def _handle_success(ns, q, lock):
qb.set_state(app_name=ns.app_name, job_id=ns.job_id, completed=True)
q.consume()
lock.release()
log.info("successfully completed job",
extra=dict(app_name=ns.app_name, job_id=ns.job_id,
completed=True))
def test_stolos_submit_readd(app1, job_id1, tasks_json_tmpfile):
qb.set_state(app1, job_id1, failed=True)
validate_zero_queued_task(app1)
run("stolos-submit -a %s -j %s" % (app1, job_id1), tasks_json_tmpfile)
validate_zero_queued_task(app1)
run("stolos-submit -a %s -j %s --readd" % (app1, job_id1),
tasks_json_tmpfile)
validate_one_queued_task(app1, job_id1)
def _handle_success(ns, q, lock):
qb.set_state(
app_name=ns.app_name, job_id=ns.job_id, completed=True)
q.consume()
lock.release()
log.info(
"successfully completed job",
extra=dict(app_name=ns.app_name, job_id=ns.job_id, completed=True))
def test_stolos_submit_readd(app1, job_id1, tasks_json_tmpfile):
qb.set_state(app1, job_id1, failed=True)
validate_zero_queued_task(app1)
run("stolos-submit -a %s -j %s" % (app1, job_id1),
tasks_json_tmpfile)
validate_zero_queued_task(app1)
run("stolos-submit -a %s -j %s --readd" % (app1, job_id1),
tasks_json_tmpfile)
validate_one_queued_task(app1, job_id1)
def test_rerun_push_tasks1(app1, app2, job_id1):
# this tests recursively deleteing parent status on child nodes
# queue and complete app 1. it queues a child
enqueue(app1, job_id1)
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
# complete app 2
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
# readd app 1
qb.readd_subtask(app1, job_id1)
validate_one_queued_task(app1, job_id1)
validate_zero_queued_task(app2)
nose.tools.assert_true(
qb.check_state(app2, job_id1, pending=True))
# complete app 1
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
# complete app 2
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
def test_create_child_task_after_one_parent_completed(
app1, app2, app3, job_id1, log, tasks_json_tmpfile, func_name):
# if you modify the tasks.json file in the middle of processing the dag
# modifications to the json file should be recognized
# the child task should run if another parent completes
# but otherwise should not run until it's manually queued
injected_app = app3
dct = {
injected_app: {
"depends_on": {"app_name": [app1, app2]},
},
}
qb.set_state(
app1, job_id1, completed=True)
validate_one_completed_task(app1, job_id1)
validate_zero_queued_task(injected_app)
consume_queue(app2)
qb.set_state(app2, job_id1, completed=True)
with inject_into_dag(func_name, dct):
validate_zero_queued_task(injected_app)
qb.set_state(app2, job_id1, completed=True)
validate_zero_queued_task(injected_app)
qb.set_state(app1, job_id1, completed=True)
validate_one_queued_task(injected_app, job_id1)
run_code(log, tasks_json_tmpfile, injected_app, '--bash_cmd echo 123')
validate_one_completed_task(injected_app, job_id1)
def test_race_condition_when_parent_queues_child(
app1, app2, job_id1, log, tasks_json_tmpfile):
# The parent queues the child and the child runs before the parent gets
# a chance to mark itself as completed
qb.set_state(app1, job_id1, pending=True)
lock = qb.obtain_execute_lock(app1, job_id1)
assert lock
qb.set_state(app1, job_id1, completed=True)
qb.set_state(app1, job_id1, pending=True)
validate_one_queued_task(app2, job_id1)
validate_zero_queued_task(app1)
# should not complete child. should de-queue child
# should not queue parent.
# should exit gracefully
# stays in the queue (forever) until parent state is completed
run_code(log, tasks_json_tmpfile, app2)
validate_zero_queued_task(app1)
validate_one_queued_task(app2, job_id1)
qb.set_state(
app1, job_id1, completed=True,
_disable_maybe_queue_children_for_testing_only=True)
lock.release()
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
run_code(log, tasks_json_tmpfile, app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
def test_complex_dependencies_pull_push(
depends_on1, depends_on_job_id1, log, tasks_json_tmpfile):
job_id = depends_on_job_id1
enqueue(depends_on1, job_id)
run_code(log, tasks_json_tmpfile, depends_on1, '--bash_cmd echo 123')
parents = api.get_parents(depends_on1, job_id)
parents = list(api.topological_sort(parents))
for parent, pjob_id in parents[:-1]:
qb.set_state(parent, pjob_id, completed=True)
validate_zero_queued_task(depends_on1)
qb.set_state(*parents[-1], completed=True)
validate_one_queued_task(depends_on1, job_id)
run_code(log, tasks_json_tmpfile, depends_on1, '--bash_cmd echo 123')
validate_one_completed_task(depends_on1, job_id)
def test_rerun_push_tasks_when_manually_queuing_child_and_parent(
app1, app2, job_id1):
_test_rerun_tasks_when_manually_queuing_child_and_parent(
app1, app2, job_id1)
# complete parent first
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
# child completes normally
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
def test_check_state1(app1, job_id1, job_id2):
nt.assert_false(api.check_state(app1, 'doesnotexist', pending=True))
with nt.assert_raises(NoNodeError):
api.check_state(app1, 'doesnotexist', raise_if_not_exists=True)
qb.set_state(app1, job_id1, pending=True)
# also: create an invalid state (one that stolos does not recognize)
api.get_qbclient().create(qb.get_job_path(app1, job_id2), '')
with nt.assert_raises(UserWarning):
api.check_state(app1, job_id1)
nt.assert_true(api.check_state(app1, job_id1, pending=True))
nt.assert_true(api.check_state(app1, job_id1, pending=True,
completed=True))
nt.assert_false(api.check_state(app1, job_id1, completed=True))
nt.assert_true(api.check_state(app1, job_id1, all=True))
# the invalid job:
nt.assert_false(api.check_state(app1, job_id2, all=True))
def test_complex_dependencies_readd(depends_on1, depends_on_job_id1,
log, tasks_json_tmpfile):
job_id = depends_on_job_id1
# mark everything completed. ensure only last completed parent queues child
parents = list(api.topological_sort(api.get_parents(depends_on1, job_id)))
for parent, pjob_id in parents[:-1]:
qb.set_state(parent, pjob_id, completed=True)
validate_zero_queued_task(depends_on1)
qb.set_state(parents[-1][0], parents[-1][1], completed=True)
validate_one_queued_task(depends_on1, job_id)
# --> parents should queue our app
consume_queue(depends_on1)
qb.set_state(depends_on1, job_id, completed=True)
validate_one_completed_task(depends_on1, job_id)
# ok great - ran through pipeline once.
log.warn("OK... Now try complex dependency test with a readd")
# re-complete the very first parent.
# we assume that this parent is a root task
parent, pjob_id = parents[0]
qb.readd_subtask(parent, pjob_id)
validate_one_queued_task(parent, pjob_id)
validate_zero_queued_task(depends_on1)
consume_queue(parent)
qb.set_state(parent, pjob_id, completed=True)
validate_one_completed_task(parent, pjob_id)
# since the parent that re-queues children that may be depends_on1's
# parents, complete those too!
for p2, pjob2 in api.get_children(parent, pjob_id, False):
if p2 == depends_on1:
continue
consume_queue(p2)
qb.set_state(p2, pjob2, completed=True)
# now, that last parent should have queued our application
validate_n_queued_task(
depends_on1, job_id,
job_id.replace('testID1', 'testID3')) # this replace is a hack
run_code(log, tasks_json_tmpfile, depends_on1, '--bash_cmd echo 123')
run_code(log, tasks_json_tmpfile, depends_on1, '--bash_cmd echo 123')
validate_one_completed_task(depends_on1, job_id)
def _handle_manually_given_job_id(ns):
"""This process was given a specific --job_id arg.
Decide whether it's okay to execute this job_id,
and if its okay to go forward, set job_id state appropriately
"""
log.warn(
('using specific job_id and'
' blindly assuming this job is not already queued.'),
extra=dict(app_name=ns.app_name, job_id=ns.job_id))
if qb.get_qbclient().exists(qb.get_job_path(ns.app_name, ns.job_id)):
msg = ('Will not execute this task because it might be already'
' queued or completed!')
log.critical(
msg, extra=dict(app_name=ns.app_name, job_id=ns.job_id))
raise UserWarning(msg)
lock = qb.obtain_execute_lock(
ns.app_name, ns.job_id, safe=False, raise_on_error=True,
blocking=False)
qb.set_state(ns.app_name, ns.job_id, pending=True)
return lock
def _handle_manually_given_job_id(ns):
"""This process was given a specific --job_id arg.
Decide whether it's okay to execute this job_id,
and if its okay to go forward, set job_id state appropriately
"""
log.warn(('using specific job_id and'
' blindly assuming this job is not already queued.'),
extra=dict(app_name=ns.app_name, job_id=ns.job_id))
if qb.get_qbclient().exists(qb.get_job_path(ns.app_name, ns.job_id)):
msg = ('Will not execute this task because it might be already'
' queued or completed!')
log.critical(msg, extra=dict(app_name=ns.app_name, job_id=ns.job_id))
raise UserWarning(msg)
lock = qb.obtain_execute_lock(ns.app_name,
ns.job_id,
safe=False,
raise_on_error=True,
blocking=False)
qb.set_state(ns.app_name, ns.job_id, pending=True)
return lock
def _test_rerun_tasks_when_manually_queuing_child_and_parent(
app1, app2, job_id1):
# complete parent and child
enqueue(app1, job_id1)
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
# manually re-add child
qb.readd_subtask(app2, job_id1)
validate_one_queued_task(app2, job_id1)
validate_one_completed_task(app1, job_id1)
# manually re-add parent
qb.readd_subtask(app1, job_id1)
validate_one_queued_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
def test_rerun_pull_tasks1(app1, app2, job_id1, log, tasks_json_tmpfile):
# queue and complete app 1. it queues a child
enqueue(app1, job_id1)
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
validate_zero_queued_task(app1)
validate_one_queued_task(app2, job_id1)
# complete app 2
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_zero_queued_task(app2)
# readd app 2
qb.readd_subtask(app2, job_id1)
validate_zero_queued_task(app1)
validate_one_queued_task(app2, job_id1)
# run app 2. the parent was previously completed
run_code(log, tasks_json_tmpfile, app2)
validate_one_completed_task(app1, job_id1) # previously completed
validate_one_completed_task(app2, job_id1)
def test_check_state1(app1, job_id1, job_id2):
nt.assert_false(api.check_state(app1, 'doesnotexist', pending=True))
with nt.assert_raises(NoNodeError):
api.check_state(app1, 'doesnotexist', raise_if_not_exists=True)
qb.set_state(app1, job_id1, pending=True)
# also: create an invalid state (one that stolos does not recognize)
api.get_qbclient().create(
qb.get_job_path(app1, job_id2), '')
with nt.assert_raises(UserWarning):
api.check_state(app1, job_id1)
nt.assert_true(
api.check_state(app1, job_id1, pending=True))
nt.assert_true(
api.check_state(app1, job_id1, pending=True, completed=True))
nt.assert_false(api.check_state(app1, job_id1, completed=True))
nt.assert_true(api.check_state(app1, job_id1, all=True))
# the invalid job:
nt.assert_false(api.check_state(app1, job_id2, all=True))
def test_create_parent_task_after_child_completed(app1, app3, job_id1,
func_name):
# if you modify the tasks.json file in the middle of processing the dag
# modifications to the json file should be recognized appropriately
# we do not re-schedule the child unless parent is completed
qb.set_state(app1, job_id1, completed=True)
validate_one_completed_task(app1, job_id1)
injected_app = app3
child_injapp = 'test_stolos/%s/testX' % func_name
dct = {
injected_app: {
},
child_injapp: {
"depends_on": {"app_name": [injected_app]}
}
}
with inject_into_dag(func_name, dct):
validate_zero_queued_task(injected_app)
qb.set_state(injected_app, job_id1, completed=True)
validate_one_completed_task(injected_app, job_id1)
validate_one_queued_task(child_injapp, job_id1)
def test_pull_tasks_with_many_children(app1, app2, app3, app4, job_id1,
log, tasks_json_tmpfile):
enqueue(app4, job_id1)
validate_one_queued_task(app4, job_id1)
validate_zero_queued_task(app1)
validate_zero_queued_task(app2)
validate_zero_queued_task(app3)
run_code(log, tasks_json_tmpfile, app4, '--bash_cmd echo app4helloworld')
validate_zero_queued_task(app4)
validate_one_queued_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
validate_one_queued_task(app3, job_id1)
consume_queue(app1)
qb.set_state(app1, job_id1, completed=True)
validate_zero_queued_task(app4)
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
validate_one_queued_task(app3, job_id1)
consume_queue(app2)
qb.set_state(app2, job_id1, completed=True)
validate_zero_queued_task(app4)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
validate_one_queued_task(app3, job_id1)
consume_queue(app3)
qb.set_state(app3, job_id1, completed=True)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
validate_one_completed_task(app3, job_id1)
validate_one_queued_task(app4, job_id1)
consume_queue(app4)
qb.set_state(app4, job_id1, completed=True)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
validate_one_completed_task(app3, job_id1)
validate_one_completed_task(app4, job_id1)
def test_check_state2(app1, job_id1, job_id2, job_id3):
"""Does check_state support multiple job_ids?"""
qb.set_state(app1, job_id1, pending=True)
qb.set_state(app1, job_id2, completed=True)
qb.set_state(app1, job_id3, failed=True)
nt.assert_list_equal([True, True],
api.check_state(app1, [job_id1, job_id2],
pending=True,
completed=True,
raise_if_not_exists=True))
nt.assert_list_equal([True, True],
api.check_state(app1, [job_id1, job_id2],
all=True,
raise_if_not_exists=True))
def test_skipped_parent_and_queued_child(app1, app2, app3, app4, job_id1,
log, tasks_json_tmpfile):
qb.set_state(app1, job_id1, skipped=True)
qb.set_state(app3, job_id1, skipped=True)
qb.set_state(app2, job_id1, skipped=True)
enqueue(app4, job_id1)
validate_zero_queued_task(app2)
validate_one_queued_task(app4, job_id1)
# ensure child unqueues itself and raises warning
out, err = run_code(log, tasks_json_tmpfile, app4, capture=True)
nose.tools.assert_in(
"parent_job_id is marked as 'skipped', so should be impossible for me,"
" the child, to exist", err)
validate_zero_queued_task(app4)
validate_zero_queued_task(app2)
def test_check_state2(app1, job_id1, job_id2, job_id3):
"""Does check_state support multiple job_ids?"""
qb.set_state(app1, job_id1, pending=True)
qb.set_state(app1, job_id2, completed=True)
qb.set_state(app1, job_id3, failed=True)
nt.assert_list_equal(
[True, True],
api.check_state(
app1, [job_id1, job_id2], pending=True, completed=True,
raise_if_not_exists=True
))
nt.assert_list_equal(
[True, True],
api.check_state(
app1, [job_id1, job_id2], all=True,
raise_if_not_exists=True
))
def test_rerun_pull_tasks_when_manually_queuing_child_and_parent(
app1, app2, job_id1):
_test_rerun_tasks_when_manually_queuing_child_and_parent(
app1, app2, job_id1)
# complete child first
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
# --> parent still queued
validate_one_queued_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
# then complete parent
qb.set_state(app1, job_id1, completed=True)
consume_queue(app1)
# --> child gets re-queued
validate_one_completed_task(app1, job_id1)
validate_one_queued_task(app2, job_id1)
# complete child second time
qb.set_state(app2, job_id1, completed=True)
consume_queue(app2)
validate_one_completed_task(app1, job_id1)
validate_one_completed_task(app2, job_id1)
