Files
fm-orchestrator/module_build_service/scheduler/producer.py
Chenxiong Qi c36bd7ebac Name component builds by states
This patch attemps to make code shorter and easier to read by naming
component builds for different states.

Signed-off-by: Chenxiong Qi <cqi@redhat.com>
2019-09-16 15:29:21 +08:00

513 lines
23 KiB
Python

# -*- coding: utf-8 -*-
# Copyright (c) 2016 Red Hat, Inc.
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
#
""" The PollingProducer class that acts as a producer entry point for
fedmsg-hub. This class polls the database for tasks to do.
"""
import koji
import operator
from datetime import timedelta, datetime
from sqlalchemy.orm import lazyload
from moksha.hub.api.producer import PollingProducer
import module_build_service.messaging
import module_build_service.scheduler
import module_build_service.scheduler.consumer
from module_build_service import conf, models, log
from module_build_service.builder import GenericBuilder
from module_build_service.builder.KojiModuleBuilder import KojiModuleBuilder
from module_build_service.utils.greenwave import greenwave
class MBSProducer(PollingProducer):
frequency = timedelta(seconds=conf.polling_interval)
def poll(self):
with models.make_db_session(conf) as db_session:
try:
self.log_summary(db_session)
self.process_waiting_module_builds(db_session)
self.process_open_component_builds(db_session)
self.fail_lost_builds(db_session)
self.process_paused_module_builds(conf, db_session)
self.retrigger_new_repo_on_failure(conf, db_session)
self.delete_old_koji_targets(conf, db_session)
self.cleanup_stale_failed_builds(conf, db_session)
self.sync_koji_build_tags(conf, db_session)
self.poll_greenwave(conf, db_session)
except Exception:
msg = "Error in poller execution:"
log.exception(msg)
log.info('Poller will now sleep for "{}" seconds'.format(conf.polling_interval))
def fail_lost_builds(self, db_session):
# This function is supposed to be handling only the part which can't be
# updated through messaging (e.g. srpm-build failures). Please keep it
# fit `n` slim. We do want rest to be processed elsewhere
# TODO re-use
if conf.system == "koji":
# We don't do this on behalf of users
koji_session = KojiModuleBuilder.get_session(conf, login=False)
log.info("Querying tasks for statuses:")
res = (
db_session.query(models.ComponentBuild)
.filter_by(state=koji.BUILD_STATES["BUILDING"])
.options(lazyload("module_build"))
.all()
)
log.info("Checking status for {0} tasks".format(len(res)))
for component_build in res:
log.debug(component_build.json(db_session))
# Don't check tasks which haven't been triggered yet
if not component_build.task_id:
continue
# Don't check tasks for components which have been reused,
# they may have BUILDING state temporarily before we tag them
# to new module tag. Checking them would be waste of resources.
if component_build.reused_component_id:
log.debug(
'Skipping check for task "{0}", '
'the component has been reused ("{1}").'.format(
component_build.task_id, component_build.reused_component_id)
)
continue
task_id = component_build.task_id
log.info('Checking status of task_id "{0}"'.format(task_id))
task_info = koji_session.getTaskInfo(task_id)
state_mapping = {
# Cancelled and failed builds should be marked as failed.
koji.TASK_STATES["CANCELED"]: koji.BUILD_STATES["FAILED"],
koji.TASK_STATES["FAILED"]: koji.BUILD_STATES["FAILED"],
# Completed tasks should be marked as complete.
koji.TASK_STATES["CLOSED"]: koji.BUILD_STATES["COMPLETE"],
}
# If it is a closed/completed task, then we can extract the NVR
build_version, build_release = None, None # defaults
if task_info["state"] == koji.TASK_STATES["CLOSED"]:
builds = koji_session.listBuilds(taskID=task_id)
if not builds:
log.warning(
"Task ID %r is closed, but we found no builds in koji." % task_id)
elif len(builds) > 1:
log.warning(
"Task ID %r is closed, but more than one build is present!" % task_id)
else:
build_version = builds[0]["version"]
build_release = builds[0]["release"]
log.info(" task {0!r} is in state {1!r}".format(task_id, task_info["state"]))
if task_info["state"] in state_mapping:
# Fake a fedmsg message on our internal queue
msg = module_build_service.messaging.KojiBuildChange(
msg_id="producer::fail_lost_builds fake msg",
build_id=component_build.task_id,
task_id=component_build.task_id,
build_name=component_build.package,
build_new_state=state_mapping[task_info["state"]],
build_release=build_release,
build_version=build_version,
)
module_build_service.scheduler.consumer.work_queue_put(msg)
elif conf.system == "mock":
pass
def cleanup_stale_failed_builds(self, conf, db_session):
""" Does various clean up tasks on stale failed module builds
:param conf: the MBS configuration object
:param db_session: a SQLAlchemy database session
"""
if conf.system == "koji":
stale_date = datetime.utcnow() - timedelta(days=conf.cleanup_failed_builds_time)
stale_module_builds = (
db_session.query(models.ModuleBuild)
.filter(
models.ModuleBuild.state == models.BUILD_STATES["failed"],
models.ModuleBuild.time_modified <= stale_date,
)
.all()
)
if stale_module_builds:
log.info(
"{0} stale failed module build(s) will be cleaned up".format(
len(stale_module_builds))
)
for module in stale_module_builds:
log.info("{0!r} is stale and is being cleaned up".format(module))
# Find completed artifacts in the stale build
artifacts = [c for c in module.component_builds if c.is_completed]
# If there are no completed artifacts, then there is nothing to tag
if artifacts:
# Set buildroot_connect=False so it doesn't recreate the Koji target and etc.
builder = GenericBuilder.create_from_module(
db_session, module, conf, buildroot_connect=False
)
builder.untag_artifacts([c.nvr for c in artifacts])
# Mark the artifacts as untagged in the database
for c in artifacts:
c.tagged = False
c.tagged_in_final = False
db_session.add(c)
state_reason = (
"The module was garbage collected since it has failed over {0}"
" day(s) ago".format(conf.cleanup_failed_builds_time)
)
module.transition(
db_session,
conf,
models.BUILD_STATES["garbage"],
state_reason=state_reason,
failure_type="user",
)
db_session.add(module)
db_session.commit()
def log_summary(self, db_session):
log.info("Current status:")
consumer = module_build_service.scheduler.consumer.get_global_consumer()
backlog = consumer.incoming.qsize()
log.info(" * internal queue backlog is {0}".format(backlog))
states = sorted(models.BUILD_STATES.items(), key=operator.itemgetter(1))
for name, code in states:
query = db_session.query(models.ModuleBuild).filter_by(state=code)
count = query.count()
if count:
log.info(" * {0} module builds in the {1} state".format(count, name))
if name == "build":
for module_build in query.all():
log.info(" * {0!r}".format(module_build))
# First batch is number '1'.
for i in range(1, module_build.batch + 1):
n = len([c for c in module_build.component_builds if c.batch == i])
log.info(" * {0} components in batch {1}".format(n, i))
def _nudge_module_builds_in_state(self, db_session, state_name, older_than_minutes):
"""
Finds all the module builds in the `state` with `time_modified` older
than `older_than_minutes` and adds fake MBSModule message to the
work queue.
"""
log.info("Looking for module builds stuck in the %s state", state_name)
builds = models.ModuleBuild.by_state(db_session, state_name)
log.info(" %r module builds in the %s state...", len(builds), state_name)
now = datetime.utcnow()
time_modified_threshold = timedelta(minutes=older_than_minutes)
for build in builds:
# Only give builds a nudge if stuck for more than ten minutes
if (now - build.time_modified) < time_modified_threshold:
continue
# Pretend the build is modified, so we don't tight spin.
build.time_modified = now
# Fake a message to kickstart the build anew in the consumer
state = module_build_service.models.BUILD_STATES[state_name]
msg = module_build_service.messaging.MBSModule(
"nudge_module_builds_fake_message", build.id, state)
log.info(" Scheduling faked event %r" % msg)
module_build_service.scheduler.consumer.work_queue_put(msg)
db_session.commit()
def process_waiting_module_builds(self, db_session):
for state in ["init", "wait"]:
self._nudge_module_builds_in_state(db_session, state, 10)
def process_open_component_builds(self, db_session):
log.warning("process_open_component_builds is not yet implemented...")
def process_paused_module_builds(self, config, db_session):
log.info("Looking for paused module builds in the build state")
if module_build_service.utils.at_concurrent_component_threshold(config, db_session):
log.debug(
"Will not attempt to start paused module builds due to "
"the concurrent build threshold being met"
)
return
ten_minutes = timedelta(minutes=10)
# Check for module builds that are in the build state but don't have any active component
# builds. Exclude module builds in batch 0. This is likely a build of a module without
# components.
module_builds = (
db_session.query(models.ModuleBuild)
.filter(
models.ModuleBuild.state == models.BUILD_STATES["build"],
models.ModuleBuild.batch > 0,
)
.all()
)
for module_build in module_builds:
now = datetime.utcnow()
# Only give builds a nudge if stuck for more than ten minutes
if (now - module_build.time_modified) < ten_minutes:
continue
# If there are no components in the build state on the module build,
# then no possible event will start off new component builds.
# But do not try to start new builds when we are waiting for the
# repo-regen.
if not module_build.current_batch(koji.BUILD_STATES["BUILDING"]):
# Initialize the builder...
builder = GenericBuilder.create_from_module(db_session, module_build, config)
if _has_missed_new_repo_message(module_build, builder.koji_session):
log.info(" Processing the paused module build %r", module_build)
further_work = module_build_service.utils.start_next_batch_build(
config, module_build, db_session, builder)
for event in further_work:
log.info(" Scheduling faked event %r" % event)
module_build_service.scheduler.consumer.work_queue_put(event)
# Check if we have met the threshold.
if module_build_service.utils.at_concurrent_component_threshold(config, db_session):
break
def retrigger_new_repo_on_failure(self, config, db_session):
"""
Retrigger failed new repo tasks for module builds in the build state.
The newRepo task may fail for various reasons outside the scope of MBS.
This method will detect this scenario and retrigger the newRepo task
if needed to avoid the module build from being stuck in the "build" state.
"""
if config.system != "koji":
return
koji_session = module_build_service.builder.KojiModuleBuilder.KojiModuleBuilder.get_session(
config)
for module_build in (
db_session.query(models.ModuleBuild).filter_by(state=models.BUILD_STATES["build"]).all()
):
if not module_build.new_repo_task_id:
continue
task_info = koji_session.getTaskInfo(module_build.new_repo_task_id)
if task_info["state"] in [koji.TASK_STATES["CANCELED"], koji.TASK_STATES["FAILED"]]:
log.info(
"newRepo task %s for %r failed, starting another one",
str(module_build.new_repo_task_id), module_build,
)
taginfo = koji_session.getTag(module_build.koji_tag + "-build")
module_build.new_repo_task_id = koji_session.newRepo(taginfo["name"])
db_session.commit()
def delete_old_koji_targets(self, config, db_session):
"""
Deletes targets older than `config.koji_target_delete_time` seconds
from Koji to cleanup after the module builds.
"""
if config.system != "koji":
return
log.info("Looking for module builds which Koji target can be removed")
now = datetime.utcnow()
koji_session = KojiModuleBuilder.get_session(config)
for target in koji_session.getBuildTargets():
koji_tag = target["dest_tag_name"]
module = db_session.query(models.ModuleBuild).filter_by(koji_tag=koji_tag).first()
if (
not module
or module.name in conf.base_module_names
or module.state in [
models.BUILD_STATES["init"],
models.BUILD_STATES["wait"],
models.BUILD_STATES["build"],
]
):
continue
# Double-check that the target we are going to remove is prefixed
# by our prefix, so we won't remove f26 when there is some garbage
# in DB or Koji.
for allowed_prefix in config.koji_tag_prefixes:
if target["name"].startswith(allowed_prefix + "-"):
break
else:
log.error("Module %r has Koji target with not allowed prefix.", module)
continue
delta = now - module.time_completed
if delta.total_seconds() > config.koji_target_delete_time:
log.info("Removing target of module %r", module)
koji_session.deleteBuildTarget(target["id"])
def cancel_stuck_module_builds(self, config, db_session):
"""
Method transitions builds which are stuck in one state too long to the "failed" state.
The states are defined with the "cleanup_stuck_builds_states" config option and the
time is defined by the "cleanup_stuck_builds_time" config option.
"""
log.info(
'Looking for module builds stuck in the states "{states}" more than {days} days'
.format(
states=" and ".join(config.cleanup_stuck_builds_states),
days=config.cleanup_stuck_builds_time,
)
)
delta = timedelta(days=config.cleanup_stuck_builds_time)
now = datetime.utcnow()
threshold = now - delta
states = [
module_build_service.models.BUILD_STATES[state]
for state in config.cleanup_stuck_builds_states
]
module_builds = (
db_session.query(models.ModuleBuild)
.filter(
models.ModuleBuild.state.in_(states), models.ModuleBuild.time_modified < threshold
)
.all()
)
log.info(" {0!r} module builds are stuck...".format(len(module_builds)))
for build in module_builds:
nsvc = ":".join([build.name, build.stream, build.version, build.context])
log.info('Transitioning build "{nsvc}" to "Failed" state.'.format(nsvc=nsvc))
state_reason = "The module was in {state} for more than {days} days".format(
state=build.state, days=config.cleanup_stuck_builds_time
)
build.transition(
db_session,
config,
state=models.BUILD_STATES["failed"],
state_reason=state_reason,
failure_type="user",
)
db_session.commit()
def sync_koji_build_tags(self, config, db_session):
"""
Method checking the "tagged" and "tagged_in_final" attributes of
"complete" ComponentBuilds in the current batch of module builds
in "building" state against the Koji.
In case the Koji shows the build as tagged/tagged_in_final,
fake "tagged" message is added to work queue.
"""
if conf.system != "koji":
return
koji_session = KojiModuleBuilder.get_session(conf, login=False)
threshold = datetime.utcnow() - timedelta(minutes=10)
module_builds = db_session.query(models.ModuleBuild).filter(
models.ModuleBuild.time_modified < threshold,
models.ModuleBuild.state == models.BUILD_STATES["build"]
).all()
for module_build in module_builds:
complete_components = module_build.current_batch(koji.BUILD_STATES["COMPLETE"])
for c in complete_components:
# In case the component is tagged in the build tag and
# also tagged in the final tag (or it is build_time_only
# and therefore should not be tagged in final tag), skip it.
if c.tagged and (c.tagged_in_final or c.build_time_only):
continue
log.info(
"%r: Component %r is complete, but not tagged in the "
"final and/or build tags.",
module_build, c,
)
# Check in which tags the component is tagged.
tag_dicts = koji_session.listTags(c.nvr)
tags = [tag_dict["name"] for tag_dict in tag_dicts]
# If it is tagged in final tag, but MBS does not think so,
# schedule fake message.
if not c.tagged_in_final and module_build.koji_tag in tags:
msg = module_build_service.messaging.KojiTagChange(
"sync_koji_build_tags_fake_message", module_build.koji_tag, c.package, c.nvr
)
log.info(" Scheduling faked event %r" % msg)
module_build_service.scheduler.consumer.work_queue_put(msg)
# If it is tagged in the build tag, but MBS does not think so,
# schedule fake message.
build_tag = module_build.koji_tag + "-build"
if not c.tagged and build_tag in tags:
msg = module_build_service.messaging.KojiTagChange(
"sync_koji_build_tags_fake_message", build_tag, c.package, c.nvr)
log.info(" Scheduling faked event %r" % msg)
module_build_service.scheduler.consumer.work_queue_put(msg)
def poll_greenwave(self, config, db_session):
"""
Polls Greenwave for all builds in done state
:param db_session: SQLAlchemy DB session
:return: None
"""
if greenwave is None:
return
module_builds = (
db_session.query(models.ModuleBuild)
.filter_by(state=models.BUILD_STATES["done"], scratch=False).all()
)
log.info("Checking Greenwave for %d builds", len(module_builds))
for build in module_builds:
if greenwave.check_gating(build):
build.transition(db_session, config, state=models.BUILD_STATES["ready"])
else:
build.state_reason = "Gating failed (MBS will retry in {0} seconds)".format(
conf.polling_interval
)
if greenwave.error_occurred:
build.state_reason += " (Error occured while querying Greenwave)"
build.time_modified = datetime.utcnow()
db_session.commit()
def _has_missed_new_repo_message(module_build, koji_session):
"""
Returns whether or not a new repo message has probably been missed.
"""
if not module_build.new_repo_task_id:
# A newRepo task has incorrectly not been requested. Treat it as a missed
# message so module build can recover.
return True
log.debug(
'Checking status of newRepo task "%d" for %s', module_build.new_repo_task_id, module_build)
task_info = koji_session.getTaskInfo(module_build.new_repo_task_id)
# Other final states, FAILED and CANCELED, are handled by retrigger_new_repo_on_failure
return task_info["state"] == koji.TASK_STATES["CLOSED"]