def task_loop():
try:
# reset tasks
local.task_active = None
local.task_push = None
# get SQS work queue
q = aws.get_sqs_queue(conf)
# Loop over tasks. There are up to two different tasks at any
# given moment that we are processing concurrently:
#
# 1. Active task -- usually a blender render operation.
# 2. S3 push task -- a task which pushes the products of the
# previous active task (such as rendered
# frames) to S3.
while True:
# reset active task
local.task_active = None
# initialize active task object
task = State()
task.msg = None
task.proc = None
task.retcode = None
task.outdir = None
task.id = 0
# Get a task from the SQS work queue. This is normally
# a short script that runs blender to render one
# or more frames.
queuemsg = q.get_messages(message_attributes=['config'])
# output some debug info
print "queue read:", task.msg
if local.task_push:
print "push task:", local.task_push.__dict__
else:
print "no task push task"
# process task
if len(queuemsg) > 0:
task.msg = queuemsg[0]
# assign an ID to task
local.task_id_counter += 1
task.id = local.task_id_counter
# register active task
local.task_active = task
# create output directory
task.outdir = os.path.join(work_dir, "brenda-outdir%d.tmp" % (task.id,))
utils.rmtree(task.outdir)
utils.mkdir(task.outdir)
# Create a config dictionary using combination of global and task-specific config values
taskconfig = conf.copy()
if 'config' in task.msg.message_attributes:
taskconfig.update(json.loads(task.msg.message_attributes['config']['string_value']))
# Store outdir in task config for later use
taskconfig['OUTDIR'] = task.outdir
if not 'BLENDER_FILE' in taskconfig:
taskconfig['BLENDER_FILE'] = '*.blend'
print "task-specific config:", taskconfig
# get the task script
script = task.msg.get_body()
print "script len:", len(script)
# do macro substitution on the task script
for k in taskconfig:
script = script.replace('$' + k, taskconfig[k])
# add shebang if absent
if not script.startswith("#!"):
script = "#!/bin/bash\n" + script
# Make sure we're working with the correct project directory
# FIXME - this is likely not the most efficient way of doing it, and probably leads to unnecessary
# downloads from s3. Ideally we would keep all project directories and switch between them,
# but currently brenda only supports one working project directory at a time
proj_dir = get_project(taskconfig, taskconfig['BLENDER_PROJECT'])
# mount additional EBS volumes
aws.mount_additional_ebs(taskconfig, proj_dir)
# cd to project directory, where we will run blender from
with utils.Cd(proj_dir) as cd:
# write script file and make it executable
script_fn = "./brenda-go"
with open(script_fn, 'w') as f:
f.write(script)
st = os.stat(script_fn)
os.chmod(script_fn, st.st_mode | (stat.S_IEXEC|stat.S_IXGRP|stat.S_IXOTH))
# run the script
print "------- Run script %s -------" % (os.path.realpath(script_fn),)
print script,
print "--------------------------"
task.proc = Subprocess([script_fn])
print "active task:", local.task_active.__dict__
# Wait for active and S3-push tasks to complete,
# while periodically reasserting with SQS to
# acknowledge that tasks are still pending.
# (If we don't reassert with SQS frequently enough,
# it will assume we died, and put our tasks back
# in the queue. "frequently enough" means within
# visibility_timeout.)
count = 0
while True:
reassert = (count >= visibility_timeout_reassert)
for i, task in enumerate((local.task_active, local.task_push)):
if task:
name = task_names[i]
if task.proc is not None:
# test if process has finished
task.retcode = task.proc.poll()
if task.retcode is not None:
# process has finished
task.proc = None
# did process finish with errors?
if task.retcode != 0:
errtxt = "fatal error in %s task" % (name,)
if name == 'active':
raise error.ValueErrorRetry(errtxt)
else:
raise ValueError(errtxt)
# Process finished successfully. If S3-push process,
# tell SQS that the task completed successfully.
if name == 'push':
print "******* TASK", task.id, "COMMITTED to S3"
q.delete_message(task.msg)
task.msg = None
local.task_count += 1
task_complete_accounting(local.task_count)
# active task completed?
if name == 'active':
print "******* TASK", task.id, "READY-FOR-PUSH"
# tell SQS that we are still working on the task
if reassert and task.proc is not None:
print "******* REASSERT", name, task.id
task.msg.change_visibility(visibility_timeout)
# break out of loop only when no pending tasks remain
if ((not local.task_active or local.task_active.proc is None)
and (not local.task_push or local.task_push.proc is None)):
break
# setup for next process poll iteration
if reassert:
count = 0
time.sleep(1)
count += 1
# clean up the S3-push task
cleanup(local.task_push, 'push')
local.task_push = None
# start a concurrent push task to commit files generated by
# just-completed active task (such as blender render frames) to S3
if local.task_active:
local.task_active.proc = start_s3_push_process(opts, args, taskconfig, local.task_active.outdir)
local.task_push = local.task_active
local.task_active = None
# if no active task and no S3-push task, we are done (unless DONE is set to "poll")
if not local.task_active and not local.task_push:
action = read_done_file()
if action == "poll":
print "Polling for more work..."
time.sleep(15)
elif action == "smart":
now = time.time()
try:
instance_id = aws.get_instance_id_self()
spot_request_id = aws.get_spot_request_from_instance_id(conf, instance_id)
launch_time = aws.get_launch_time(conf, spot_request_id)
if launch_time:
spottime = aws.get_uptime(now, launch_time)
minutes_after_hour = (spottime / 60) % 60
print "Smart poll: ", minutes_after_hour
if minutes_after_hour >= smart_shutdown_threshold:
print "Smart poll threshold passed, shutting down (%d minutes after the hour with no work in queue)" % (minutes_after_hour)
# update the value of DONE config var for clean shutdown
conf['DONE'] = 'shutdown'
write_done_file()
break;
else:
print "Smart poll: no launch_time for spot request %s" % (spot_request_id)
except Exception, e:
print "Smart poll failed!", e
time.sleep(15)
else:
break
finally:
cleanup_all()
# get configuration parameters
work_dir = aws.get_work_dir(conf)
visibility_timeout_reassert = int(conf.get('VISIBILITY_TIMEOUT_REASSERT', '30'))
visibility_timeout = int(conf.get('VISIBILITY_TIMEOUT', '120'))
# validate RENDER_OUTPUT bucket
#aws.get_s3_output_bucket(conf)
# file cleanup
utils.rm('task_count')
utils.rm('task_last')
# create Blender temporary directory
tmp_dir = os.path.join(work_dir, 'tmp')
if not os.path.isdir(tmp_dir):
utils.mkdir(tmp_dir)
os.environ['TMP'] = tmp_dir
# save the value of DONE config var
write_done_file()
# Prepare for smart shutdown
smart_shutdown_threshold = int(conf.get('SMART_SHUTDOWN_THRESHOLD', 58))
spot_request_create_time = datetime.datetime.now().isoformat()
# Get our spot instance request, if it exists
spot_request_id = None
if int(conf.get('RUNNING_ON_EC2', '1')):
try:
instance_id = aws.get_instance_id_self()
spot_request_id = aws.get_spot_request_from_instance_id(conf, instance_id)
def task_loop():
try:
# reset tasks
local.task_active = None
local.task_push = None
# get SQS work queue
q = aws.get_sqs_queue(conf)
# Loop over tasks. There are up to two different tasks at any
# given moment that we are processing concurrently:
#
# 1. Active task -- usually a blender render operation.
# 2. S3 push task -- a task which pushes the products of the
# previous active task (such as rendered
# frames) to S3.
while True:
# reset active task
local.task_active = None
# initialize active task object
task = State()
task.msg = None
task.proc = None
task.retcode = None
task.outdir = None
task.id = 0
# Get a task from the SQS work queue. This is normally
# a short script that runs blender to render one
# or more frames.
task.msg = q.read()
# output some debug info
print "queue read:", task.msg
if local.task_push:
print "push task:", local.task_push.__dict__
else:
print "no task push task"
# process task
if task.msg is not None:
# assign an ID to task
local.task_id_counter += 1
task.id = local.task_id_counter
# register active task
local.task_active = task
# create output directory
task.outdir = os.path.join(
work_dir, "brenda-outdir%d.tmp" % (task.id, ))
utils.rmtree(task.outdir)
utils.mkdir(task.outdir)
# get the task script
script = task.msg.get_body()
print "script len:", len(script)
# do macro substitution on the task script
script = script.replace('$OUTDIR', task.outdir)
# add shebang if absent
if not script.startswith("#!"):
script = "#!/bin/bash\n" + script
# cd to project directory, where we will run blender from
with utils.Cd(proj_dir) as cd:
# write script file and make it executable
script_fn = "./brenda-go"
with open(script_fn, 'w') as f:
f.write(script)
st = os.stat(script_fn)
os.chmod(
script_fn, st.st_mode |
(stat.S_IEXEC | stat.S_IXGRP | stat.S_IXOTH))
# run the script
print "------- Run script %s -------" % (
os.path.realpath(script_fn), )
print script,
print "--------------------------"
task.proc = Subprocess([script_fn])
print "active task:", local.task_active.__dict__
# Wait for active and S3-push tasks to complete,
# while periodically reasserting with SQS to
# acknowledge that tasks are still pending.
# (If we don't reassert with SQS frequently enough,
# it will assume we died, and put our tasks back
# in the queue. "frequently enough" means within
# visibility_timeout.)
count = 0
while True:
reassert = (count >= visibility_timeout_reassert)
for i, task in enumerate(
(local.task_active, local.task_push)):
if task:
name = task_names[i]
if task.proc is not None:
# test if process has finished
task.retcode = task.proc.poll()
if task.retcode is not None:
# process has finished
task.proc = None
# did process finish with errors?
if task.retcode != 0:
errtxt = "fatal error in %s task" % (
name, )
if name == 'active':
raise error.ValueErrorRetry(errtxt)
else:
raise ValueError(errtxt)
# Process finished successfully. If S3-push process,
# tell SQS that the task completed successfully.
if name == 'push':
print "******* TASK", task.id, "COMMITTED to S3"
q.delete_message(task.msg)
task.msg = None
local.task_count += 1
task_complete_accounting(
local.task_count)
# active task completed?
if name == 'active':
print "******* TASK", task.id, "READY-FOR-PUSH"
# tell SQS that we are still working on the task
if reassert and task.proc is not None:
print "******* REASSERT", name, task.id
task.msg.change_visibility(visibility_timeout)
# break out of loop only when no pending tasks remain
if ((not local.task_active
or local.task_active.proc is None) and
(not local.task_push or local.task_push.proc is None)):
break
# setup for next process poll iteration
if reassert:
count = 0
time.sleep(1)
count += 1
# clean up the S3-push task
cleanup(local.task_push, 'push')
local.task_push = None
# start a concurrent push task to commit files generated by
# just-completed active task (such as blender render frames) to S3
if local.task_active:
local.task_active.proc = start_s3_push_process(
opts, args, conf, local.task_active.outdir)
local.task_push = local.task_active
local.task_active = None
# if no active task and no S3-push task, we are done (unless DONE is set to "poll")
if not local.task_active and not local.task_push:
if read_done_file() == "poll":
print "Polling for more work..."
time.sleep(15)
else:
break
finally:
cleanup_all()
work_dir = aws.get_work_dir(conf)
visibility_timeout_reassert = int(
conf.get('VISIBILITY_TIMEOUT_REASSERT', '30'))
visibility_timeout = int(conf.get('VISIBILITY_TIMEOUT', '120'))
# validate RENDER_OUTPUT bucket
aws.get_s3_output_bucket(conf)
# file cleanup
utils.rm('task_count')
utils.rm('task_last')
# create Blender temporary directory
tmp_dir = os.path.join(work_dir, 'tmp')
if not os.path.isdir(tmp_dir):
utils.mkdir(tmp_dir)
os.environ['TMP'] = tmp_dir
# save the value of DONE config var
write_done_file()
# Get our spot instance request, if it exists
spot_request_id = None
if int(conf.get('RUNNING_ON_EC2', '1')):
try:
instance_id = aws.get_instance_id_self()
spot_request_id = aws.get_spot_request_from_instance_id(
conf, instance_id)
print "Spot request ID:", spot_request_id
except Exception, e:
print "Error determining spot instance request:", e
def task_loop():
try:
# reset tasks
local.task_render = None
local.task_upload = None
# get SQS work queue
q = aws.get_sqs_conn_queue(conf)[0]
# Loop over tasks. There are up to two different tasks at any
# given moment that we are processing concurrently:
#
# 1. Render task -- usually a render operation.
# 2. Upload task -- a task which uploads results to S3.
while True:
# reset render task
local.task_render = None
# initialize render task object
task = State()
task.msg = None
task.proc = None
task.retcode = None
task.outdir = None
task.id = 0
task.script_name = None
# Get a task from the SQS work queue. This is normally
# a short script that renders one or more frames.
task.msg = q.read(message_attributes=['All'])
# output some debug info
logging.debug('Reading work queue')
if local.task_upload:
logging.info("Running upload task #%d", local.task_upload.id)
logging.debug(local.task_upload.__dict__)
else:
logging.info('No upload task available')
# process task
if task.msg is not None:
# assign an ID to task
local.task_id_counter += 1
task.id = local.task_id_counter
task.script_name = task.msg.message_attributes['script_name']['string_value']
# register render task
local.task_render = task
# create output directory
task.outdir = os.path.join(work_dir, "{}_out_{}".format(task.script_name, task.id))
utils.rmtree(task.outdir)
utils.mkdir(task.outdir)
# get the task script
script = task.msg.get_body()
# cd to output directory, where we will run render task from
with utils.Cd(task.outdir):
# write script file and make it executable
script_fn = "./{}".format(task.script_name)
with open(script_fn, 'w') as f:
f.write(script)
st = os.stat(script_fn)
os.chmod(script_fn, st.st_mode | (stat.S_IEXEC|stat.S_IXGRP|stat.S_IXOTH))
# run the script
task.proc = Subprocess([script_fn])
logging.info('Running render task \"%s #%d\"', local.task_render.script_name, local.task_render.id)
logging.info(script.replace("\n"," "))
logging.debug(local.task_render.__dict__)
# Wait for render & upload tasks to complete, while periodically reasserting with SQS to
# acknowledge that tasks are still pending. (If we don't reassert with SQS frequently enough,
# it will assume we died, and put our tasks back in the queue. "frequently enough" means within
# visibility_timeout.)
count = 0
while True:
reassert = (count >= visibility_timeout_reassert)
for i, task in enumerate((local.task_render, local.task_upload)):
if task:
name = task_names[i]
if task.proc is not None:
# test if process has finished
task.retcode = task.proc.poll()
if task.retcode is not None:
# process has finished
task.proc = None
# did process finish with errors?
if task.retcode != 0:
if name == 'render':
errtxt = "Render task \"{} #{}\" exited with status code {}".format(
task.script_name, task.id, task.retcode)
raise error.ValueErrorRetry(errtxt)
else:
errtxt = "Upload task #{} exited with status code {}".format(
task.id, task.retcode)
raise ValueError(errtxt)
# Process finished successfully. If upload process,
# tell SQS that the task completed successfully.
if name == 'upload':
logging.info('Finished upload task #%d', task.id)
q.delete_message(task.msg)
task.msg = None
local.task_count += 1
task_complete_accounting(local.task_count)
# Render task completed?
if name == 'render':
logging.info('Finished render task \"%s #%d\"', task.script_name, task.id)
# tell SQS that we are still working on the task
if reassert and task.proc is not None:
logging.debug('Reasserting %s task %d with SQS', name, task.id)
task.msg.change_visibility(visibility_timeout)
# break out of loop only when no pending tasks remain
if ((not local.task_render or local.task_render.proc is None)
and (not local.task_upload or local.task_upload.proc is None)):
break
# setup for next process poll iteration
if reassert:
count = 0
time.sleep(1)
count += 1
# clean up the upload task
cleanup(local.task_upload, 'upload')
local.task_upload = None
# start a concurrent upload task to commit files generated by just-completed render task to S3
if local.task_render:
local.task_render.proc = start_upload_process(opts, args, conf, local.task_render)
local.task_upload = local.task_render
local.task_render = None
# if no render or upload task, we are done (unless DONE is set to "poll")
if not local.task_render and not local.task_upload:
if read_done_file() == "poll":
logging.info('Waiting for tasks...')
time.sleep(15)
else:
logging.info('Exiting')
break
finally:
cleanup_all()
def task_loop():
try:
# reset tasks
local.task_active = None
local.task_push = None
# get SQS work queue
q = aws.get_sqs_queue(conf)
# Loop over tasks. There are up to two different tasks at any
# given moment that we are processing concurrently:
#
# 1. Active task -- usually a blender render operation.
# 2. S3 push task -- a task which pushes the products of the
# previous active task (such as rendered
# frames) to S3.
while True:
# reset active task
local.task_active = None
# initialize active task object
task = State()
task.msg = None
task.proc = None
task.retcode = None
task.outdir = None
task.id = 0
# Get a task from the SQS work queue. This is normally
# a short script that runs blender to render one
# or more frames.
task.msg = q.read()
# output some debug info
print "queue read:", task.msg
if local.task_push:
print "push task:", local.task_push.__dict__
else:
print "no task push task"
# process task
if task.msg is not None:
# assign an ID to task
local.task_id_counter += 1
task.id = local.task_id_counter
# register active task
local.task_active = task
# create output directory
task.outdir = os.path.join(work_dir, "brenda-outdir%d.tmp" % (task.id,))
utils.rmtree(task.outdir)
utils.mkdir(task.outdir)
# get the task script
script = task.msg.get_body()
print "script len:", len(script)
# do macro substitution on the task script
script = script.replace('$OUTDIR', task.outdir)
# add shebang if absent
if not script.startswith("#!"):
script = "#!/bin/bash\n" + script
# cd to project directory, where we will run blender from
with utils.Cd(proj_dir) as cd:
# write script file and make it executable
script_fn = "./brenda-go"
with open(script_fn, 'w') as f:
f.write(script)
st = os.stat(script_fn)
os.chmod(script_fn, st.st_mode | (stat.S_IEXEC|stat.S_IXGRP|stat.S_IXOTH))
# run the script
print "------- Run script %s -------" % (os.path.realpath(script_fn),)
print script,
print "--------------------------"
task.proc = Subprocess([script_fn])
print "active task:", local.task_active.__dict__
# Wait for active and S3-push tasks to complete,
# while periodically reasserting with SQS to
# acknowledge that tasks are still pending.
# (If we don't reassert with SQS frequently enough,
# it will assume we died, and put our tasks back
# in the queue. "frequently enough" means within
# visibility_timeout.)
count = 0
while True:
reassert = (count >= visibility_timeout_reassert)
for i, task in enumerate((local.task_active, local.task_push)):
if task:
name = task_names[i]
if task.proc is not None:
# test if process has finished
task.retcode = task.proc.poll()
if task.retcode is not None:
# process has finished
task.proc = None
# did process finish with errors?
if task.retcode != 0:
errtxt = "fatal error in %s task" % (name,)
if name == 'active':
raise error.ValueErrorRetry(errtxt)
else:
raise ValueError(errtxt)
# Process finished successfully. If S3-push process,
# tell SQS that the task completed successfully.
if name == 'push':
print "******* TASK", task.id, "COMMITTED to S3"
q.delete_message(task.msg)
task.msg = None
local.task_count += 1
task_complete_accounting(local.task_count)
# active task completed?
if name == 'active':
print "******* TASK", task.id, "READY-FOR-PUSH"
# tell SQS that we are still working on the task
if reassert and task.proc is not None:
print "******* REASSERT", name, task.id
task.msg.change_visibility(visibility_timeout)
# break out of loop only when no pending tasks remain
if ((not local.task_active or local.task_active.proc is None)
and (not local.task_push or local.task_push.proc is None)):
break
# setup for next process poll iteration
if reassert:
count = 0
time.sleep(1)
count += 1
# clean up the S3-push task
cleanup(local.task_push, 'push')
local.task_push = None
# start a concurrent push task to commit files generated by
# just-completed active task (such as blender render frames) to S3
if local.task_active:
local.task_active.proc = start_s3_push_process(opts, args, conf, local.task_active.outdir)
local.task_push = local.task_active
local.task_active = None
# if no active task and no S3-push task, we are done (unless DONE is set to "poll")
if not local.task_active and not local.task_push:
if read_done_file() == "poll":
print "Polling for more work..."
time.sleep(15)
else:
break
finally:
cleanup_all()
