X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/99d5b548b50c573281d9780ecd631ecf83f4eefe..b4dde1b5648aa71180282b7dfa0fc70ba7c87755:/crunch_scripts/run-command diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command index 27ce8227ee..74ad8a87fc 100755 --- a/crunch_scripts/run-command +++ b/crunch_scripts/run-command @@ -1,15 +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) @@ -23,24 +38,23 @@ 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'] + 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()) def sub_jobid(v): return os.environ['JOB_UUID'] @@ -48,17 +62,15 @@ def sub_jobid(v): def sub_taskid(v): return os.environ['TASK_UUID'] -subst.default_subs["link "] = sub_link +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.id"] = sub_jobid -subst.default_subs["task.id"] = sub_taskid - -rcode = 1 - -def machine_progress(bytes_written, bytes_expected): - return "run-command: wrote {} total {}\n".format( - bytes_written, -1 if (bytes_expected is None) else bytes_expected) +subst.default_subs["job.uuid"] = sub_jobid +subst.default_subs["task.uuid"] = sub_taskid class SigHandler(object): def __init__(self): @@ -81,7 +93,7 @@ def expand_item(p, c): return r elif isinstance(c, list): return expand_list(p, c) - elif isinstance(c, str): + elif isinstance(c, str) or isinstance(c, unicode): return [subst.do_substitution(p, c)] return [] @@ -94,28 +106,36 @@ def get_items(p, value): return expand_list(p, value) fn = subst.do_substitution(p, value) - mode = os.stat(fn) + mode = os.stat(fn).st_mode prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:] - if mode != None: + if mode is not None: if stat.S_ISDIR(mode): - items = ["$(dir %s/%s)" % (prefix, l) for l in os.listdir(fn)] - else stat.S_ISREG(mode): + items = ["$(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] + items = [line.rstrip("\r\n") for line in f] return items else: return None -if "task.foreach" in jobp: - if arvados.current_task()['sequence'] == 0: - var = jobp["foreach"] - items = get_items(jobp, jobp[var]) - if items != None: - print("run-command: parallelizing on %s with items %s" % (var, items)) +stdoutname = None +stdoutfile = None +stdinname = None +stdinfile = None +rcode = 1 - for i in items: - params = copy.copy(jobp) - params[var] = i +def recursive_foreach(params, fvars): + var = fvars[0] + fvars = fvars[1:] + items = get_items(params, params[var]) + logger.info("parallelizing on %s with items %s" % (var, items)) + if items is not None: + for i in items: + params = copy.copy(params) + params[var] = i + if len(fvars) > 0: + recursive_foreach(params, fvars) + else: arvados.api().job_tasks().create(body={ 'job_uuid': arvados.current_job()['uuid'], 'created_by_job_task_uuid': arvados.current_task()['uuid'], @@ -123,24 +143,64 @@ if "task.foreach" in jobp: 'parameters': params } ).execute() - sys.exit(0) - else: - sys.exit(1) -else: - p = jobp + else: + logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var])) + sys.exit(1) try: - cmd = expand_list(p, p["command"]) + if "task.foreach" in jobp: + if arvados.current_task()['sequence'] == 0: + # This is the first task to start the other tasks and exit + fvars = jobp["task.foreach"] + if isinstance(fvars, basestring): + fvars = [fvars] + if not isinstance(fvars, list) or len(fvars) == 0: + logger.error("value of task.foreach must be a string or non-empty list") + sys.exit(1) + recursive_foreach(jobp, jobp["task.foreach"]) + if "task.vwd" in jobp: + # Set output of the first task to the base vwd collection so it + # 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: + # This is the only task so taskp/jobp are the same + 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"])) - stdoutname = None - stdoutfile = None - if "stdout" in p: - stdoutname = subst.do_substitution(p, p["stdout"]) + 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 were:") + logger.error(pprint.pformat(taskp)) + sys.exit(1) +except Exception as e: + logger.exception("caught exception") + logger.error("task parameters were:") + logger.error(pprint.pformat(taskp)) + sys.exit(1) - sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile) +try: + sp = subprocess.Popen(cmd, shell=False, stdin=stdinfile, stdout=stdoutfile) sig = SigHandler() # forward signals to the process. @@ -151,14 +211,14 @@ try: # wait for process to complete. rcode = sp.wait() - if sig.sig != None: - print("run-command: terminating on signal %s" % sig.sig) + if sig.sig is not None: + logger.critical("terminating on signal %s" % sig.sig) sys.exit(2) else: - print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed")) + 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)) + logger.exception("caught exception") # restore default signal handlers. signal.signal(signal.SIGINT, signal.SIG_DFL) @@ -168,34 +228,27 @@ signal.signal(signal.SIGQUIT, signal.SIG_DFL) for l in links: os.unlink(l) -print("run-command: the following 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") - -done = False -resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint")) -reporter = put.progress_writer(machine_progress) -bytes_expected = put.expected_bytes_for(".") -while not done: - try: - out = put.ArvPutCollectionWriter.from_cache(resume_cache, reporter, bytes_expected) - out.do_queued_work() - out.write_directory_tree(".", max_manifest_depth=0) - 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 KeyboardInterrupt: - print("run-command: terminating on signal 2") - sys.exit(2) - except Exception as e: - print("run-command: caught exception: {}".format(e)) - time.sleep(5) +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: + outcollection = robust_put.upload(outdir, logger) + +api.job_tasks().update(uuid=arvados.current_task()['uuid'], + body={ + 'output': outcollection, + 'success': (rcode == 0), + 'progress':1.0 + }).execute() sys.exit(rcode)