection(queue_projects): items.append(i) index = items.index(item) # Quantize on a logarithmic scale so that we don't constantly # needlessly adjust thousands of node requests. # If we're in the top 10 changes, return the accurate number. if index < 10: return index # After 10, batch then in groups of 10 (so items 10-19 are all # at node priority 10, 20-29 at 20, etc). if index < 100: return index // 10 * 10 # After 100, batch in groups of 100. return index // 100 * 100 def resolveChangeReferences(self, change_references): return self.resolveChangeKeys( [ChangeKey.fromReference(r) for r in change_references]) def resolveChangeKeys(self, change_keys): resolved_changes = [] for key in change_keys: change = self._change_cache.get(key.reference) if change is None: source = self.sched.connections.getSource(key.connection_name) change = source.getChange(key) if change is None: self.log.error("Unable to resolve change from key %s", key) if isinstance(change, model.Change): update_commit_dependencies = ( change.commit_needs_changes is None) update_topic_dependencies = ( change.topic_needs_changes is None and self.useDependenciesByTopic(change.project)) if (update_commit_dependencies or update_topic_dependencies): self.updateCommitDependencies(change, event=None) self._change_cache[change.cache_key] = change resolved_changes.append(change) return resolved_changes def clearCache(self): self._change_cache.clear() def _maintainCache(self): active_layout_uuids = set() referenced_change_keys = set() for item in self.pipeline.getAllItems(): if item.layout_uuid: active_layout_uuids.add(item.layout_uuid) for change in item.changes: if isinstance(change, model.Change): referenced_change_keys.update(change.getNeedsChanges( self.useDependenciesByTopic(change.project))) referenced_change_keys.update( change.getNeededByChanges()) # Clean up unused layouts in the cache unused_layouts = set(self._layout_cache.keys()) - active_layout_uuids if unused_layouts: self.log.debug("Removing unused layouts %s from cache", unused_layouts) for uid in unused_layouts: with contextlib.suppress(KeyError): del self._layout_cache[uid] # Clean up change cache unused_keys = set(self._change_cache.keys()) - referenced_change_keys for key in unused_keys: with contextlib.suppress(KeyError): del self._change_cache[key] def isChangeAlreadyInPipeline(self, change): # Checks live items in the pipeline for item in self.pipeline.getAllItems(): if not item.live: continue for c in item.changes: if change.equals(c): return True return False def isChangeRelevantToPipeline(self, change): # Checks if any version of the change or its deps matches any # item in the pipeline. for change_key in self.pipeline.change_list.getChangeKeys(): if change.cache_stat.key.isSameChange(change_key): return True if isinstance(change, model.Change): for dep_change_ref in change.getNeedsChanges( self.useDependenciesByTopic(change.project)): dep_change_key = ChangeKey.fromReference(dep_change_ref) if change.cache_stat.key.isSameChange(dep_change_key): return True return False def isChangeAlreadyInQueue(self, change, change_queue): # Checks any item in the specified change queue for item in change_queue.queue: for c in item.changes: if change.equals(c): return True return False def refreshDeps(self, change, event): if not isinstance(change, model.Change): return to_refresh = set() for item in self.pipeline.getAllItems(): for item_change in item.changes: if not isinstance(item_change, model.Change): continue if item_change.equals(change): to_refresh.add(item_change) for dep_change_ref in item_change.commit_needs_changes: dep_change_key = ChangeKey.fromReference(dep_change_ref) if dep_change_key.isSameChange(change.cache_stat.key): to_refresh.add(item_change) for existing_change in to_refresh: self.updateCommitDependencies(existing_change, event) def reportEnqueue(self, item): if not self.pipeline.state.disabled: self.log.info("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.state.disabled: self.log.info("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 reportNormalBuildsetEnd(self, build_set, action, final, result=None): # Report a buildset end if there are jobs or errors if ((build_set.job_graph and len(build_set.job_graph.job_uuids) > 0) or build_set.has_blocking_errors or build_set.unable_to_merge): self.sql.reportBuildsetEnd(build_set, action, final, result) def reportDequeue(self, item, quiet=False): if not (self.pipeline.state.disabled or quiet): self.log.info( "Reporting dequeue, action %s item%s", self.pipeline.dequeue_actions, item, ) ret = self.sendReport(self.pipeline.dequeue_actions, item) if ret: self.log.error( "Reporting item dequeue %s received: %s", item, ret ) # This might be called after canceljobs, which also sets a # non-final 'cancel' result. self.reportNormalBuildsetEnd(item.current_build_set, 'dequeue', final=False) def sendReport(self, action_reporters, item, phase1=True, phase2=True): """Sends the built message off to configured reporters. Takes the action_reporters and item 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: try: ret = reporter.report(item, phase1=phase1, phase2=phase2) if ret: for r in ret: if r: report_errors.append(r) except Exception as e: item.setReportedResult('ERROR') log.exception("Exception while reporting") report_errors.append(str(e)) return report_errors def areChangesReadyToBeEnqueued(self, changes, event): return True def enqueueChangesAhead(self, change, event, quiet, ignore_requirements, change_queue, history=None, dependency_graph=None, warnings=None): return True def enqueueChangesBehind(self, change, event, quiet, ignore_requirements, change_queue, history=None, dependency_graph=None): return True def getMissingNeededChanges(self, changes, change_queue, event, dependency_graph=None): """Check that all needed changes are ahead in the queue. Return a list of any that are missing. If it is not possible to correct the missing changes, "abort" will be true. :returns: (abort, needed_changes) """ return False, [] def getFailingDependentItems(self, item): return [] def getItemForChange(self, change, change_queue=None): if change_queue is not None: items = change_queue.queue else: items = self.pipeline.getAllItems() for item in items: for c in item.changes: if change.equals(c): return item return None def findOldVersionOfChangeAlreadyInQueue(self, change): # Return the item and the old version of the change for item in self.pipeline.getAllItems(): if not item.live: continue for item_change in item.changes: if change.isUpdateOf(item_change): return item, item_change return None, None def removeOldVersionsOfChange(self, change, event): if not self.pipeline.dequeue_on_new_patchset: return old_item, old_change = self.findOldVersionOfChangeAlreadyInQueue( change) if old_item: log = get_annotated_logger(self.log, event) log.debug("Change %s is a new version, removing %s", change, old_item) self.removeItem(old_item) if old_item.live: changes = old_item.changes[:] changes.remove(old_change) self.reEnqueueChanges(old_item, changes) 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 item.live: continue for item_change in item.changes: if item_change.equals(change): if len(item.changes) > 1: msg = ("Dependency cycle change " f"{change.url} abandoned.") item.setDequeuedNeedingChange(msg) try: self.reportItem(item) except exceptions.MergeFailure: pass self.removeItem(item) def reEnqueueChanges(self, item, changes): for change in changes: orig_ref = None if item.event: orig_ref = item.event.ref event = EnqueueEvent(self.pipeline.tenant.name, self.pipeline.name, change.project.canonical_hostname, change.project.name, change=change._id(), orig_ref=orig_ref) event.zuul_event_id = item.event.zuul_event_id self.sched.pipeline_management_events[ self.pipeline.tenant.name][self.pipeline.name].put(event) @abstractmethod def getChangeQueue(self, change, event, existing=None): pass def reEnqueueItem(self, item, last_head, old_item_ahead, item_ahead_valid): log = get_annotated_logger(self.log, item.event) with self.getChangeQueue(item.changes[0], item.event, last_head.queue) as change_queue: if change_queue: log.debug("Re-enqueing %s in queue %s", item, change_queue) change_queue.enqueueItem(item) # 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, but if we have a # job graph already, then keep it (our repo state and # jobs are frozen and will now only update if the item # ahead changes). 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 there # isn't one; 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.current_build_set.job_graph) if item.live: # Only reset the layout for live items as we don't need to # re-create the layout in independent pipelines. item.updateAttributes(self.current_context, layout_uuid=None) # If the item is no longer active, but has a job graph we # will make sure to update it. if item.active or has_job_graph: self.prepareItem(item) # 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: item.setResult(build) # Similarly, reset the item state. if item.current_build_set.unable_to_merge: item.setUnableToMerge() if item.current_build_set.has_blocking_errors: item.setConfigErrors(item.getConfigErrors()) if item.dequeued_needing_change: item.setDequeuedNeedingChange(item.dequeued_needing_change) if item.dequeued_missing_requirements: item.setDequeuedMissingRequirements() # 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. self._resumeBuilds(item.current_build_set) self.reportStats(item) return True else: log.error("Unable to find change queue for project %s", item.change[0].project) 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, skip_presence_check=False): 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) and not skip_presence_check): log.debug("Change %s is already in pipeline, ignoring" % change) return True if ((not self.pipeline.allow_other_connections) and (change.project.connection_name not in self.pipeline.connections)): log.debug("Change %s is not from a connection known to %s ", change, self.pipeline) return False try: self.getDependencyGraph(change, dependency_graph, event, update_deps=True) except exceptions.DependencyLimitExceededError: 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 cycle = [] if isinstance(change, model.Change): cycle = self.cycleForChange(change, dependency_graph, event) cycle = self.sortCycleByGitDepends(cycle) if not cycle: cycle = [change] if not ignore_requirements: for f in self.ref_filters: for cycle_change in cycle: if (f.connection_name != cycle_change.project.connection_name): log.debug("Filter %s skipped for change %s due " "to mismatched connections", f, cycle_change) continue match_result = f.matches(cycle_change) if not match_result: log.debug("Change %s does not match pipeline " "requirement %s because %s", cycle_change, f, str(match_result)) return False if not self.areChangesReadyToBeEnqueued(cycle, event): log.debug("Cycle %s is not ready to be enqueued, ignoring" % cycle) return False if len(cycle) > 1: for cycle_change in cycle: if not self.canProcessCycle(cycle_change.project): log.info("Dequeing change %s since the project " "does not allow circular dependencies", cycle_change) warnings = ["Dependency cycle detected and project " f"{cycle_change.project.name} " "doesn't allow circular dependencies"] self._reportNonEqueuedItem( change_queue, change, event, warnings) return False warnings = [] if not self.enqueueChangesAhead( cycle, event, quiet, ignore_requirements, change_queue, history=history, dependency_graph=dependency_graph, warnings=warnings): log.debug("Failed to enqueue changes ahead of %s" % change) if warnings: self._reportNonEqueuedItem(change_queue, change, event, warnings) return False log.debug("History after enqueuing changes ahead: %s", history) if self.isChangeAlreadyInQueue(change, change_queue): if not skip_presence_check: log.debug("Change %s is already in queue, ignoring", change) return True log.info("Adding %s to queue %s in %s" % (cycle, change_queue, self.pipeline)) if enqueue_time is None: enqueue_time = time.time() event_span = tracing.restoreSpanContext(event.span_context) link_attributes = {"rel": type(event).__name__} link = trace.Link(event_span.get_span_context(), attributes=link_attributes) span_info = tracing.startSavedSpan( 'QueueItem', start_time=enqueue_time, links=[link]) item = change_queue.enqueueChanges(cycle, event, span_info=span_info, enqueue_time=enqueue_time) with item.activeContext(self.current_context): if enqueue_time: item.enqueue_time = enqueue_time item.live = live self.reportStats(item, trigger_event=event) item.quiet = quiet if item.live: self.reportEnqueue(item) for c in cycle: self.enqueueChangesBehind(c, event, quiet, ignore_requirements, change_queue, history, dependency_graph) zuul_driver = self.sched.connections.drivers['zuul'] tenant = self.pipeline.tenant with trace.use_span(tracing.restoreSpan(item.span_info)): for c in item.changes: zuul_driver.onChangeEnqueued( tenant, c, self.pipeline, event) self.dequeueSupercededItems(item) return True def _reportNonEqueuedItem(self, change_queue, change, event, warnings): # Enqueue an item which otherwise can not be enqueued in order # to report a message to the user. actions = self.pipeline.failure_actions ci = change_queue.enqueueChanges([change], event) try: for w in warnings: ci.warning(w) ci.setReportedResult('FAILURE') # Only report the item 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( ci, change): self.sendReport(actions, ci) finally: # Ensure that the item is dequeued in any case. Otherwise we # might leak items caused by this temporary enqueue. self.dequeueItem(ci) # We don't use reportNormalBuildsetEnd here because we want to # report even with no jobs. self.sql.reportBuildsetEnd(ci.current_build_set, 'failure', final=True) def cycleForChange(self, change, dependency_graph, event, debug=True): if debug: log = get_annotated_logger(self.log, event) log.debug("Running Tarjan's algorithm on current dependencies: %s", dependency_graph) sccs = [s for s in strongly_connected_components(dependency_graph) if len(s) > 1] if debug: log.debug("Strongly connected components (cyles): %s", sccs) for scc in sccs: if change in scc: if debug: 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 return [] def sortCycleByGitDepends(self, cycle): new_cycle = [] cycle = list(cycle) while cycle: self._sortCycleByGitDepends(cycle[0], cycle, new_cycle) return new_cycle def _sortCycleByGitDepends(self, change, cycle, new_cycle): cycle.remove(change) for needed_change in self.resolveChangeReferences( change.git_needs_changes): if needed_change not in cycle: continue self._sortCycleByGitDepends(needed_change, cycle, new_cycle) new_cycle.append(change) def getCycleDependencies(self, change, dependency_graph, event): cycle = self.cycleForChange(change, dependency_graph, event) return set( itertools.chain.from_iterable( dependency_graph[c] for c in cycle if c != change) ) - set(cycle) def getDependencyGraph(self, change, dependency_graph, event, update_deps=False, history=None, quiet=False, indent=''): log = get_annotated_logger(self.log, event) if not quiet: log.debug("%sChecking for changes needed by %s:", indent, change) if self.pipeline.ignore_dependencies: return if not isinstance(change, model.Change): return if not change.getNeedsChanges( self.useDependenciesByTopic(change.project)): return if history is None: history = set() history.add(change) if update_deps: self.updateCommitDependencies(change, event) for needed_change in self.resolveChangeReferences( change.getNeedsChanges( self.useDependenciesByTopic(change.project))): if not quiet: log.debug("%sChange %s needs change %s:", indent, change, needed_change) if needed_change.is_merged: if not quiet: log.debug("%sNeeded change is merged", indent) continue if (self.pipeline.tenant.max_dependencies is not None and (len(dependency_graph) > self.pipeline.tenant.max_dependencies)): log.info("%sDependency graph for change %s is too large", indent, change) raise exceptions.DependencyLimitExceededError( "Dependency graph is too large") node = dependency_graph.setdefault(change, []) if needed_change not in node: if not quiet: log.debug("%sAdding change %s to dependency graph for " "change %s", indent, needed_change, change) node.append(needed_change) if needed_change not in history: self.getDependencyGraph(needed_change, dependency_graph, event, update_deps, history, quiet, indent + ' ') def getQueueConfig(self, project): layout = self.pipeline.tenant.layout queue_name = None for project_config in layout.getAllProjectConfigs( project.canonical_name ): if not queue_name: queue_name = project_config.queue_name project_pipeline_config = project_config.pipelines.get( self.pipeline.name) if project_pipeline_config is None: continue return layout.queues.get(queue_name) def canProcessCycle(self, project): queue_config = self.getQueueConfig(project) if queue_config is None: return False return queue_config.allow_circular_dependencies def useDependenciesByTopic(self, project): queue_config = self.getQueueConfig(project) if queue_config is None: return False return queue_config.dependencies_by_topic def getNonMergeableCycleChanges(self, item): """Return changes in the cycle that do not fulfill the pipeline's ready criteria.""" return [] def dequeueItem(self, item, quiet=False): log = get_annotated_logger(self.log, item.event) log.debug("Removing %s from queue", item) # 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 item.live: item.setReportedResult('DEQUEUED') if not item.reported_start: # If we haven't reported start, we don't know if Zuul # was supposed to act on the item at all, so keep it # quiet. quiet = True self.reportDequeue(item, quiet) item.queue.dequeueItem(item) span_attrs = { 'zuul_event_id': item.event.zuul_event_id, 'zuul_tenant': self.pipeline.tenant.name, 'zuul_pipeline': self.pipeline.name, } for change in item.changes: for k, v in change.getSafeAttributes().toDict().items(): span_attrs.setdefault(f'ref_{k}', []).append(v) tracing.endSavedSpan(item.current_build_set.span_info) tracing.endSavedSpan(item.span_info, attributes=span_attrs) 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 item: %s " "because it is being removed.", item) self.cancelJobs(item) self.dequeueItem(item) self.reportStats(item) def dequeueSupercededItems(self, item): for other_name in self.pipeline.supercedes: other_pipeline = self.pipeline.tenant.layout.pipelines.get( other_name) if not other_pipeline: continue for change in item.changes: change_id = ( change._id() if isinstance(change, Change) else None ) event = model.SupercedeEvent( other_pipeline.tenant.name, other_pipeline.name, change.project.canonical_hostname, change.project.name, change_id, change.ref) self.sched.pipeline_trigger_events[ self.pipeline.tenant.name][other_pipeline.name ].put_supercede(event) def updateCommitDependencies(self, change, event): log = get_annotated_logger(self.log, event) must_update_commit_deps = ( not hasattr(event, "zuul_event_ltime") or change.commit_needs_changes is None or change.cache_stat.mzxid <= event.zuul_event_ltime ) must_update_topic_deps = ( self.useDependenciesByTopic(change.project) and ( not hasattr(event, "zuul_event_ltime") or change.topic_needs_changes is None or change.cache_stat.mzxid <= event.zuul_event_ltime ) ) update_attrs = {} if must_update_commit_deps: # Search for Depends-On headers and find appropriate changes log.debug(" Updating commit dependencies for %s", change) dependencies = [] seen = set() for match in find_dependency_headers(change.message): log.debug(" Found Depends-On header: %s", match) if match in seen: continue seen.add(match) try: url = urllib.parse.urlparse(match) except ValueError: continue source = self.sched.connections.getSourceByHostname( url.hostname) if not source: continue log.debug(" Found source: %s", source) dep = source.getChangeByURLWithRetry(match, event) if dep and (not dep.is_merged) and dep not in dependencies: log.debug(" Adding dependency: %s", dep) dependencies.append(dep) new_commit_needs_changes = [d.cache_key for d in dependencies] update_attrs = dict(commit_needs_changes=new_commit_needs_changes) # Ask the source for any tenant-specific changes (this allows # drivers to implement their own way of collecting deps): source = self.sched.connections.getSource( change.project.connection_name) if must_update_topic_deps: log.debug(" Updating topic dependencies for %s", change) new_topic_needs_changes = [] for dep in source.getChangesByTopic(change.topic): if dep and (not dep.is_merged): log.debug(" Adding dependency: %s", dep) new_topic_needs_changes.append(dep.cache_key) update_attrs['topic_needs_changes'] = new_topic_needs_changes if update_attrs: source.setChangeAttributes(change, **update_attrs) 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 %s", item) parent_span = tracing.restoreSpan(build_set.span_info) with trace.use_span(parent_span): for job in jobs: if self.sched.globals.use_relative_priority: relative_priority = self.getNodePriority( item, item.getChangeForJob(job)) else: relative_priority = 0 self._makeNodepoolRequest( log, build_set, job, relative_priority) return True def _makeNodepoolRequest(self, log, build_set, job, relative_priority, alternative=0): provider = self._getPausedParentProvider(build_set, job) priority = self._calculateNodeRequestPriority(build_set, job) tenant_name = build_set.item.pipeline.tenant.name pipeline_name = build_set.item.pipeline.name item = build_set.item req = self.sched.nodepool.requestNodes( build_set.uuid, job, tenant_name, pipeline_name, provider, priority, relative_priority, event=item.event) log.debug("Adding node request %s for job %s to item %s", req, job, item) build_set.setJobNodeRequestID(job, req.id) if req.fulfilled: nodeset = self.sched.nodepool.getNodeSet(req, job.nodeset) job = build_set.item.getJob(req.job_uuid) build_set.jobNodeRequestComplete(job, nodeset) else: job.setWaitingStatus(f'node request: {req.id}') def _getPausedParent(self, build_set, job): job_graph = build_set.job_graph if job_graph: for parent in job_graph.getParentJobsRecursively(job): build = build_set.getBuild(parent) if build.paused: return build return None def _getPausedParentProvider(self, build_set, job): parent_build = self._getPausedParent(build_set, job) if parent_build: return build_set.getJobNodeProvider(parent_build.job) return None def _calculateNodeRequestPriority(self, build_set, job): precedence_adjustment = 0 precedence = build_set.item.pipeline.precedence if self._getPausedParent(build_set, job): precedence_adjustment = -1 initial_precedence = model.PRIORITY_MAP[precedence] return max(0, initial_precedence + precedence_adjustment) def _executeJobs(self, item, jobs): log = get_annotated_logger(self.log, item.event) log.debug("Executing jobs for %s", item) build_set = item.current_build_set for job in jobs: log.debug("Found job %s for %s", job, item) try: zone = build_set.getJobNodeExecutorZone(job) nodes = build_set.getJobNodeList(job) self.sched.executor.execute( job, nodes, item, self.pipeline, zone, build_set.dependent_changes, build_set.merger_items) job.setWaitingStatus('executor') except Exception: log.exception("Exception while executing job %s " "for %s:", job, item) try: # 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 pipeline = build_set.item.pipeline event_queue = self.sched.pipeline_result_events[ tenant.name][pipeline.name] tenant.semaphore_handler.release(event_queue, 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.current_build_set.job_graph: return False jobs = item.findJobsToRun( item.pipeline.tenant.semaphore_handler) 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 %s", item) canceled = False old_build_set = item.current_build_set jobs_to_cancel = item.getJobs() for job in jobs_to_cancel: self.sched.cancelJob(old_build_set, job, final=True) if (prime and item.current_build_set.ref): # Force a dequeued result here because we haven't actually # reported the item, but we are done with this buildset. self.reportNormalBuildsetEnd( item.current_build_set, 'dequeue', final=False, result='DEQUEUED') tracing.endSavedSpan(item.current_build_set.span_info) item.resetAllBuilds() for item_behind in item.items_behind: log.debug("Canceling jobs for %s, behind %s", item_behind, item) if self.cancelJobs(item_behind, prime=prime): canceled = True return canceled def _findRelevantErrors(self, item, layout): # First collect all the config errors that are not related to the # current item. parent_error_keys = list( self.pipeline.tenant.layout.loading_errors.error_keys) for item_ahead in item.items_ahead: parent_error_keys.extend( e.key for e in item.item_ahead.getConfigErrors()) # Then find config errors which aren't in the parent. But # include errors in this project-branch because the error # detection hash is imperfect and someone attempting to fix an # error may create a near duplicate error and it would go # undetected. Or if there are two errors and the user only # fixes one, then they may not realize their work is # incomplete. relevant_errors = [] severity_error = False for err in layout.loading_errors.errors: econtext = err.key.context matches_project_branch = False if econtext: for change in item.changes: if (econtext.project_name == change.project.name and econtext.branch == change.branch): matches_project_branch = True break if (err.key not in parent_error_keys or matches_project_branch): relevant_errors.append(err) if err.severity == model.SEVERITY_ERROR: severity_error = True return relevant_errors, severity_error 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.zk_client, self.sched.globals, self.sched.unparsed_config_cache, self.sched.statsd, self.sched) 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 additional_project_branches = self._getProjectBranches(item) try: # 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( item, build_set.files, additional_project_branches, self.sched.ansible_manager, include_config_projects=True, zuul_event_id=None) trusted_errors = len(filter_severity( trusted_layout.loading_errors.errors, errors=True, warnings=False)) > 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( item, build_set.files, additional_project_branches, self.sched.ansible_manager, include_config_projects=False, zuul_event_id=None) untrusted_errors = len(filter_severity( untrusted_layout.loading_errors.errors, errors=True, warnings=False)) > 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): log.info("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.""")) item.setConfigError(msg) 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.changes and only report # if one is found. relevant_errors, severity_error = self._findRelevantErrors( item, untrusted_layout) if relevant_errors: item.setConfigErrors(relevant_errors) if severity_error: return None log.info( "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.changes and only report # if one is found. relevant_errors, severity_error = self._findRelevantErrors( item, trusted_layout) if relevant_errors: item.setConfigErrors(relevant_errors) if severity_error: return None log.info( "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 else: 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 _getProjectBranches(self, item): # Get the branches of the item and all items ahead project_branches = {} while item: for change in item.changes: if hasattr(change, 'branch'): this_project_branches = project_branches.setdefault( change.project.canonical_name, set()) this_project_branches.add(change.branch) item = item.item_ahead return project_branches def getFallbackLayout(self, item): parent_item = item.item_ahead if not parent_item: return item.pipeline.tenant.layout return self.getLayout(parent_item) def getLayout(self, item): log = get_annotated_logger(self.log, item.event) layout = self._layout_cache.get(item.layout_uuid) if layout: log.debug("Using cached layout %s for item %s", layout.uuid, item) return layout if item.layout_uuid: log.debug("Re-calculating layout for item %s", item) layout = self._getLayout(item) if layout: item.updateAttributes(self.current_context, layout_uuid=layout.uuid) self._layout_cache[item.layout_uuid] = layout return layout def _getLayout(self, item): log = get_annotated_logger(self.log, item.event) if item.item_ahead: if ( (item.item_ahead.live and not item.item_ahead.current_build_set.job_graph) or (not item.item_ahead.live and not item.item_ahead.layout_uuid) ): # We're probably waiting on a merge job for the item ahead. return None # If the current item does not update the layout, use its parent. if not item.updatesConfig(): return self.getFallbackLayout(item) # 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 self.getFallbackLayout(item) log.debug("Preparing dynamic layout for: %s" % item) start = time.time() layout = self._loadDynamicLayout(item) self.reportPipelineTiming('layout_generation_time', start) return layout def _branchesForRepoState(self, projects, tenant, items=None): items = items or [] if all(tenant.getExcludeUnprotectedBranches(project) for project in projects): branches = set() # Add all protected branches of all involved projects for project in projects: branches.update( tenant.getProjectBranches(project.canonical_name)) # Additionally add all target branches of all involved items. branches.update(change.branch for item in items for change in item.changes if hasattr(change, 'branch')) # Make sure override-checkout targets are part of the repo state for item in items: if not item.current_build_set.job_graph: continue for job in item.current_build_set.job_graph.getJobs(): if job.override_checkout: branches.add(job.override_checkout) for p in job.required_projects.values(): if p.override_checkout: branches.add(p.override_checkout) branches = list(branches) else: branches = None return branches def scheduleMerge(self, item, files=None, dirs=None): log = item.annotateLogger(self.log) build_set = item.current_build_set log.debug("Scheduling merge for item %s (files: %s, dirs: %s)" % (item, files, dirs)) # 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] projects = { change.project for i in items for change in i.changes if tenant.getProject(change.project.canonical_name)[1] } branches = self._branchesForRepoState(projects=projects, tenant=tenant, items=items) # Using the first change as representative of whether this # pipeline is handling changes or refs if isinstance(item.changes[0], model.Change): self.sched.merger.mergeChanges(build_set.merger_items, build_set, files, dirs, precedence=self.pipeline.precedence, event=item.event, branches=branches) else: self.sched.merger.getRepoState(build_set.merger_items, build_set, precedence=self.pipeline.precedence, event=item.event, branches=branches) build_set.updateAttributes(self.current_context, merge_state=build_set.PENDING) 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 self.sched.merger.getFilesChanges(item.changes, build_set=build_set, event=item.event) build_set.updateAttributes(self.current_context, files_state=build_set.PENDING) return False def scheduleGlobalRepoState(self, item: QueueItem) -> bool: log = item.annotateLogger(self.log) tenant = item.pipeline.tenant jobs = item.current_build_set.job_graph.getJobs() project_cnames = set() for job in jobs: log.debug('Processing job %s', job.name) project_cnames.update(job.affected_projects) log.debug('Needed projects: %s', project_cnames) # Filter projects for ones that are already in repo state repo_state = item.current_build_set.repo_state connections = self.sched.connections.connections for connection in repo_state.keys(): canonical_hostname = connections[connection].canonical_hostname for project in repo_state[connection].keys(): canonical_project_name = canonical_hostname + '/' + project project_cnames.discard(canonical_project_name) if not project_cnames: item.current_build_set.updateAttributes( self.current_context, repo_state_state=item.current_build_set.COMPLETE) return True log.info('Scheduling global repo state for item %s', item) # At this point we know we're going to request a merge job; # set the waiting state on all the item's jobs so users know # what we're waiting on. for job in jobs: job.setWaitingStatus('repo state') projects = [] for project_cname in project_cnames: projects.append(tenant.getProject(project_cname)[1]) branches = self._branchesForRepoState( projects=projects, tenant=tenant, items=[item]) new_items = list() build_set = item.current_build_set # If we skipped the initial repo state (for branch/ref items), # we need to include the merger items for the final repo state. if build_set.merge_repo_state is None: new_items.extend(build_set.merger_items) for project in projects: new_item = dict() new_item['project'] = project.name new_item['connection'] = project.connection_name new_items.append(new_item) # Get state for not yet tracked projects self.sched.merger.getRepoState(items=new_items, build_set=item.current_build_set, event=item.event, branches=branches) item.current_build_set.updateAttributes( self.current_context, repo_state_request_time=time.time(), repo_state_state=item.current_build_set.PENDING) return True 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. if not build_set.ref: build_set.setConfiguration(self.current_context) # 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 item.item_ahead.current_build_set.has_blocking_errors): msg = "This change depends on a change "\ "with an invalid configuration.\n" item.setConfigError(msg) # Find our layout since the reporter will need it to # determine if the project is in the pipeline. self.getLayout(item) 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. in_tenant = False for c in item.changes: if tenant.project_configs.get(c.project.canonical_name): in_tenant = True break if in_tenant: 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 build_set.merge_state == build_set.NEW: if item.live or item.updatesConfig(): # Collect extra config files and dirs of required changes. extra_config_files = set() extra_config_dirs = set() for merger_item in item.current_build_set.merger_items: source = self.sched.connections.getSource( merger_item["connection"]) project = source.getProject(merger_item["project"]) tpc = tenant.project_configs.get(project.canonical_name) if tpc: extra_config_files.update(tpc.extra_config_files) extra_config_dirs.update(tpc.extra_config_dirs) ready = self.scheduleMerge( item, files=(['zuul.yaml', '.zuul.yaml'] + list(extra_config_files)), dirs=(['zuul.d', '.zuul.d'] + list(extra_config_dirs))) 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 # If the change can not be merged or has config errors, don't # run jobs. if build_set.unable_to_merge or build_set.has_blocking_errors: # Find our layout since the reporter will need it to # determine if the project is in the pipeline. self.getLayout(item) 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 item. if not (item.layout_uuid or item.current_build_set.job_graph): layout = self.getLayout(item) if not layout: return False # We don't need to build a job graph for a non-live item, we # just need the layout. if not item.live: 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.current_build_set.job_graph: try: log.debug("Freezing job graph for %s" % (item,)) start = time.time() item.freezeJobGraph(self.getLayout(item), self.current_context, skip_file_matcher=False, redact_secrets_and_keys=False) self.reportPipelineTiming('job_freeze_time', start) 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" % (str(e))) return False if (item.current_build_set.job_graph and len(item.current_build_set.job_graph.job_uuids) > 0): self.sql.reportBuildsetStart(build_set) # At this point we know all frozen jobs and their repos so update the # repo state with all missing repos. if build_set.repo_state_state == build_set.NEW: self.scheduleGlobalRepoState(item) if build_set.repo_state_state == build_set.PENDING: 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.live): item_ahead = None change_queue = item.queue meets_reqs = self.areChangesReadyToBeEnqueued(item.changes, item.event) dependency_graph = collections.OrderedDict() try: self.getDependencyGraph(item.changes[0], dependency_graph, item.event, quiet=True) except exceptions.DependencyLimitExceededError: self.removeItem(item) return True, nnfi # Verify that the cycle dependency graph is correct cycle = self.cycleForChange( item.changes[0], dependency_graph, item.event, debug=False) cycle = set(cycle or [item.changes[0]]) item_cycle = set(item.changes) # We don't consider merged dependencies when building the # dependency graph, so we need to ignore differences resulting # from changes that have been merged in the meantime. Put any # missing merged changes back in the cycle for comparison # purposes. merged_changes = set(c for c in (item_cycle - cycle) if c.is_merged) cycle |= merged_changes if cycle != item_cycle: log.info("Item cycle has changed: %s, now: %s, was: %s", item, cycle, item_cycle) self.removeItem(item) if item.live: self.reEnqueueChanges(item, item.changes) return (True, nnfi) abort, needs_changes = self.getMissingNeededChanges( item.changes, change_queue, item.event, dependency_graph=dependency_graph) if not (meets_reqs and not needs_changes): # It's not okay to enqueue this change, we should remove it. log.info("Dequeuing %s because " "it can no longer merge" % item) self.cancelJobs(item) if not meets_reqs: item.setDequeuedMissingRequirements() else: clist = ', '.join([c.url for c in needs_changes]) if len(needs_changes) > 1: msg = f'Changes {clist} are needed.' else: msg = f'Change {clist} is needed.' item.setDequeuedNeedingChange(msg) if item.live: try: self.reportItem(item) except exceptions.MergeFailure: pass self.dequeueItem(item) return (True, nnfi) actionable = change_queue.isActionable(item) item.updateAttributes(self.current_context, active=actionable) dep_items = self.getFailingDependentItems(item) if dep_items: failing_reasons.append('a needed change is failing') self.cancelJobs(item, prime=False) else: item_ahead_merged = (item_ahead.areAllChangesMerged() if item_ahead else False) 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. log.info("Resetting builds for changes %s because the " "item ahead, %s, is not the nearest non-failing " "item, %s" % (item.changes, item_ahead, nnfi)) change_queue.moveItem(item, nnfi) changed = True self.cancelJobs(item) 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(item.current_build_set.job_graph.job_uuids) > 0 and not item.reported_start and not item.quiet ): self.reportStart(item) item.updateAttributes(self.current_context, reported_start=True) if item.current_build_set.unable_to_merge: failing_reasons.append("it has a merge conflict") if item.current_build_set.has_blocking_errors: failing_reasons.append("it has an invalid configuration") if ready and self.provisionNodes(item): changed = True if ready and self.executeJobs(item): changed = True if item.hasAnyJobFailed(): failing_reasons.append("at least one job failed") if (not item.live) and (not item.items_behind) and (not dequeued): failing_reasons.append("is a non-live item with no items behind") self.dequeueItem(item) changed = dequeued = True can_report = not item_ahead and item.areAllJobsComplete() and item.live is_cycle = len(item.changes) > 1 if can_report and is_cycle: # Before starting to merge the cycle items, make sure they # can still be merged, to reduce the chance of a partial merge. non_mergeable_cycle_changes = self.getNonMergeableCycleChanges( item) if non_mergeable_cycle_changes: clist = ', '.join([ c.url for c in non_mergeable_cycle_changes]) if len(non_mergeable_cycle_changes) > 1: msg = f'Changes {clist} can not be merged.' else: msg = f'Change {clist} can not be merged.' item.setDequeuedNeedingChange(msg) failing_reasons.append("cycle can not be merged") log.debug( "Dequeuing item %s because cycle can no longer merge", item ) try: self.reportItem(item) except exceptions.MergeFailure: pass for item_behind in item.items_behind: log.info("Resetting builds for %s because the " "item ahead, %s, can not be merged" % (item_behind, item)) self.cancelJobs(item_behind) self.dequeueItem(item) return (True, nnfi) if can_report: succeeded = item.didAllJobsSucceed() try: self.reportItem(item) except exceptions.MergeFailure: failing_reasons.append("it did not merge") for item_behind in item.items_behind: log.info("Resetting builds for %s because the " "item ahead, %s, failed to merge" % (item_behind, item)) self.cancelJobs(item_behind) # Only re-report items in the cycle when we encounter a merge # failure for a successful cycle. if is_cycle and succeeded: self.sendReport(self.pipeline.failure_actions, item) self.dequeueItem(item) changed = dequeued = True elif not failing_reasons and item.live: nnfi = item if not dequeued: item.current_build_set.updateAttributes( self.current_context, failing_reasons=failing_reasons) if failing_reasons: log.debug("%s is a failing item because %s" % (item, failing_reasons)) if (item.live and not dequeued and self.sched.globals.use_relative_priority): for job, request_id in \ item.current_build_set.getNodeRequests(): node_request = self.sched.nodepool.zk_nodepool.getNodeRequest( request_id, cached=True) if not node_request: continue if node_request.state != model.STATE_REQUESTED: # If the node request was locked and accepted by a # provider, we can no longer update the relative priority. continue priority = self.getNodePriority( item, item.getChangeForJob(job)) if node_request.relative_priority != priority: self.sched.nodepool.reviseRequest( node_request, priority) return (changed, nnfi) def processQueue(self): # Do whatever needs to be done for each change in the queue self.log.debug("Starting queue processor: %s" % self.pipeline.name) changed = False change_keys = set() 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 self.reportStats(item) for change in item.changes: change_keys.add(change.cache_stat.key) 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" % (queue.name, status)) self.pipeline.change_list.setChangeKeys( self.pipeline.manager.current_context, change_keys) self._maintainCache() self.log.debug("Finished queue processor: %s (changed: %s)" % (self.pipeline.name, changed)) return changed def onBuildStarted(self, build): log = get_annotated_logger(self.log, build.zuul_event_id) log.debug("Build %s started", build) self.sql.reportBuildStart(build) self.reportPipelineTiming('job_wait_time', build.execute_time, build.start_time) if not build.build_set.item.first_job_start_time: # Only report this for the first job in a queue item so # that we don't include gate resets. build.build_set.item.updateAttributes( self.current_context, first_job_start_time=build.start_time) self.reportPipelineTiming('event_job_time', build.build_set.item.event.timestamp, build.start_time) 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) item.setResult(build) # We need to resume builds because we could either have no children # or have children that are already skipped. self._resumeBuilds(build.build_set) return True def _resumeBuilds(self, build_set): """ Resumes all paused builds of a buildset that may be resumed. """ for build in build_set.getBuilds(): if not build.paused: continue # check if all child jobs are finished child_builds = [] item = build.build_set.item job_graph = item.current_build_set.job_graph child_builds += [ item.current_build_set.getBuild(x) for x in job_graph.getDependentJobsRecursively( build.job)] all_completed = True for child_build in child_builds: if not child_build or not child_build.result: all_completed = False break if all_completed: self.sched.executor.resumeBuild(build) with build.activeContext(self.current_context): build.paused = False build.addEvent( model.BuildEvent( event_time=time.time(), event_type=model.BuildEvent.TYPE_RESUMED)) def _resetDependentBuilds(self, build_set, build): job_graph = build_set.job_graph for job in job_graph.getDependentJobsRecursively(build.job): self.sched.cancelJob(build_set, job) build = build_set.getBuild(job) if build: build_set.removeBuild(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: build_set.item.setResult(build) def _cancelRunningBuilds(self, build_set): item = build_set.item for job in item.getJobs(): build = build_set.getBuild(job) 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) event_queue = self.sched.pipeline_result_events[ item.pipeline.tenant.name][item.pipeline.name] item.pipeline.tenant.semaphore_handler.release( event_queue, item, build.job) if item.getJob(build.job.uuid) is None: log.info("Build %s no longer in job graph for item %s", build, item) return item.setResult(build) log.debug("Item %s status is now:\n %s", item, item.formatStatus()) build_set = build.build_set if build.retry: if build_set.getJobNodeSetInfo(build.job): build_set.removeJobNodeSetInfo(build.job) # in case this was a paused build we need to retry all # child jobs self._resetDependentBuilds(build_set, build) self._resumeBuilds(build_set) if (build_set.fail_fast and build.failed and build.job.voting 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) self._cancelRunningBuilds(build_set) return True def onFilesChangesCompleted(self, event, build_set): item = build_set.item for i, change in enumerate(item.changes): source = self.sched.connections.getSource( change.project.connection_name) if event.files: change_files = event.files[i] else: change_files = None source.setChangeAttributes(change, files=change_files) build_set.updateAttributes(self.current_context, files_state=build_set.COMPLETE) if build_set.merge_state == build_set.COMPLETE: # We're the second of the files/merger pair, report the stat self.reportPipelineTiming('merge_request_time', build_set.configured_time) if event.elapsed_time: self.reportPipelineTiming('merger_files_changes_op_time', event.elapsed_time, elapsed=True) def onMergeCompleted(self, event, build_set): if build_set.merge_state == build_set.COMPLETE: self._onGlobalRepoStateCompleted(event, build_set) self.reportPipelineTiming('repo_state_time', build_set.repo_state_request_time) if event.elapsed_time: self.reportPipelineTiming('merger_repo_state_op_time', event.elapsed_time, elapsed=True) else: self._onMergeCompleted(event, build_set) if build_set.files_state == build_set.COMPLETE: # We're the second of the files/merger pair, report the stat self.reportPipelineTiming('merge_request_time', build_set.configured_time) if event.elapsed_time: self.reportPipelineTiming('merger_merge_op_time', event.elapsed_time, elapsed=True) def _onMergeCompleted(self, event, build_set): item = build_set.item log = get_annotated_logger(self.log, item.event) if isinstance(item.changes[0], model.Tag): # Since this is only used for Tag items, we know that # circular dependencies are not in play and there is only # one change in the list of changes. if len(item.changes) > 1: raise Exception("Tag item with more than one change") change = item.changes[0] source = self.sched.connections.getSource( change.project.connection_name) source.setChangeAttributes( change, containing_branches=event.item_in_branches) with build_set.activeContext(self.current_context): build_set.setMergeRepoState(event.repo_state) build_set.merge_state = build_set.COMPLETE if event.merged: build_set.setFiles(event.files) if not (event.merged or event.updated): log.info("Unable to merge %s" % item) item.setUnableToMerge(event.errors) def _onGlobalRepoStateCompleted(self, event, build_set): item = build_set.item if not event.updated: self.log.info("Unable to get global repo state for %s", item) item.setUnableToMerge(event.errors) else: self.log.info("Received global repo state for %s", item) with build_set.activeContext(self.current_context): build_set.setExtraRepoState(event.repo_state) build_set.repo_state_state = build_set.COMPLETE def _handleNodeRequestFallback(self, log, build_set, job, old_request): if len(job.nodeset_alternatives) <= job.nodeset_index + 1: # No alternatives to fall back upon return False # Increment the nodeset index and remove the old request with job.activeContext(self.current_context): job.nodeset_index = job.nodeset_index + 1 log.info("Re-attempting node request for job " f"{job.name} of item {build_set.item} " f"with nodeset alternative {job.nodeset_index}") build_set.removeJobNodeRequestID(job) # Make a new request if self.sched.globals.use_relative_priority: relative_priority = self.getNodePriority( build_set.item, build_set.item.getChangeForJob(job)) else: relative_priority = 0 log = build_set.item.annotateLogger(self.log) self._makeNodepoolRequest(log, build_set, job, relative_priority) return True def onNodesProvisioned(self, request, nodeset, build_set): log = get_annotated_logger(self.log, request.event_id) self.reportPipelineTiming('node_request_time', request.created_time) job = build_set.item.getJob(request.job_uuid) # First see if we need to retry the request if not request.fulfilled: log.info("Node request %s: failure for %s", request, job.name) if self._handleNodeRequestFallback(log, build_set, job, request): return # No more fallbacks -- tell the buildset the request is complete if nodeset is not None: build_set.jobNodeRequestComplete(job, nodeset) # Put a fake build through the cycle to clean it up. if not request.fulfilled: build_set.item.setNodeRequestFailure( job, f'Node request {request.id} failed') self._resumeBuilds(build_set) pipeline = build_set.item.pipeline tenant = pipeline.tenant event_queue = self.sched.pipeline_result_events[ tenant.name][pipeline.name] tenant.semaphore_handler.release( event_queue, build_set.item, job) log.info("Completed node request %s for job %s of item %s " "with nodes %s", request, job.name, build_set.item, request.nodes) def reportItem(self, item): log = get_annotated_logger(self.log, item.event) action = None phase1 = True phase2 = True is_cycle = len(item.changes) > 1 succeeded = item.didAllJobsSucceed() already_reported = item.reported if (self.changes_merge and is_cycle and succeeded): phase2 = False if not already_reported: action, reported = self._reportItem( item, phase1=phase1, phase2=phase2) if phase2 is False: phase1 = False phase2 = True action, reported = self._reportItem( item, phase1=phase1, phase2=phase2) item.updateAttributes(self.current_context, reported=reported) if not phase2: return if self.changes_merge: merged = item.reported if merged: for change in item.changes: source = change.project.source merged = source.isMerged(change, change.branch) if not merged: break if action: if action == 'success' and not merged: log.debug("Overriding result for %s to merge failure", item) action = 'merge-failure' item.setReportedResult('MERGE_FAILURE') self.reportNormalBuildsetEnd(item.current_build_set, action, final=True) change_queue = item.queue if not (succeeded and merged): if (not item.current_build_set.job_graph or not item.current_build_set.job_graph.job_uuids): error_reason = "did not have any jobs configured" elif not succeeded: error_reason = "failed tests" else: error_reason = "failed to merge" log.info("Changes for %s did not merge because it %s, " "status: all-succeeded: %s, merged: %s", item, error_reason, succeeded, merged) if not succeeded: change_queue.decreaseWindowSize() log.debug("%s window size decreased to %s", change_queue, change_queue.window) raise exceptions.MergeFailure( "Changes for %s failed to merge" % item) else: self.reportNormalBuildsetEnd(item.current_build_set, action, final=True) log.info("Reported %s status: all-succeeded: %s, " "merged: %s", item, succeeded, merged) change_queue.increaseWindowSize() log.debug("%s window size increased to %s", change_queue, change_queue.window) zuul_driver = self.sched.connections.drivers['zuul'] tenant = self.pipeline.tenant with trace.use_span(tracing.restoreSpan(item.span_info)): for change in item.changes: source = change.project.source zuul_driver.onChangeMerged(tenant, change, source) elif action: self.reportNormalBuildsetEnd(item.current_build_set, action, final=True) def _reportItem(self, item, phase1, phase2): log = get_annotated_logger(self.log, item.event) log.debug("Reporting phase1: %s phase2: %s item: %s", phase1, phase2, item) ret = True # Means error as returned by trigger.report # In the case of failure, we may not have 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. However, if we ran jobs, then we obviously # were in the pipeline config. project_in_pipeline = bool(item.getJobs()) if not project_in_pipeline: layout = None if item.layout_uuid: layout = self.getLayout(item) if not layout: layout = self.pipeline.tenant.layout try: for change in item.changes: project_in_pipeline = bool( layout.getProjectPipelineConfig(item, change)) if project_in_pipeline: break except Exception: log.exception("Invalid config for %s", item) if not project_in_pipeline: log.debug("Project not in pipeline %s for %s", self.pipeline, item) project_in_pipeline = False action = 'no-jobs' actions = self.pipeline.no_jobs_actions item.setReportedResult('NO_JOBS') elif item.current_build_set.has_blocking_errors: log.debug("Invalid config for %s", item) action = 'config-error' actions = self.pipeline.config_error_actions item.setReportedResult('CONFIG_ERROR') elif item.didMergerFail(): log.debug("Merge conflict") action = 'merge-conflict' actions = self.pipeline.merge_conflict_actions item.setReportedResult('MERGE_CONFLICT') elif item.wasDequeuedNeedingChange(): log.debug("Dequeued needing change") action = 'failure' actions = self.pipeline.failure_actions item.setReportedResult('FAILURE') elif item.wasDequeuedMissingRequirements(): log.debug("Dequeued missing merge requirements") action = 'failure' actions = self.pipeline.failure_actions item.setReportedResult('FAILURE') elif not item.getJobs(): # We don't send empty reports with +1 log.debug("No jobs for %s", item) action = 'no-jobs' actions = self.pipeline.no_jobs_actions item.setReportedResult('NO_JOBS') elif item.didAllJobsSucceed(): log.debug("success %s", self.pipeline.success_actions) action = 'success' actions = self.pipeline.success_actions item.setReportedResult('SUCCESS') with self.pipeline.state.activeContext(self.current_context): self.pipeline.state.consecutive_failures = 0 else: action = 'failure' actions = self.pipeline.failure_actions item.setReportedResult('FAILURE') with self.pipeline.state.activeContext(self.current_context): self.pipeline.state.consecutive_failures += 1 if project_in_pipeline and self.pipeline.state.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.state.disabled and self.pipeline.state.consecutive_failures >= self.pipeline.disable_at): self.pipeline.state.updateAttributes( self.current_context, disabled=True) if actions: log.info("Reporting item %s, actions: %s", item, actions) ret = self.sendReport(actions, item, phase1, phase2) if ret: log.error("Reporting item %s received: %s", item, ret) return action, (not ret) def reportStats(self, item, trigger_event=None): if not self.sched.statsd: return try: # Update the gauge on enqueue and dequeue, but timers only # when dequeing. if item.dequeue_time: dt = (item.dequeue_time - item.enqueue_time) * 1000 item_changes = len(item.changes) else: dt = None item_changes = 0 changes = sum(len(i.changes) for i in self.pipeline.getAllItems()) # TODO(jeblair): add items keys like changes tenant = self.pipeline.tenant basekey = 'zuul.tenant.%s' % tenant.name key = '%s.pipeline.%s' % (basekey, self.pipeline.name) # stats.timers.zuul.tenant..pipeline..resident_time # stats_counts.zuul.tenant..pipeline..total_changes # stats.gauges.zuul.tenant..pipeline..current_changes # stats.gauges.zuul.tenant..pipeline..window self.sched.statsd.gauge(key + '.current_changes', changes) self.sched.statsd.gauge(key + '.window', item.pipeline.window) if dt: self.sched.statsd.timing(key + '.resident_time', dt) self.sched.statsd.incr(key + '.total_changes', item_changes) if item.queue and item.queue.name: queuename = (item.queue.name. replace('.', '_').replace('/', '.')) # stats.gauges.zuul.tenant..pipeline..queue..resident_time # stats.gauges.zuul.tenant..pipeline..queue..total_changes # stats.gauges.zuul.tenant..pipeline..queue..current_changes # stats.gauges.zuul.tenant..pipeline..queue..window queuekey = '%s.queue.%s' % (key, queuename) queue_changes = sum(len(i.changes) for i in item.queue.queue) self.sched.statsd.gauge(queuekey + '.current_changes', queue_changes) self.sched.statsd.gauge(queuekey + '.window', item.queue.window) if dt: self.sched.statsd.timing(queuekey + '.resident_time', dt) self.sched.statsd.incr(queuekey + '.total_changes', item_changes) for change in item.changes: if hasattr(change, 'branch'): hostname = (change.project.canonical_hostname. replace('.', '_')) projectname = (change.project.name. replace('.', '_').replace('/', '.')) projectname = projectname.replace('.', '_').replace( '/', '.') branchname = change.branch.replace('.', '_').replace( '/', '.') # stats.timers.zuul.tenant..pipeline.. # project....resident_time # stats_counts.zuul.tenant..pipeline.. # project....total_changes key += '.project.%s.%s.%s' % (hostname, projectname, branchname) if dt: self.sched.statsd.timing(key + '.resident_time', dt) self.sched.statsd.incr(key + '.total_changes') if ( trigger_event and hasattr(trigger_event, 'arrived_at_scheduler_timestamp') ): now = time.time() arrived = trigger_event.arrived_at_scheduler_timestamp processing = (now - arrived) * 1000 elapsed = (now - trigger_event.timestamp) * 1000 self.sched.statsd.timing( basekey + '.event_enqueue_processing_time', processing) self.sched.statsd.timing( basekey + '.event_enqueue_time', elapsed) self.reportPipelineTiming('event_enqueue_time', trigger_event.timestamp) except Exception: self.log.exception("Exception reporting pipeline stats") def reportPipelineTiming(self, key, start, end=None, elapsed=False): if not self.sched.statsd: return if not start: return if end is None: end = time.time() pipeline = self.pipeline tenant = pipeline.tenant stats_key = f'zuul.tenant.{tenant.name}.pipeline.{pipeline.name}' if elapsed: dt = start else: dt = (end - start) * 1000 self.sched.statsd.timing(f'{stats_key}.{key}', dt)