def pipeline(task, wkname, conf=None):
logindent(2)
if not task: # in this case, conf is expected
source_seqtype = "aa" if "aa" in GLOBALS["seqtypes"] else "nt"
all_seqs = GLOBALS["target_sequences"]
initial_task = Msf(set(all_seqs), set(),
seqtype=source_seqtype)
initial_task.main_tree = None
initial_task.threadid = generate_runid()
initial_task.configid = initial_task.threadid
initial_task.target_wkname = wkname
# Register node
db.add_node(initial_task.threadid, initial_task.nodeid,
initial_task.cladeid, initial_task.target_seqs,
initial_task.out_seqs)
new_tasks = [initial_task]
else:
conf = GLOBALS[task.configid]
npr_conf = IterConfig(conf, wkname, task.size, task.seqtype)
new_tasks = process_task(task, wkname, npr_conf, conf["_nodeinfo"])
process_new_tasks(task, new_tasks, conf)
logindent(-2)
return new_tasks
def pipeline(task, wkname, conf=None):
logindent(2)
# Points to npr parameters according to task properties
if not task:
source_seqtype = "aa" if "aa" in GLOBALS["seqtypes"] else "nt"
npr_conf = IterConfig(conf, wkname,
len(GLOBALS["target_species"]),
source_seqtype)
cogconf, cogclass = npr_conf.cog_selector
initial_task = cogclass(GLOBALS["target_species"], set(),
source_seqtype, conf, cogconf)
initial_task.main_tree = main_tree = None
initial_task.threadid = generate_runid()
initial_task.configid = initial_task.threadid
initial_task.target_wkname = wkname
# Register node
db.add_node(initial_task.threadid, initial_task.nodeid,
initial_task.cladeid, initial_task.targets,
initial_task.outgroups)
new_tasks = [initial_task]
else:
conf = GLOBALS[task.configid]
npr_conf = IterConfig(conf, wkname, task.size, task.seqtype)
new_tasks = process_task(task, wkname, npr_conf, conf['_nodeinfo'])
process_new_tasks(task, new_tasks, conf)
logindent(-2)
return new_tasks
def show_task_info(task):
log.log(26, "")
set_logindent(1)
log.log(28, "(%s) %s" % (color_status(task.status), task))
logindent(2)
st_info = ', '.join(["%d(%s)" % (v, k) for k, v in
task.job_status.iteritems()])
log.log(26, "%d jobs: %s" %(len(task.jobs), st_info))
tdir = task.taskid
tdir = tdir.lstrip("/")
log.log(20, "TaskDir: %s" %tdir)
if task.status == "L":
logindent(-2)
log.warning("Some jobs within the task [%s] are marked as (L)ost,"
" meaning that although they look as running,"
" its execution could not be tracked. NPR will"
" continue execution with other pending tasks."
%task)
logindent(2)
logindent(2)
# Shows details about jobs
for j in task.jobs:
if j.status == "D":
log.log(20, "(%s): %s", j.status, j)
else:
log.log(24, "(%s): %s", j.status, j)
logindent(-2)
def get_jobs_status(self, sge_jobs=None):
""" Check the status of all children jobs. """
self.cores_used = 0
all_states = defaultdict(int)
jobs_to_check = set(reversed(self.jobs))
while jobs_to_check:
j = jobs_to_check.pop()
logindent(1)
jobid = j.taskid if istask(j) else j.jobid
if jobid in GLOBALS["cached_status"]:
log.log(22, "@@8:Recycling status@@1: %s" % j)
st = GLOBALS["cached_status"][jobid]
all_states[st] += 1
elif j not in self._donejobs:
st = j.get_status(sge_jobs)
GLOBALS["cached_status"][jobid] = st
all_states[st] += 1
if st == "D":
self._donejobs.add(j)
# If task has an internal worflow processor,
# launch it and populate with new jobs
if istask(j) and j.task_processor:
pipeline = j.task_processor
target_workflow = j.target_wkname
for new_job in pipeline(j, target_workflow):
jobs_to_check.add(new_job)
self.jobs.append(new_job)
elif st in set("QRL"):
if isjob(j) and not j.host.startswith("@sge"):
self.cores_used += j.cores
elif istask(j):
self.cores_used += j.cores_used
elif st == "E":
errorpath = j.jobdir if isjob(j) else j.taskid
raise TaskError(j, "Job execution error %s" % errorpath)
else:
all_states["D"] += 1
logindent(-1)
if not all_states:
all_states["D"] += 1
return all_states
def get_status(self, sge_jobs=None):
# If another tasks with the same id (same work to be done) has
# been checked in the same cycle, reuse its information
if self.taskid in GLOBALS["cached_status"]:
return GLOBALS["cached_status"][self.taskid]
# Otherwise check the status or all its children jobs and
# tasks
logindent(2)
# last_status = db.get_last_task_status(self.taskid)
task_saved = db.task_is_saved(self.taskid)
# If task is processed and saved, just return its state
# without checking children
if task_saved and self.status == "D":
log.log(24, "@@8:Task is done and processed@@1:")
self.status = "D"
# If I have just noticed the task is done and saved, load its
# stored data.
elif task_saved and self.status != "D":
log.log(26, "@@8:Loading pre-computed data@@1:")
self.status = "D"
self.load_stored_data()
else:
# Otherwise, we need to check for all children
self.job_status = self.get_jobs_status(sge_jobs)
job_statuses = set(self.job_status.keys())
# If all children jobs have just finished, we process the
# task, and save it into the database
if job_statuses == set("D"):
logindent(-2)
log.log(22, "Processing done task: %s", self)
logindent(2)
try:
self.finish()
except Exception, e:
print traceback.print_exc()
raise TaskError(self, e)
else:
# store in database .......
if self.check():
self.status = "D"
elif self.status == "!":
# this means the finish procedure has generate
# new jobs associated to the task, so it
# requires relaunching
self.status = "W"
else:
# Otherwise, everything point to errors when
# processing
raise TaskError(self, "Task check not passed")
# Otherwise, update the ongoing task status, but do not
# store result yet.
else:
else:
# Order matters
if "E" in job_statuses:
self.status = "E"
elif "L" in job_statuses:
self.status = "L"
elif "R" in job_statuses:
self.status = "R"
elif "Q" in job_statuses:
self.status = "Q"
elif "W" in job_statuses:
self.status = "W"
else:
log.error("unknown task state %s" % (job_statuses))
logindent(-2)
GLOBALS["cached_status"][self.taskid] = self.status
return self.status
def init(self):
# List of associated jobs necessary to complete the task. Job
# and Task classes are accepted as elements in the list.
self.jobs = []
self._donejobs = set()
self._running_jobs = set()
# Prepare required jobs
self.load_jobs()
def schedule(workflow_task_processor, pending_tasks, schedule_time, execution, debug, norender):
# Adjust debug mode
if debug == "all":
log.setLevel(10)
pending_tasks = set(pending_tasks)
## ===================================
## INITIALIZE BASIC VARS
execution, run_detached = execution
thread2tasks = defaultdict(list)
for task in pending_tasks:
thread2tasks[task.configid].append(task)
expected_threads = set(thread2tasks.keys())
past_threads = {}
thread_errors = defaultdict(list)
## END OF VARS AND SHORTCUTS
## ===================================
cores_total = GLOBALS["_max_cores"]
if cores_total > 0:
job_queue = Queue()
back_launcher = Process(target=background_job_launcher,
args=(job_queue, run_detached,
GLOBALS["launch_time"], cores_total))
back_launcher.start()
else:
job_queue = None
back_launcher = None
GLOBALS["_background_scheduler"] = back_launcher
GLOBALS["_job_queue"] = job_queue
# Captures Ctrl-C for debuging DEBUG
#signal.signal(signal.SIGINT, control_c)
last_report_time = None
BUG = set()
try:
# Enters into task scheduling
while pending_tasks:
wtime = schedule_time
# ask SGE for running jobs
if execution == "sge":
sgeid2jobs = db.get_sge_tasks()
qstat_jobs = sge.qstat()
else:
qstat_jobs = None
# Show summary of pending tasks per thread
thread2tasks = defaultdict(list)
for task in pending_tasks:
thread2tasks[task.configid].append(task)
set_logindent(0)
log.log(28, "@@13: Updating tasks status:@@1: (%s)" % (ctime()))
info_lines = []
for tid, tlist in thread2tasks.iteritems():
threadname = GLOBALS[tid]["_name"]
sizelist = ["%s" %getattr(_ts, "size", "?") for _ts in tlist]
info = "Thread @@13:%s@@1:: pending tasks: @@8:%s@@1: of sizes: %s" %(
threadname, len(tlist), ', '.join(sizelist))
info_lines.append(info)
for line in info_lines:
log.log(28, line)
if GLOBALS["email"] and last_report_time is None:
last_report_time = time()
send_mail(GLOBALS["email"], "Your NPR process has started", '\n'.join(info_lines))
## ================================
## CHECK AND UPDATE CURRENT TASKS
checked_tasks = set()
check_start_time = time()
to_add_tasks = set()
GLOBALS["cached_status"] = {}
for task in sorted(pending_tasks, sort_tasks):
# Avoids endless periods without new job submissions
elapsed_time = time() - check_start_time
#if not back_launcher and pending_tasks and \
# elapsed_time > schedule_time * 2:
# log.log(26, "@@8:Interrupting task checks to schedule new jobs@@1:")
# db.commit()
# wtime = launch_jobs(sorted(pending_tasks, sort_tasks),
# execution, run_detached)
# check_start_time = time()
# Enter debuging mode if necessary
if debug and log.level > 10 and task.taskid.startswith(debug):
log.setLevel(10)
log.debug("ENTERING IN DEBUGGING MODE")
thread2tasks[task.configid].append(task)
# Update tasks and job statuses
if task.taskid not in checked_tasks:
try:
show_task_info(task)
task.status = task.get_status(qstat_jobs)
db.dataconn.commit()
if back_launcher and task.status not in set("DE"):
for j, cmd in task.iter_waiting_jobs():
j.status = "Q"
GLOBALS["cached_status"][j.jobid] = "Q"
if j.jobid not in BUG:
if not os.path.exists(j.jobdir):
os.makedirs(j.jobdir)
for ifile, outpath in j.input_files.iteritems():
try:
_tid, _did = ifile.split(".")
_did = int(_did)
except (IndexError, ValueError):
dataid = ifile
else:
dataid = db.get_dataid(_tid, _did)
if not outpath:
outfile = pjoin(GLOBALS["input_dir"], ifile)
else:
outfile = pjoin(outpath, ifile)
if not os.path.exists(outfile):
open(outfile, "w").write(db.get_data(dataid))
log.log(24, " @@8:Queueing @@1: %s from %s" %(j, task))
job_queue.put([j.jobid, j.cores, cmd, j.status_file])
BUG.add(j.jobid)
update_task_states_recursively(task)
db.commit()
checked_tasks.add(task.taskid)
except TaskError, e:
log.error("Errors found in %s" %task)
import traceback
traceback.print_exc()
if GLOBALS["email"]:
threadname = GLOBALS[task.configid]["_name"]
send_mail(GLOBALS["email"], "Errors found in %s!" %threadname,
'\n'.join(map(str, [task, e.value, e.msg])))
pending_tasks.discard(task)
thread_errors[task.configid].append([task, e.value, e.msg])
continue
else:
# Set temporary Queued state to avoids launching
# jobs from clones
task.status = "Q"
if log.level < 24:
show_task_info(task)
if task.status == "D":
#db.commit()
show_task_info(task)
logindent(3)
# Log commands of every task
if 'cmd_log_file' not in GLOBALS[task.configid]:
GLOBALS[task.configid]['cmd_log_file'] = pjoin(GLOBALS[task.configid]["_outpath"], "cmd.log")
O = open(GLOBALS[task.configid]['cmd_log_file'], "w")
O.close()
cmd_lines = get_cmd_log(task)
CMD_LOG = open(GLOBALS[task.configid]['cmd_log_file'], "a")
print >>CMD_LOG, task
for c in cmd_lines:
print >>CMD_LOG, ' '+'\t'.join(map(str, c))
CMD_LOG.close()
#
try:
#wkname = GLOBALS[task.configid]['_name']
create_tasks = workflow_task_processor(task, task.target_wkname)
except TaskError, e:
log.error("Errors found in %s" %task)
pending_tasks.discard(task)
thread_errors[task.configid].append([task, e.value, e.msg])
continue
else:
logindent(-3)
to_add_tasks.update(create_tasks)
pending_tasks.discard(task)
elif task.status == "E":
log.error("task contains errors: %s " %task)
log.error("Errors found in %s")
pending_tasks.discard(task)
thread_errors[task.configid].append([task, None, "Found (E) task status"])
