blob: a582ead248bb800ad186cef0846e2785b12573fb [file] [log] [blame]
# Copyright 2012 Hewlett-Packard Development Company, L.P.
#
# 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 re
import time
from uuid import uuid4
import extras
OrderedDict = extras.try_imports(['collections.OrderedDict',
'ordereddict.OrderedDict'])
MERGER_MERGE = 1 # "git merge"
MERGER_MERGE_RESOLVE = 2 # "git merge -s resolve"
MERGER_CHERRY_PICK = 3 # "git cherry-pick"
MERGER_MAP = {
'merge': MERGER_MERGE,
'merge-resolve': MERGER_MERGE_RESOLVE,
'cherry-pick': MERGER_CHERRY_PICK,
}
PRECEDENCE_NORMAL = 0
PRECEDENCE_LOW = 1
PRECEDENCE_HIGH = 2
PRECEDENCE_MAP = {
None: PRECEDENCE_NORMAL,
'low': PRECEDENCE_LOW,
'normal': PRECEDENCE_NORMAL,
'high': PRECEDENCE_HIGH,
}
class Pipeline(object):
"""A top-level pipeline such as check, gate, post, etc."""
def __init__(self, name):
self.name = name
self.description = None
self.failure_message = None
self.success_message = None
self.dequeue_on_new_patchset = True
self.dequeue_on_conflict = True
self.job_trees = {} # project -> JobTree
self.manager = None
self.queues = []
self.precedence = PRECEDENCE_NORMAL
self.trigger = None
def __repr__(self):
return '<Pipeline %s>' % self.name
def setManager(self, manager):
self.manager = manager
def addProject(self, project):
job_tree = JobTree(None) # Null job == job tree root
self.job_trees[project] = job_tree
return job_tree
def getProjects(self):
return self.job_trees.keys()
def addQueue(self, queue):
self.queues.append(queue)
def getQueue(self, project):
for queue in self.queues:
if project in queue.projects:
return queue
return None
def getJobTree(self, project):
tree = self.job_trees.get(project)
return tree
def getJobs(self, changeish):
tree = self.getJobTree(changeish.project)
if not tree:
return []
return changeish.filterJobs(tree.getJobs())
def _findJobsToRun(self, job_trees, item):
torun = []
if item.item_ahead:
# Only run jobs if any 'hold' jobs on the change ahead
# have completed successfully.
if self.isHoldingFollowingChanges(item.item_ahead):
return []
for tree in job_trees:
job = tree.job
result = None
if job:
if not job.changeMatches(item.change):
continue
build = item.current_build_set.getBuild(job.name)
if build:
result = build.result
else:
# There is no build for the root of this job tree,
# so we should run it.
torun.append(job)
# If there is no job, this is a null job tree, and we should
# run all of its jobs.
if result == 'SUCCESS' or not job:
torun.extend(self._findJobsToRun(tree.job_trees, item))
return torun
def findJobsToRun(self, item):
tree = self.getJobTree(item.change.project)
if not tree:
return []
return self._findJobsToRun(tree.job_trees, item)
def haveAllJobsStarted(self, item):
for job in self.getJobs(item.change):
build = item.current_build_set.getBuild(job.name)
if not build or not build.start_time:
return False
return True
def areAllJobsComplete(self, item):
for job in self.getJobs(item.change):
build = item.current_build_set.getBuild(job.name)
if not build or not build.result:
return False
return True
def didAllJobsSucceed(self, item):
for job in self.getJobs(item.change):
if not job.voting:
continue
build = item.current_build_set.getBuild(job.name)
if not build:
return False
if build.result != 'SUCCESS':
return False
return True
def didAnyJobFail(self, item):
for job in self.getJobs(item.change):
if not job.voting:
continue
build = item.current_build_set.getBuild(job.name)
if build and build.result and (build.result != 'SUCCESS'):
return True
return False
def isHoldingFollowingChanges(self, item):
for job in self.getJobs(item.change):
if not job.hold_following_changes:
continue
build = item.current_build_set.getBuild(job.name)
if not build:
return True
if build.result != 'SUCCESS':
return True
if not item.item_ahead:
return False
return self.isHoldingFollowingChanges(item.item_ahead)
def setResult(self, item, build):
if build.retry:
item.removeBuild(build)
elif build.result != 'SUCCESS':
# Get a JobTree from a Job so we can find only its dependent jobs
root = self.getJobTree(item.change.project)
tree = root.getJobTreeForJob(build.job)
for job in tree.getJobs():
fakebuild = Build(job, None)
fakebuild.result = 'SKIPPED'
item.addBuild(fakebuild)
def setUnableToMerge(self, item, msg):
item.current_build_set.unable_to_merge = True
item.current_build_set.unable_to_merge_message = msg
root = self.getJobTree(item.change.project)
for job in root.getJobs():
fakebuild = Build(job, None)
fakebuild.result = 'SKIPPED'
item.addBuild(fakebuild)
def setDequeuedNeedingChange(self, item):
item.dequeued_needing_change = True
root = self.getJobTree(item.change.project)
for job in root.getJobs():
fakebuild = Build(job, None)
fakebuild.result = 'SKIPPED'
item.addBuild(fakebuild)
def getChangesInQueue(self):
changes = []
for shared_queue in self.queues:
changes.extend([x.change for x in shared_queue.queue])
return changes
def getAllItems(self):
items = []
for shared_queue in self.queues:
items.extend(shared_queue.queue)
items.extend(shared_queue.severed_heads)
return items
def formatStatusHTML(self):
ret = ''
for queue in self.queues:
if len(self.queues) > 1:
s = 'Change queue: %s' % queue.name
ret += s + '\n'
ret += '-' * len(s) + '\n'
for head in queue.getHeads():
ret += self.formatStatus(head, html=True)
return ret
def formatStatusJSON(self):
j_pipeline = dict(name=self.name,
description=self.description)
j_queues = []
j_pipeline['change_queues'] = j_queues
for queue in self.queues:
j_queue = dict(name=queue.name)
j_queues.append(j_queue)
j_queue['heads'] = []
for head in queue.getHeads():
j_changes = []
e = head
while e:
j_changes.append(self.formatItemJSON(e))
if (len(j_changes) > 1 and
(j_changes[-2]['remaining_time'] is not None) and
(j_changes[-1]['remaining_time'] is not None)):
j_changes[-1]['remaining_time'] = max(
j_changes[-2]['remaining_time'],
j_changes[-1]['remaining_time'])
e = e.item_behind
j_queue['heads'].append(j_changes)
return j_pipeline
def formatStatus(self, item, indent=0, html=False):
changeish = item.change
indent_str = ' ' * indent
ret = ''
if html and hasattr(changeish, 'url') and changeish.url is not None:
ret += '%sProject %s change <a href="%s">%s</a>\n' % (
indent_str,
changeish.project.name,
changeish.url,
changeish._id())
else:
ret += '%sProject %s change %s\n' % (indent_str,
changeish.project.name,
changeish._id())
for job in self.getJobs(changeish):
build = item.current_build_set.getBuild(job.name)
if build:
result = build.result
else:
result = None
job_name = job.name
if not job.voting:
voting = ' (non-voting)'
else:
voting = ''
if html:
if build:
url = build.url
else:
url = None
if url is not None:
job_name = '<a href="%s">%s</a>' % (url, job_name)
ret += '%s %s: %s%s' % (indent_str, job_name, result, voting)
ret += '\n'
if item.item_behind:
ret += '%sFollowed by:\n' % (indent_str)
ret += self.formatStatus(item.item_behind, indent + 2, html)
return ret
def formatItemJSON(self, item):
changeish = item.change
ret = {}
if hasattr(changeish, 'url') and changeish.url is not None:
ret['url'] = changeish.url
else:
ret['url'] = None
ret['id'] = changeish._id()
ret['project'] = changeish.project.name
ret['enqueue_time'] = int(item.enqueue_time * 1000)
ret['jobs'] = []
max_remaining = 0
for job in self.getJobs(changeish):
now = time.time()
build = item.current_build_set.getBuild(job.name)
elapsed = None
remaining = None
result = None
url = None
if build:
result = build.result
url = build.url
if build.start_time:
if build.end_time:
elapsed = int((build.end_time -
build.start_time) * 1000)
remaining = 0
else:
elapsed = int((now - build.start_time) * 1000)
if build.estimated_time:
remaining = max(
int(build.estimated_time * 1000) - elapsed,
0)
if remaining and remaining > max_remaining:
max_remaining = remaining
ret['jobs'].append(
dict(
name=job.name,
elapsed_time=elapsed,
remaining_time=remaining,
url=url,
result=result,
voting=job.voting))
if self.haveAllJobsStarted(item):
# if a change ahead has failed, we are unknown.
item_ahead_failed = False
i = item.item_ahead
while i:
if self.didAnyJobFail(i):
item_ahead_failed = True
i = None # safe to stop looking
else:
i = i.item_ahead
if item_ahead_failed:
ret['remaining_time'] = None
else:
ret['remaining_time'] = max_remaining
else:
ret['remaining_time'] = None
return ret
class ChangeQueue(object):
"""DependentPipelines have multiple parallel queues shared by
different projects; this is one of them. For instance, there may
a queue shared by interrelated projects foo and bar, and a second
queue for independent project baz. Pipelines have one or more
PipelineQueues."""
def __init__(self, pipeline, dependent=True):
self.pipeline = pipeline
self.name = ''
self.projects = []
self._jobs = set()
self.queue = []
self.severed_heads = []
self.dependent = dependent
def __repr__(self):
return '<ChangeQueue %s: %s>' % (self.pipeline.name, self.name)
def getJobs(self):
return self._jobs
def addProject(self, project):
if project not in self.projects:
self.projects.append(project)
names = [x.name for x in self.projects]
names.sort()
self.name = ', '.join(names)
self._jobs |= set(self.pipeline.getJobTree(project).getJobs())
def enqueueChange(self, change):
item = QueueItem(self.pipeline, change)
self.enqueueItem(item)
item.enqueue_time = time.time()
return item
def enqueueItem(self, item):
if self.dependent and self.queue:
item.item_ahead = self.queue[-1]
item.item_ahead.item_behind = item
self.queue.append(item)
def dequeueItem(self, item):
if item in self.queue:
self.queue.remove(item)
if item in self.severed_heads:
self.severed_heads.remove(item)
if item.item_ahead:
item.item_ahead.item_behind = item.item_behind
if item.item_behind:
item.item_behind.item_ahead = item.item_ahead
item.item_ahead = None
item.item_behind = None
item.dequeue_time = time.time()
def addSeveredHead(self, item):
self.severed_heads.append(item)
def mergeChangeQueue(self, other):
for project in other.projects:
self.addProject(project)
def getHead(self):
if not self.queue:
return None
return self.queue[0]
def getHeads(self):
heads = []
if self.dependent:
h = self.getHead()
if h:
heads.append(h)
else:
heads.extend(self.queue)
heads.extend(self.severed_heads)
return heads
class Project(object):
def __init__(self, name):
self.name = name
self.merge_mode = MERGER_MERGE_RESOLVE
def __str__(self):
return self.name
def __repr__(self):
return '<Project %s>' % (self.name)
class Job(object):
def __init__(self, name):
# If you add attributes here, be sure to add them to the copy method.
self.name = name
self.failure_message = None
self.success_message = None
self.failure_pattern = None
self.success_pattern = None
self.parameter_function = None
self.hold_following_changes = False
self.voting = True
self.branches = []
self._branches = []
self.files = []
self._files = []
def __str__(self):
return self.name
def __repr__(self):
return '<Job %s>' % (self.name)
def copy(self, other):
if other.failure_message:
self.failure_message = other.failure_message
if other.success_message:
self.success_message = other.success_message
if other.failure_pattern:
self.failure_pattern = other.failure_pattern
if other.success_pattern:
self.success_pattern = other.success_pattern
if other.parameter_function:
self.parameter_function = other.parameter_function
if other.branches:
self.branches = other.branches[:]
self._branches = other._branches[:]
if other.files:
self.files = other.files[:]
self._files = other._files[:]
self.hold_following_changes = other.hold_following_changes
self.voting = other.voting
def changeMatches(self, change):
matches_branch = False
for branch in self.branches:
if hasattr(change, 'branch') and branch.match(change.branch):
matches_branch = True
if hasattr(change, 'ref') and branch.match(change.ref):
matches_branch = True
if self.branches and not matches_branch:
return False
matches_file = False
for f in self.files:
if hasattr(change, 'files'):
for cf in change.files:
if f.match(cf):
matches_file = True
if self.files and not matches_file:
return False
return True
class JobTree(object):
""" A JobTree represents an instance of one Job, and holds JobTrees
whose jobs should be run if that Job succeeds. A root node of a
JobTree will have no associated Job. """
def __init__(self, job):
self.job = job
self.job_trees = []
def addJob(self, job):
if job not in [x.job for x in self.job_trees]:
t = JobTree(job)
self.job_trees.append(t)
return t
def getJobs(self):
jobs = []
for x in self.job_trees:
jobs.append(x.job)
jobs.extend(x.getJobs())
return jobs
def getJobTreeForJob(self, job):
if self.job == job:
return self
for tree in self.job_trees:
ret = tree.getJobTreeForJob(job)
if ret:
return ret
return None
class Build(object):
def __init__(self, job, uuid):
self.job = job
self.uuid = uuid
self.url = None
self.number = None
self.result = None
self.build_set = None
self.launch_time = time.time()
self.start_time = None
self.end_time = None
self.estimated_time = None
self.pipeline = None
self.canceled = False
self.retry = False
self.parameters = {}
def __repr__(self):
return '<Build %s of %s>' % (self.uuid, self.job.name)
class BuildSet(object):
def __init__(self, item):
self.item = item
self.other_changes = []
self.builds = {}
self.result = None
self.next_build_set = None
self.previous_build_set = None
self.ref = None
self.commit = None
self.unable_to_merge = False
self.unable_to_merge_message = None
def setConfiguration(self):
# The change isn't enqueued until after it's created
# so we don't know what the other changes ahead will be
# until jobs start.
if not self.other_changes:
next_item = self.item.item_ahead
while next_item:
self.other_changes.append(next_item.change)
next_item = next_item.item_ahead
if not self.ref:
self.ref = 'Z' + uuid4().hex
def addBuild(self, build):
self.builds[build.job.name] = build
build.build_set = self
def removeBuild(self, build):
del self.builds[build.job.name]
def getBuild(self, job_name):
return self.builds.get(job_name)
def getBuilds(self):
keys = self.builds.keys()
keys.sort()
return [self.builds.get(x) for x in keys]
class QueueItem(object):
"""A changish inside of a Pipeline queue"""
def __init__(self, pipeline, change):
self.pipeline = pipeline
self.change = change # a changeish
self.build_sets = []
self.dequeued_needing_change = False
self.current_build_set = BuildSet(self)
self.build_sets.append(self.current_build_set)
self.item_ahead = None
self.item_behind = None
self.enqueue_time = None
self.dequeue_time = None
self.reported = False
def resetAllBuilds(self):
old = self.current_build_set
self.current_build_set.result = 'CANCELED'
self.current_build_set = BuildSet(self)
old.next_build_set = self.current_build_set
self.current_build_set.previous_build_set = old
self.build_sets.append(self.current_build_set)
def addBuild(self, build):
self.current_build_set.addBuild(build)
build.pipeline = self.pipeline
def removeBuild(self, build):
self.current_build_set.removeBuild(build)
def setReportedResult(self, result):
self.current_build_set.result = result
class Changeish(object):
"""Something like a change; either a change or a ref"""
is_reportable = False
def __init__(self, project):
self.project = project
def equals(self, other):
raise NotImplementedError()
def isUpdateOf(self, other):
raise NotImplementedError()
def filterJobs(self, jobs):
return filter(lambda job: job.changeMatches(self), jobs)
def getRelatedChanges(self):
return set()
class Change(Changeish):
is_reportable = True
def __init__(self, project):
super(Change, self).__init__(project)
self.branch = None
self.number = None
self.url = None
self.patchset = None
self.refspec = None
self.files = []
self.needs_change = None
self.needed_by_changes = []
self.is_current_patchset = True
self.can_merge = False
self.is_merged = False
self.failed_to_merge = False
def _id(self):
return '%s,%s' % (self.number, self.patchset)
def __repr__(self):
return '<Change 0x%x %s>' % (id(self), self._id())
def equals(self, other):
if self.number == other.number and self.patchset == other.patchset:
return True
return False
def isUpdateOf(self, other):
if ((hasattr(other, 'number') and self.number == other.number) and
(hasattr(other, 'patchset') and
self.patchset is not None and
other.patchset is not None and
int(self.patchset) > int(other.patchset))):
return True
return False
def getRelatedChanges(self):
related = set()
if self.needs_change:
related.add(self.needs_change)
for c in self.needed_by_changes:
related.add(c)
related.update(c.getRelatedChanges())
return related
class Ref(Changeish):
is_reportable = False
def __init__(self, project):
super(Ref, self).__init__(project)
self.ref = None
self.oldrev = None
self.newrev = None
def _id(self):
return self.newrev
def __repr__(self):
rep = None
if self.newrev == '0000000000000000000000000000000000000000':
rep = '<Ref 0x%x deletes %s from %s' % (
id(self), self.ref, self.oldrev)
elif self.oldrev == '0000000000000000000000000000000000000000':
rep = '<Ref 0x%x creates %s on %s>' % (
id(self), self.ref, self.newrev)
else:
# Catch all
rep = '<Ref 0x%x %s updated %s..%s>' % (
id(self), self.ref, self.oldrev, self.newrev)
return rep
def equals(self, other):
if (self.project == other.project
and self.ref == other.ref
and self.newrev == other.newrev):
return True
return False
def isUpdateOf(self, other):
return False
class NullChange(Changeish):
is_reportable = False
def __init__(self, project):
super(NullChange, self).__init__(project)
def _id(self):
return 'None'
def equals(self, other):
return False
def isUpdateOf(self, other):
return False
class TriggerEvent(object):
def __init__(self):
self.data = None
# common
self.type = None
self.project_name = None
self.trigger_name = None
# Representation of the user account that performed the event.
self.account = None
# patchset-created, comment-added, etc.
self.change_number = None
self.change_url = None
self.patch_number = None
self.refspec = None
self.approvals = []
self.branch = None
self.comment = None
# ref-updated
self.ref = None
self.oldrev = None
self.newrev = None
# timer
self.timespec = None
def __repr__(self):
ret = '<TriggerEvent %s %s' % (self.type, self.project_name)
if self.branch:
ret += " %s" % self.branch
if self.change_number:
ret += " %s,%s" % (self.change_number, self.patch_number)
if self.approvals:
ret += ' ' + ', '.join(
['%s:%s' % (a['type'], a['value']) for a in self.approvals])
ret += '>'
return ret
def getChange(self, project, trigger):
if self.change_number:
change = trigger.getChange(self.change_number, self.patch_number)
elif self.ref:
change = Ref(project)
change.ref = self.ref
change.oldrev = self.oldrev
change.newrev = self.newrev
change.url = trigger.getGitwebUrl(project, sha=self.newrev)
else:
change = NullChange(project)
return change
class EventFilter(object):
def __init__(self, types=[], branches=[], refs=[], approvals={},
comment_filters=[], email_filters=[], timespecs=[]):
self._types = types
self._branches = branches
self._refs = refs
self._comment_filters = comment_filters
self._email_filters = email_filters
self.types = [re.compile(x) for x in types]
self.branches = [re.compile(x) for x in branches]
self.refs = [re.compile(x) for x in refs]
self.comment_filters = [re.compile(x) for x in comment_filters]
self.email_filters = [re.compile(x) for x in email_filters]
self.approvals = approvals
self.timespecs = timespecs
def __repr__(self):
ret = '<EventFilter'
if self._types:
ret += ' types: %s' % ', '.join(self._types)
if self._branches:
ret += ' branches: %s' % ', '.join(self._branches)
if self._refs:
ret += ' refs: %s' % ', '.join(self._refs)
if self.approvals:
ret += ' approvals: %s' % ', '.join(
['%s:%s' % a for a in self.approvals.items()])
if self._comment_filters:
ret += ' comment_filters: %s' % ', '.join(self._comment_filters)
if self._email_filters:
ret += ' email_filters: %s' % ', '.join(self._email_filters)
if self.timespecs:
ret += ' timespecs: %s' % ', '.join(self.timespecs)
ret += '>'
return ret
def matches(self, event):
def normalizeCategory(name):
name = name.lower()
return re.sub(' ', '-', name)
# event types are ORed
matches_type = False
for etype in self.types:
if etype.match(event.type):
matches_type = True
if self.types and not matches_type:
return False
# branches are ORed
matches_branch = False
for branch in self.branches:
if branch.match(event.branch):
matches_branch = True
if self.branches and not matches_branch:
return False
# refs are ORed
matches_ref = False
for ref in self.refs:
if ref.match(event.ref):
matches_ref = True
if self.refs and not matches_ref:
return False
# comment_filters are ORed
matches_comment_filter = False
for comment_filter in self.comment_filters:
if (event.comment is not None and
comment_filter.search(event.comment)):
matches_comment_filter = True
if self.comment_filters and not matches_comment_filter:
return False
# We better have an account provided by Gerrit to do
# email filtering.
if event.account is not None:
account_email = event.account.get('email')
# email_filters are ORed
matches_email_filter = False
for email_filter in self.email_filters:
if (account_email is not None and
email_filter.search(account_email)):
matches_email_filter = True
if self.email_filters and not matches_email_filter:
return False
# approvals are ANDed
for category, value in self.approvals.items():
matches_approval = False
for eapproval in event.approvals:
if (normalizeCategory(eapproval['description']) == category and
int(eapproval['value']) == int(value)):
matches_approval = True
if not matches_approval:
return False
# timespecs are ORed
matches_timespec = False
for timespec in self.timespecs:
if (event.timespec == timespec):
matches_timespec = True
if self.timespecs and not matches_timespec:
return False
return True
class Layout(object):
def __init__(self):
self.projects = {}
self.pipelines = OrderedDict()
self.jobs = {}
self.metajobs = []
def getJob(self, name):
if name in self.jobs:
return self.jobs[name]
job = Job(name)
if name.startswith('^'):
# This is a meta-job
regex = re.compile(name)
self.metajobs.append((regex, job))
else:
# Apply attributes from matching meta-jobs
for regex, metajob in self.metajobs:
if regex.match(name):
job.copy(metajob)
self.jobs[name] = job
return job