def error(self, request, code, message):
"""
Writes the proper out an error response message depending on the
content type in the request
"""
response_types = self.get_accept(request)
logger.error(message)
if "text/html" in response_types:
request.setResponseCode(code)
html_error = template.load("error.html")
result = html_error.render(
code=code, code_msg=responses[code], message=message)
request.write(result.encode())
elif "application/json" in response_types:
request.setResponseCode(code)
request.write(dumps({"error": message}))
else:
request.setResponseCode(UNSUPPORTED_MEDIA_TYPE)
error = dumps(
{"error":
"Can only handle one of %s here" % self.ALLOWED_ACCEPT})
request.write(error)
request.finish()
def get(self, **kwargs):
request = kwargs.get("request")
if request is not None and request_from_master(request):
config.master_contacted()
return dumps(versions=[1]), OK
def request(self, method, **kwargs):
data = kwargs.pop("data", None)
headers = kwargs.pop("headers", {})
uri = kwargs.pop("uri", self.URI)
request = DummyRequest(uri)
request.method = method.upper()
if data is not None:
request.content = StringIO()
request.content.write(dumps(data))
request.content.seek(0)
if self.DEFAULT_HEADERS is not NotImplemented:
headers.update(self.DEFAULT_HEADERS)
if headers:
self.failUnlessIsInstance(headers, dict)
for key, value in headers.items():
if isinstance(value, STRING_TYPES):
value = [value]
self.failUnlessIsInstance(value, list)
request.requestHeaders.setRawHeaders(key, value)
self.failUnlessEqual(kwargs, {}, "Unknown keywords %s" % kwargs.keys())
return request
def get(self, **_):
# Get counts for child processes and grandchild processes
process = psutil.Process()
direct_child_processes = len(process.children(recursive=False))
all_child_processes = len(process.children(recursive=True))
grandchild_processes = all_child_processes - direct_child_processes
# Determine the last time we talked to the master (if ever)
contacted = config.master_contacted(update=False)
if isinstance(contacted, datetime): # pragma: no cover
contacted = datetime.utcnow() - contacted
# Determine the last time we announced ourselves to the
# master (if ever)
last_announce = config.get("last_announce", None)
if isinstance(last_announce, datetime): # pragma: no cover
last_announce = datetime.utcnow() - last_announce
data = {"state": config["state"],
"agent_hostname": config["agent_hostname"],
"free_ram": memory.free_ram(),
"agent_process_ram": memory.process_memory(),
"consumed_ram": memory.total_consumption(),
"child_processes": direct_child_processes,
"grandchild_processes": grandchild_processes,
"pids": config["pids"],
"agent_id": config["agent_id"],
"last_master_contact": contacted,
"last_announce": last_announce,
"agent_lock_file": config["agent_lock_file"],
"uptime": total_seconds(
timedelta(seconds=time.time() - config["start"])),
"jobs": list(config["jobtypes"].keys())}
if config["farm_name"]:
data["farm_name"] = config["farm_name"]
return dumps(data)
def fake_work():
parser = argparse.ArgumentParser(
description="Quick and dirty script to create a job type, a job, and "
"some tasks which are then posted directly to the "
"agent. The primary purpose of this script is to test "
"the internal of the job types")
parser.add_argument(
"--master-api", default="http://127.0.0.1/api/v1",
help="The url to the master's api [default: %(default)s]")
parser.add_argument(
"--agent-api", default="http://127.0.0.1:50000/api/v1",
help="The url to the agent's api [default: %(default)s]")
parser.add_argument(
"--jobtype", default="FakeRender",
help="The job type to use [default: %(default)s]")
parser.add_argument(
"--job", type=int,
help="If provided then this will be the job we pull tasks from "
"and assign to the agent. Please note we'll only be pulling "
"tasks that aren't running or assigned.")
args = parser.parse_args()
logger.info("Master args.master_api: %s", args.master_api)
logger.info("Agent args.master_api: %s", args.agent_api)
assert not args.agent_api.endswith("/")
assert not args.master_api.endswith("/")
# session to make requests with
session = requests.Session()
session.headers.update({"content-type": "application/json"})
existing_jobtype = session.get(
args.master_api + "/jobtypes/%s" % args.jobtype)
# Create a FakeRender job type if one does not exist
if not existing_jobtype.ok:
sourcecode = dedent("""
from pyfarm.jobtypes.examples import %s as _%s
class %s(_%s):
pass""" % (args.jobtype, args.jobtype, args.jobtype, args.jobtype))
response = session.post(
args.master_api + "/jobtypes/",
data=dumps({
"name": args.jobtype,
"classname": args.jobtype,
"code": sourcecode,
"max_batch": 1}))
assert response.ok, response.json()
jobtype_data = response.json()
logger.info(
"Created job type %r, id %r", args.jobtype, jobtype_data["id"])
else:
jobtype_data = existing_jobtype.json()
logger.info(
"Job type %r already exists, id %r",
args.jobtype, jobtype_data["id"])
jobtype_version = jobtype_data["version"]
if args.job is None:
job = session.post(
args.master_api + "/jobs/",
data=dumps({
"start": 1,
"end": 1,
"title": "Fake Job - %s" % int(time.time()),
"jobtype": args.jobtype}))
assert job.ok, job.json()
job = job.json()
logger.info("Job %r created", job["id"])
else:
job = session.get(args.master_api + "/jobs/%s" % args.job)
if not job.ok:
logger.error("No such job with id %r", args.job)
return
else:
job = job.json()
logger.info("Job %r exists", job["id"])
tasks = session.get(args.master_api + "/jobs/%s/tasks/" % job["id"])
assert tasks.ok
job_tasks = []
for task in tasks.json():
if task["state"] not in ("queued", "failed"):
logger.info(
"Can't use task %s, it's state is not 'queued' or 'failed'",
task["id"])
continue
if task["agent_id"] is not None:
logger.info(
"Can't use task %s, it already has an agent assigned",
task["id"])
job_tasks.append({"id": task["id"], "frame": task["frame"]})
if not job_tasks:
logger.error("Could not find any tasks to send for job %s", job["id"])
return
logger.info(
"Found %s tasks from job %s to assign to %r",
len(job_tasks), job["id"], args.agent_api)
assignment_data = {
"job": {
"id": job["id"],
"by": job["by"],
"ram": job["ram"],
"ram_warning": job["ram_warning"],
"ram_max": job["ram_max"],
"cpus": job["cpus"],
"batch": job["batch"],
"user": job["user"],
"data": job["data"],
"environ": job["environ"],
"title": job["title"]},
"jobtype": {
"name": args.jobtype,
"version": jobtype_version},
"tasks": job_tasks}
# Drop any keys which don't have values since this
# would break the schema validation in the agent.
for key in list(assignment_data["job"]):
if assignment_data["job"][key] is None:
del assignment_data["job"][key]
response = session.post(
args.agent_api + "/assign",
data=dumps(assignment_data))
assert response.ok, response.json()
logger.info("Tasks posted to agent")
def test_dumps_uuid(self):
data = {"uuid": uuid4()}
self.assertEqual(dumps(data), dumps({"uuid": str(data["uuid"])}))
def test_dumps_decimal(self):
config["agent_pretty_json"] = False
data = {"decimal": Decimal("1.2")}
self.assertEqual(
dumps(data), dumps_(data, default=default_json_encoder))
def test_dumps_datetime(self):
config["agent_pretty_json"] = False
data = {"datetime": datetime.utcnow()}
self.assertEqual(
dumps(data), dumps_(data, default=default_json_encoder))
def test_dumps_single_argument(self):
config["agent_pretty_json"] = False
data = self.data.keys()[0]
self.assertEqual(dumps(data), dumps_(data))
def test_dumps_not_pretty(self):
config["agent_pretty_json"] = False
self.assertEqual(dumps(self.data), dumps_(self.data))
def test_dumps_pretty(self):
config["agent_pretty_json"] = True
self.assertEqual(dumps(self.data), dumps_(self.data, indent=2))
def post(self, **kwargs):
if request_from_master(kwargs["request"]):
config.master_contacted()
request = kwargs["request"]
request_data = kwargs["data"]
# First, get the resources we have *right now*. In some cases
# this means using the functions in pyfarm.core.sysinfo because
# entries in `config` could be slightly out of sync with the system.
memory_free = free_ram()
cpus = config["agent_cpus"]
requires_ram = request_data["job"].get("ram")
requires_cpus = request_data["job"].get("cpus")
if ("agent_id" in request_data and
request_data["agent_id"] != config["agent_id"]):
logger.error("Wrong agent_id in assignment: %s. Our id is %s",
request_data["agent_id"], config["agent_id"])
return (
dumps({"error": "You have the wrong agent. "
"I am %s." % config["agent_id"],
"agent_id": config["agent_id"]}),
BAD_REQUEST
)
elif self.agent.reannounce_lock.locked:
logger.warning("Temporarily rejecting assignment because we "
"are in the middle of a reannounce.")
return (
dumps({"error": "Agent cannot accept assignments because of a "
"reannounce in progress. Try again shortly."}),
SERVICE_UNAVAILABLE
)
elif self.agent.shutting_down:
logger.error("Rejecting assignment because the agent is in the "
"process of shutting down.")
return (
dumps({"error": "Agent cannot accept assignments because it is "
"shutting down."}),
SERVICE_UNAVAILABLE
)
elif "restart_requested" in config \
and config["restart_requested"] is True:
logger.error("Rejecting assignment because of scheduled restart.")
return (
dumps({"error": "Agent cannot accept assignments because of a "
"pending restart."}),
SERVICE_UNAVAILABLE
)
elif "agent_id" not in config:
logger.error(
"Agent has not yet connected to the master or `agent_id` "
"has not been set yet.")
return (
dumps({"error": "agent_id has not been set in the config"}),
SERVICE_UNAVAILABLE
)
# Do we have enough ram?
elif requires_ram is not None and requires_ram > memory_free:
logger.error(
"Task %s requires %sMB of ram, this agent has %sMB free. "
"Rejecting Task %s.",
request_data["job"]["id"], requires_ram, memory_free,
request_data["job"]["id"])
config["free_ram"] = memory_free
return (
dumps({"error": "Not enough ram",
"agent_ram": memory_free,
"requires_ram": requires_ram}),
BAD_REQUEST
)
# Do we have enough cpus (count wise)?
elif requires_cpus is not None and requires_cpus > cpus:
logger.error(
"Task %s requires %s CPUs, this agent has %s CPUs. "
"Rejecting Task %s.",
request_data["job"]["id"], requires_cpus, cpus,
request_data["job"]["id"])
return (
dumps({"error": "Not enough cpus",
"agent_cpus": cpus,
"requires_cpus": requires_cpus}),
BAD_REQUEST
)
new_task_ids = set(task["id"] for task in request_data["tasks"])
for assignment in config["current_assignments"].itervalues():
existing_task_ids = set(x["id"] for x in assignment["tasks"])
# If the assignment is identical to one we already have
if existing_task_ids == new_task_ids:
logger.debug(
"Ignoring repeated assignment of the same batch")
return dumps({"id": assignment["id"]}), ACCEPTED
# If there is only a partial overlap
elif existing_task_ids & new_task_ids:
logger.error("Rejecting assignment with partial overlap with "
"existing assignment.")
unknown_task_ids = new_task_ids - existing_task_ids
return (
dumps({"error": "Partial overlap of tasks",
"rejected_task_ids": list(unknown_task_ids)}),
CONFLICT
)
if not config["agent_allow_sharing"]:
for jobtype in config["jobtypes"].itervalues():
num_finished_tasks = (len(jobtype.finished_tasks) +
len(jobtype.failed_tasks))
if len(jobtype.assignment["tasks"]) > num_finished_tasks:
logger.error("Rejecting an assignment that would require "
"agent sharing")
return (
dumps({
"error": "Agent does not allow multiple "
"assignments",
"rejected_task_ids": list(new_task_ids)}),
CONFLICT
)
assignment_uuid = uuid4()
request_data.update(id=assignment_uuid)
config["current_assignments"][assignment_uuid] = request_data
logger.debug("Accepted assignment %s: %r",
assignment_uuid, request_data)
logger.info("Accept assignment from job %s with %s tasks",
request_data["job"]["title"], len(request_data["tasks"]))
def assignment_failed(result, assign_id):
logger.error(
"Assignment %s failed, result: %r, removing.", assign_id, result)
logger.error(result.getTraceback())
if (len(config["current_assignments"]) <= 1 and
not self.agent.shutting_down):
config["state"] = AgentState.ONLINE
self.agent.reannounce(force=True)
# Do not mark the assignment as failed if the reason for failing
# was that we ran out of disk space
failed = not isinstance(result.value, InsufficientSpaceError)
assignment = config["current_assignments"].pop(assign_id)
if "jobtype" in assignment:
jobtype_id = assignment["jobtype"].pop("id", None)
if jobtype_id:
instance = config["jobtypes"].pop(jobtype_id, None)
instance.stop(
assignment_failed=failed,
avoid_reassignment=not failed,
error="Error in jobtype: %r. "
"Traceback: %s" % (result,
traceback.format_exc()))
def assignment_started(_, assign_id):
logger.debug("Assignment %s has started", assign_id)
config["state"] = AgentState.RUNNING
self.agent.reannounce(force=True)
def remove_assignment(_, assign_id):
assignment = config["current_assignments"].pop(assign_id)
if "jobtype" in assignment:
jobtype_id = assignment["jobtype"].pop("id", None)
if jobtype_id:
config["jobtypes"].pop(jobtype_id, None)
def assignment_stopped(_, assign_id):
logger.debug("Assignment %s has stopped", assign_id)
if (len(config["current_assignments"]) <= 1 and
not self.agent.shutting_down):
config["state"] = AgentState.ONLINE
self.agent.reannounce(force=True)
assignment = config["current_assignments"][assign_id]
if "jobtype" in assignment:
jobtype_id = assignment["jobtype"].pop("id", None)
if jobtype_id:
jobtype = config["jobtypes"].pop(jobtype_id, None)
updates_deferred = DeferredList(
jobtype.task_update_deferreds)
updates_deferred.addBoth(remove_assignment, assign_id)
else:
config["current_assignments"].pop(assign_id)
def restart_if_necessary(_): # pragma: no cover
if "restart_requested" in config and config["restart_requested"]:
stopping = config["agent"].stop()
stopping.addCallbacks(lambda _: reactor.stop(),
lambda _: reactor.stop())
def load_jobtype_failed(result, assign_id):
logger.error(
"Loading jobtype for assignment %s failed, removing.", assign_id)
traceback = result.getTraceback()
logger.debug("Got traceback")
logger.error(traceback)
assignment = config["current_assignments"].pop(assign_id)
# Mark all tasks as failed on master and set an error message
logger.debug("Marking tasks in assignment as failed")
def post_update(post_url, post_data, task, delay=0):
post_func = partial(post, post_url, data=post_data,
callback=lambda x: result_callback(
post_url, post_data, task, x),
errback=lambda x: error_callback(
post_url, post_data, task, x))
reactor.callLater(delay, post_func)
def result_callback(cburl, cbdata, task, response):
if 500 <= response.code < 600:
logger.error(
"Error while marking task %s as failed on master, "
"retrying", task["id"])
post_update(cburl, cbdata, task, delay=http_retry_delay())
elif response.code != OK:
logger.error(
"Could not mark task %s as failed, server response "
"code was %s", task["id"], response.code)
else:
logger.info(
"Marked task %s as failed on master", task["id"])
def error_callback(cburl, cbdata, task, failure_reason):
logger.error(
"Error while marking task %s as failed, retrying",
task["id"], failure_reason)
post_update(cburl, cbdata, task, delay=http_retry_delay())
for task in assignment["tasks"]:
url = "%s/jobs/%s/tasks/%s" % (
config["master_api"], assignment["job"]["id"], task["id"])
data = {
"state": WorkState.FAILED,
"last_error": traceback}
post_update(url, data, task)
# If the loading was partially successful for some reason, there
# might already be an entry for this jobtype in the config.
# Remove it if it exists.
if "jobtype" in assignment:
jobtype_id = assignment["jobtype"].pop("id", None)
if jobtype_id:
config["jobtypes"].pop(jobtype_id, None)
def loaded_jobtype(jobtype_class, assign_id):
# TODO: report error to master
if hasattr(jobtype_class, "getTraceback"):
logger.error(jobtype_class.getTraceback())
return
# TODO: add call to prepare_for_job
# TODO: add call to spawn_persistent_process
# Instance the job type and pass in the assignment data.
instance = jobtype_class(request_data)
if not isinstance(instance, JobType):
raise TypeError(
"Expected a subclass of "
"pyfarm.jobtypes.core.jobtype.JobType")
# TODO: add callback to cleanup_after_job
# TODO: add callback to stop persistent process
try:
started_deferred, stopped_deferred = instance._start()
started_deferred.addCallback(assignment_started, assign_id)
started_deferred.addErrback(assignment_failed, assign_id)
stopped_deferred.addCallback(assignment_stopped, assign_id)
stopped_deferred.addErrback(assignment_failed, assign_id)
stopped_deferred.addBoth(restart_if_necessary)
stopped_deferred.addBoth(
lambda *args: instance._remove_tempdirs())
stopped_deferred.addBoth(
lambda *args: instance._close_logs())
stopped_deferred.addBoth(
lambda *args: instance._upload_logfile())
except Exception as e:
logger.error("Error on starting jobtype, stopping it now. "
"Error was: %r. Traceback: %s", e,
traceback.format_exc())
instance.stop(assignment_failed=True,
error="Error while loading jobtype: %r. "
"Traceback: %s" %
(e, traceback.format_exc()))
assignment = config["current_assignments"].pop(assign_id)
if "jobtype" in assignment:
jobtype_id = assignment["jobtype"].pop("id", None)
if jobtype_id:
config["jobtypes"].pop(jobtype_id, None)
# Load the job type then pass the class along to the
# callback. No errback here because all the errors
# are handled internally in this case.
jobtype_loader = JobType.load(request_data)
jobtype_loader.addCallback(loaded_jobtype, assignment_uuid)
jobtype_loader.addErrback(load_jobtype_failed, assignment_uuid)
return dumps({"id": assignment_uuid}), ACCEPTED