def check_and_fix_integrity(wf_ex):
check_after_seconds = CONF.engine.execution_integrity_check_delay
if check_after_seconds < 0:
# Never check integrity if it's a negative value.
return
# To break cyclic dependency.
from mistral.engine import task_handler
running_task_execs = db_api.get_task_executions(
workflow_execution_id=wf_ex.id,
state=states.RUNNING
)
for t_ex in running_task_execs:
# The idea is that we take the latest known timestamp of the task
# execution and consider it eligible for checking and fixing only
# if some minimum period of time elapsed since the last update.
timestamp = t_ex.updated_at or t_ex.created_at
delta = timeutils.delta_seconds(timestamp, timeutils.utcnow())
if delta < check_after_seconds:
continue
child_executions = t_ex.executions
if not child_executions:
continue
all_finished = all(
[states.is_completed(c_ex.state) for c_ex in child_executions]
)
if all_finished:
# Find the timestamp of the most recently finished child.
most_recent_child_timestamp = max(
[c_ex.updated_at or c_ex.created_at for c_ex in
child_executions]
)
interval = timeutils.delta_seconds(
most_recent_child_timestamp,
timeutils.utcnow()
)
if interval > check_after_seconds:
# We found a task execution in RUNNING state for which all
# child executions are finished. We need to call
# "schedule_on_action_complete" on the task handler for any of
# the child executions so that the task state is calculated and
# updated properly.
LOG.warning(
"Found a task execution that is likely stuck in RUNNING"
" state because all child executions are finished,"
" will try to recover [task_execution=%s]", t_ex.id
)
task_handler.schedule_on_action_complete(child_executions[-1])
def on_action_complete(action_ex, result):
task_ex = action_ex.task_execution
action = _build_action(action_ex)
try:
action.complete(result)
except exc.MistralException as e:
msg = ("Failed to complete action [action=%s, task=%s]: %s\n%s" %
(action_ex.name, task_ex.name, e, tb.format_exc()))
LOG.error(msg)
action.fail(msg)
if task_ex:
task_handler.force_fail_task(task_ex, msg)
return
if task_ex:
task_handler.schedule_on_action_complete(action_ex)
def on_action_complete(action_ex, result):
task_ex = action_ex.task_execution
action = _build_action(action_ex)
try:
action.complete(result)
except exc.MistralException as e:
msg = (
"Failed to complete action [error=%s, action=%s, task=%s]:\n%s"
% (e, action_ex.name, task_ex.name, tb.format_exc())
)
LOG.error(msg)
action.fail(msg)
if task_ex:
task_handler.force_fail_task(task_ex, msg)
return
if task_ex:
task_handler.schedule_on_action_complete(action_ex)
def _check_and_fix_integrity(wf_ex_id):
check_after_seconds = CONF.engine.execution_integrity_check_delay
if check_after_seconds < 0:
# Never check integrity if it's a negative value.
return
# To break cyclic dependency.
from mistral.engine import task_handler
with db_api.transaction():
wf_ex = db_api.get_workflow_execution(wf_ex_id)
if states.is_completed(wf_ex.state):
return
_schedule_check_and_fix_integrity(wf_ex, delay=120)
running_task_execs = db_api.get_task_executions(
workflow_execution_id=wf_ex.id,
state=states.RUNNING,
limit=CONF.engine.execution_integrity_check_batch_size
)
for t_ex in running_task_execs:
# The idea is that we take the latest known timestamp of the task
# execution and consider it eligible for checking and fixing only
# if some minimum period of time elapsed since the last update.
timestamp = t_ex.updated_at or t_ex.created_at
delta = timeutils.delta_seconds(timestamp, timeutils.utcnow())
if delta < check_after_seconds:
continue
child_executions = t_ex.executions
if not child_executions:
continue
all_finished = all(
[states.is_completed(c_ex.state) for c_ex in child_executions]
)
if all_finished:
# Find the timestamp of the most recently finished child.
most_recent_child_timestamp = max(
[c_ex.updated_at or c_ex.created_at for c_ex in
child_executions]
)
interval = timeutils.delta_seconds(
most_recent_child_timestamp,
timeutils.utcnow()
)
if interval > check_after_seconds:
# We found a task execution in RUNNING state for which all
# child executions are finished. We need to call
# "schedule_on_action_complete" on the task handler for
# any of the child executions so that the task state is
# calculated and updated properly.
LOG.warning(
"Found a task execution that is likely stuck in"
" RUNNING state because all child executions are"
" finished, will try to recover [task_execution=%s]",
t_ex.id
)
task_handler.schedule_on_action_complete(
child_executions[-1]
)
