X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/c3b19359bd21c0b2115c4169102327762e48a93b..ac21f0d45a76294aaca0c0c0fdf06eb72d03368d:/crunch_scripts/run-command diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command index 64d9c6c931..13ae918895 100755 --- a/crunch_scripts/run-command +++ b/crunch_scripts/run-command @@ -1,47 +1,76 @@ #!/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 +import argparse +import json +import tempfile +import errno + +parser = argparse.ArgumentParser() +parser.add_argument('--dry-run', action='store_true') +parser.add_argument('--script-parameters', type=str, default="{}") +args = parser.parse_args() os.umask(0077) -t = arvados.current_task().tmpdir - -api = arvados.api('v1') +if not args.dry_run: + api = arvados.api('v1') + t = arvados.current_task().tmpdir + os.chdir(arvados.current_task().tmpdir) + os.mkdir("tmpdir") + os.mkdir("output") -os.chdir(arvados.current_task().tmpdir) -os.mkdir("tmpdir") -os.mkdir("output") + os.chdir("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 = None + jobp = arvados.current_job()['script_parameters'] + if len(arvados.current_task()['parameters']) > 0: + taskp = arvados.current_task()['parameters'] +else: + outdir = "/tmp" + jobp = json.loads(args.script_parameters) + os.environ['JOB_UUID'] = 'zzzzz-8i9sb-1234567890abcde' + os.environ['TASK_UUID'] = 'zzzzz-ot0gb-1234567890abcde' + os.environ['CRUNCH_SRC'] = '/tmp/crunche-src' + if 'TASK_KEEPMOUNT' not in os.environ: + os.environ['TASK_KEEPMOUNT'] = '/keep' 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'] @@ -49,117 +78,337 @@ 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): self.sig = None - def send_signal(self, sp, signum): - sp.send_signal(signum) + def send_signal(self, subprocesses, signum): + for sp in subprocesses: + sp.send_signal(signum) self.sig = signum +# http://rightfootin.blogspot.com/2006/09/more-on-python-flatten.html +def flatten(l, ltypes=(list, tuple)): + ltype = type(l) + l = list(l) + i = 0 + while i < len(l): + while isinstance(l[i], ltypes): + if not l[i]: + l.pop(i) + i -= 1 + break + else: + l[i:i + 1] = l[i] + i += 1 + return ltype(l) + +def add_to_group(gr, match): + m = match.groups() + if m not in gr: + gr[m] = [] + gr[m].append(match.group(0)) + +class EvaluationError(Exception): + pass + +# Return the name of variable ('var') that will take on each value in 'items' +# when performing an inner substitution +def var_items(p, c, key): + if key not in c: + raise EvaluationError("'%s' was expected in 'p' but is missing" % key) + + if "var" in c: + if not isinstance(c["var"], basestring): + raise EvaluationError("Value of 'var' must be a string") + # Var specifies the variable name for inner parameter substitution + return (c["var"], get_items(p, c[key])) + else: + # The component function ('key') value is a list, so return the list + # directly with no parameter selected. + if isinstance(c[key], list): + return (None, get_items(p, c[key])) + elif isinstance(c[key], basestring): + # check if c[key] is a string that looks like a parameter + m = re.match("^\$\((.*)\)$", c[key]) + if m and m.group(1) in p: + return (m.group(1), get_items(p, c[key])) + else: + # backwards compatible, foreach specifies bare parameter name to use + return (c[key], get_items(p, p[c[key]])) + else: + raise EvaluationError("Value of '%s' must be a string or list" % key) + +# "p" is the parameter scope, "c" is the item to be expanded. +# If "c" is a dict, apply function expansion. +# If "c" is a list, recursively expand each item and return a new list. +# If "c" is a string, apply parameter substitution 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]) + # Expand a command template for each item in the specified user + # parameter + var, items = var_items(p, c, "foreach") + if var is None: + raise EvaluationError("Must specify 'var' in foreach") r = [] for i in items: params = copy.copy(p) params[var] = i - r.extend(expand_list(params, c["command"])) + r.append(expand_item(params, c["command"])) + return r + elif "list" in c and "index" in c and "command" in c: + # extract a single item from a list + var, items = var_items(p, c, "list") + if var is None: + raise EvaluationError("Must specify 'var' in list") + params = copy.copy(p) + params[var] = items[int(c["index"])] + return expand_item(params, c["command"]) + elif "regex" in c: + pattern = re.compile(c["regex"]) + if "filter" in c: + # filter list so that it only includes items that match a + # regular expression + _, items = var_items(p, c, "filter") + return [i for i in items if pattern.match(i)] + elif "group" in c: + # generate a list of lists, where items are grouped on common + # subexpression match + _, items = var_items(p, c, "group") + groups = {} + for i in items: + match = pattern.match(i) + if match: + add_to_group(groups, match) + return [groups[k] for k in groups] + elif "extract" in c: + # generate a list of lists, where items are split by + # subexpression match + _, items = var_items(p, c, "extract") + r = [] + for i in items: + match = pattern.match(i) + if match: + r.append(list(match.groups())) + return r + elif "batch" in c and "size" in c: + # generate a list of lists, where items are split into a batch size + _, items = var_items(p, c, "batch") + sz = int(c["size"]) + r = [] + for j in xrange(0, len(items), sz): + r.append(items[j:j+sz]) return r + raise EvaluationError("Missing valid list context function") elif isinstance(c, list): - return expand_list(p, c) - elif isinstance(c, str): - 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)] + return [expand_item(p, arg) for arg in c] + elif isinstance(c, basestring): + m = re.match("^\$\((.*)\)$", c) + if m and m.group(1) in p: + return expand_item(p, p[m.group(1)]) + else: + return subst.do_substitution(p, c) + else: + raise EvaluationError("expand_item() unexpected parameter type %s" % type(c)) +# Evaluate in a list context +# "p" is the parameter scope, "value" will be evaluated +# if "value" is a list after expansion, return that +# if "value" is a path to a directory, return a list consisting of each entry in the directory +# if "value" is a path to a file, return a list consisting of each line of the file def get_items(p, value): + value = expand_item(p, value) if isinstance(value, list): - return expand_list(p, value) - - fn = subst.do_substitution(p, value) - mode = os.stat(fn).st_mode - prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:] - if mode != None: - if stat.S_ISDIR(mode): - items = ["$(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 + return value + elif isinstance(value, basestring): + mode = os.stat(value).st_mode + prefix = value[len(os.environ['TASK_KEEPMOUNT'])+1:] + if mode is not None: + if stat.S_ISDIR(mode): + items = [os.path.join(value, l) for l in os.listdir(value)] + elif stat.S_ISREG(mode): + with open(value) as f: + items = [line.rstrip("\r\n") for line in f] + return items + raise EvaluationError("get_items did not yield a list") + +stdoutname = None +stdoutfile = None +stdinname = None +stdinfile = None + +# Construct the cross product of all values of each variable listed in fvars +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: + if not args.dry_run: + 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() + else: + if isinstance(params["command"][0], list): + for c in params["command"]: + logger.info(flatten(expand_item(params, c))) + else: + logger.info(flatten(expand_item(params, params["command"]))) else: - return None - -if "task.foreach" in jobp: - if arvados.current_task()['sequence'] == 0: - var = jobp["task.foreach"] - items = get_items(jobp, jobp[var]) - if items != None: - print("run-command: parallelizing on %s with items %s" % (var, items)) + logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var])) + sys.exit(1) - 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() +try: + if "task.foreach" in jobp: + if args.dry_run or 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 not args.dry_run: + 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: - sys.exit(1) -else: - p = jobp + else: + # This is the only task so taskp/jobp are the same + taskp = jobp +except Exception as e: + logger.exception("caught exception") + logger.error("job parameters were:") + logger.error(pprint.pformat(jobp)) + sys.exit(1) try: - cmd = expand_list(p, p["command"]) + if not args.dry_run: + 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 = [] + if isinstance(taskp["command"][0], list): + for c in taskp["command"]: + cmd.append(flatten(expand_item(taskp, c))) + else: + cmd.append(flatten(expand_item(taskp, taskp["command"]))) + + if "task.stdin" in taskp: + stdinname = subst.do_substitution(taskp, taskp["task.stdin"]) + if not args.dry_run: + stdinfile = open(stdinname, "rb") + + if "task.stdout" in taskp: + stdoutname = subst.do_substitution(taskp, taskp["task.stdout"]) + if not args.dry_run: + stdoutfile = open(stdoutname, "wb") + + logger.info("{}{}{}".format(' | '.join([' '.join(c) for c in cmd]), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else "")) + + if args.dry_run: + sys.exit(0) +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) - stdoutname = None - stdoutfile = None - if "save.stdout" in p: - stdoutname = subst.do_substitution(p, p["save.stdout"]) - stdoutfile = open(stdoutname, "wb") +# rcode holds the return codes produced by each subprocess +rcode = {} +try: + subprocesses = [] + close_streams = [] + if stdinfile: + close_streams.append(stdinfile) + next_stdin = stdinfile + + for i in xrange(len(cmd)): + if i == len(cmd)-1: + # this is the last command in the pipeline, so its stdout should go to stdoutfile + next_stdout = stdoutfile + else: + # this is an intermediate command in the pipeline, so its stdout should go to a pipe + next_stdout = subprocess.PIPE - print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else "")) + sp = subprocess.Popen(cmd[i], shell=False, stdin=next_stdin, stdout=next_stdout) - sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile) - sig = SigHandler() + # Need to close the FDs on our side so that subcommands will get SIGPIPE if the + # consuming process ends prematurely. + if sp.stdout: + close_streams.append(sp.stdout) + + # Send this processes's stdout to to the next process's stdin + next_stdin = sp.stdout + + subprocesses.append(sp) - # 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)) + # File descriptors have been handed off to the subprocesses, so close them here. + for s in close_streams: + s.close() - # wait for process to complete. - rcode = sp.wait() + # Set up signal handling + sig = SigHandler() + + # Forward terminate signals to the subprocesses. + signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(subprocesses, signum)) + signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(subprocesses, signum)) + signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(subprocesses, signum)) + + active = 1 + pids = set([s.pid for s in subprocesses]) + while len(pids) > 0: + (pid, status) = os.wait() + pids.discard(pid) + if not taskp.get("task.ignore_rcode"): + rcode[pid] = (status >> 8) + else: + rcode[pid] = 0 - 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")) + for i in xrange(len(cmd)): + r = rcode[subprocesses[i].pid] + logger.info("%s completed with exit code %i (%s)" % (cmd[i][0], r, "success" if r == 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) @@ -169,34 +418,30 @@ 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) + +# Success if no non-zero return codes +success = any(rcode) and not any([status != 0 for status in rcode.values()]) + +api.job_tasks().update(uuid=arvados.current_task()['uuid'], + body={ + 'output': outcollection, + 'success': success, + 'progress':1.0 + }).execute() sys.exit(rcode)