X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/847459b3c257aba65df3e0cbf6777f7148542af2..55ed403e8bd408c70ce97cefc10156141ccd46a6:/crunch_scripts/run-command diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command index 2eb5589471..1635ecdf69 100755 --- a/crunch_scripts/run-command +++ b/crunch_scripts/run-command @@ -1,13 +1,30 @@ #!/usr/bin/env python +import logging + +logger = logging.getLogger('run-command') +log_handler = logging.StreamHandler() +log_handler.setFormatter(logging.Formatter("run-command: %(message)s")) +logger.addHandler(log_handler) +logger.setLevel(logging.INFO) + import arvados import re import os import subprocess import sys import shutil -import subst +import crunchutil.subst as subst import time +import arvados.commands.put as put +import signal +import stat +import copy +import traceback +import pprint +import multiprocessing +import crunchutil.robust_put as robust_put +import crunchutil.vwd as vwd os.umask(0077) @@ -21,84 +38,200 @@ os.mkdir("output") os.chdir("output") +outdir = os.getcwd() + +taskp = None +jobp = arvados.current_job()['script_parameters'] if len(arvados.current_task()['parameters']) > 0: - p = arvados.current_task()['parameters'] -else: - p = arvados.current_job()['script_parameters'] + taskp = arvados.current_task()['parameters'] links = [] -def sub_link(v): - r = os.path.basename(v) - os.symlink(os.path.join(os.environ['TASK_KEEPMOUNT'], v) , r) - links.append(r) - return r - def sub_tmpdir(v): return os.path.join(arvados.current_task().tmpdir, 'tmpdir') +def sub_outdir(v): + return outdir + def sub_cores(v): - return os.environ['CRUNCH_NODE_SLOTS'] + return str(multiprocessing.cpu_count()) -subst.default_subs["link "] = sub_link -subst.default_subs["tmpdir"] = sub_tmpdir -subst.default_subs["node.cores"] = sub_cores +def sub_jobid(v): + return os.environ['JOB_UUID'] + +def sub_taskid(v): + return os.environ['TASK_UUID'] +def sub_jobsrc(v): + return os.environ['CRUNCH_SRC'] + +subst.default_subs["task.tmpdir"] = sub_tmpdir +subst.default_subs["task.outdir"] = sub_outdir +subst.default_subs["job.srcdir"] = sub_jobsrc +subst.default_subs["node.cores"] = sub_cores +subst.default_subs["job.uuid"] = sub_jobid +subst.default_subs["task.uuid"] = sub_taskid + +class SigHandler(object): + def __init__(self): + self.sig = None + + def send_signal(self, sp, signum): + sp.send_signal(signum) + self.sig = signum + +def expand_item(p, c): + if isinstance(c, dict): + if "foreach" in c and "command" in c: + var = c["foreach"] + items = get_items(p, p[var]) + r = [] + for i in items: + params = copy.copy(p) + params[var] = i + r.extend(expand_list(params, c["command"])) + return r + elif isinstance(c, list): + return expand_list(p, c) + elif isinstance(c, str) or isinstance(c, unicode): + return [subst.do_substitution(p, c)] + + return [] + +def expand_list(p, l): + return [exp for arg in l for exp in expand_item(p, arg)] + +def get_items(p, value): + if isinstance(value, list): + return expand_list(p, value) + + fn = subst.do_substitution(p, value) + mode = os.stat(fn).st_mode + prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:] + if mode is not None: + if stat.S_ISDIR(mode): + items = ["$(dir %s/%s/)" % (prefix, l) for l in os.listdir(fn)] + elif stat.S_ISREG(mode): + with open(fn) as f: + items = [line for line in f] + return items + else: + return None + +stdoutname = None +stdoutfile = None +stdinname = None +stdinfile = None rcode = 1 try: - cmd = [] - for c in p["command"]: - cmd.append(subst.do_substitution(p, c)) - - stdoutname = None - stdoutfile = None - if "stdout" in p: - stdoutname = subst.do_substitution(p, p["stdout"]) + if "task.foreach" in jobp: + if arvados.current_task()['sequence'] == 0: + var = jobp["task.foreach"] + items = get_items(jobp, jobp[var]) + logger.info("parallelizing on %s with items %s" % (var, items)) + if items is not None: + for i in items: + params = copy.copy(jobp) + params[var] = i + arvados.api().job_tasks().create(body={ + 'job_uuid': arvados.current_job()['uuid'], + 'created_by_job_task_uuid': arvados.current_task()['uuid'], + 'sequence': 1, + 'parameters': params + } + ).execute() + if "task.vwd" in jobp: + # Base vwd collection will be merged with output fragments from + # the other tasks by crunch. + arvados.current_task().set_output(subst.do_substitution(jobp, jobp["task.vwd"])) + else: + arvados.current_task().set_output(None) + sys.exit(0) + else: + sys.exit(1) + else: + taskp = jobp + + if "task.vwd" in taskp: + # Populate output directory with symlinks to files in collection + vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir) + + if "task.cwd" in taskp: + os.chdir(subst.do_substitution(taskp, taskp["task.cwd"])) + + cmd = expand_list(taskp, taskp["command"]) + + if "task.stdin" in taskp: + stdinname = subst.do_substitution(taskp, taskp["task.stdin"]) + stdinfile = open(stdinname, "rb") + + if "task.stdout" in taskp: + stdoutname = subst.do_substitution(taskp, taskp["task.stdout"]) stdoutfile = open(stdoutname, "wb") - print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else "")) + logger.info("{}{}{}".format(' '.join(cmd), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else "")) +except subst.SubstitutionError as e: + logger.error(str(e)) + logger.error("task parameters was:") + logger.error(pprint.pformat(taskp)) + sys.exit(1) +except Exception as e: + logger.exception("caught exception") + logger.error("task parameters was:") + logger.error(pprint.pformat(taskp)) + sys.exit(1) + +try: + sp = subprocess.Popen(cmd, shell=False, stdin=stdinfile, stdout=stdoutfile) + sig = SigHandler() - rcode = subprocess.call(cmd, stdout=stdoutfile) + # forward signals to the process. + signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(sp, signum)) + signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(sp, signum)) + signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(sp, signum)) - print("run-command: completed with exit code %i" % rcode) + # wait for process to complete. + rcode = sp.wait() + + if sig.sig is not None: + logger.critical("terminating on signal %s" % sig.sig) + sys.exit(2) + else: + logger.info("completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed")) except Exception as e: - print("run-command: caught exception: {}".format(e)) - -finally: - for l in links: - os.unlink(l) - - print("run-command: the follow output files will be saved to keep:") - - subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"]) - - print("run-command: start writing output to keep") - - out = arvados.CollectionWriter() - out.write_directory_tree(".", max_manifest_depth=0) - done = False - while not done: - try: - outuuid = out.finish() - api.job_tasks().update(uuid=arvados.current_task()['uuid'], - body={ - 'output':outuuid, - 'success': (rcode == 0), - 'progress':1.0 - }).execute() - done = True - except Exception as e: - print("run-command: caught exception: {}".format(e)) - time.sleep(5) - -if rcode == 0: - os.chdir("..") - shutil.rmtree("tmpdir") - shutil.rmtree("output") - print("run-command: success") + logger.exception("caught exception") + +# restore default signal handlers. +signal.signal(signal.SIGINT, signal.SIG_DFL) +signal.signal(signal.SIGTERM, signal.SIG_DFL) +signal.signal(signal.SIGQUIT, signal.SIG_DFL) + +for l in links: + os.unlink(l) + +logger.info("the following output files will be saved to keep:") + +subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"], stdout=sys.stderr) + +logger.info("start writing output to keep") + +if "task.vwd" in taskp: + if "task.foreach" in jobp: + # This is a subtask, so don't merge with the original collection, that will happen at the end + outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=False).manifest_text() + else: + # Just a single task, so do merge with the original collection + outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=True).manifest_text() else: - print("run-command: failed") + outcollection = robust_put.upload(outdir) + +api.job_tasks().update(uuid=arvados.current_task()['uuid'], + body={ + 'output': outcollection, + 'success': (rcode == 0), + 'progress':1.0 + }).execute() sys.exit(rcode)