X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/c79e86aff4cb20413cf0f09c52fe5066ca197deb..1479cedd5fb5504058e9cb1f4664474a5335d64b:/crunch_scripts/run-command diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command index fbfd511df0..3fd08bf28b 100755 --- a/crunch_scripts/run-command +++ b/crunch_scripts/run-command @@ -1,4 +1,7 @@ #!/usr/bin/env python +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: Apache-2.0 import logging @@ -57,12 +60,10 @@ else: 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' + os.environ['CRUNCH_SRC'] = '/tmp/crunch-src' if 'TASK_KEEPMOUNT' not in os.environ: os.environ['TASK_KEEPMOUNT'] = '/keep' -links = [] - def sub_tmpdir(v): return os.path.join(arvados.current_task().tmpdir, 'tmpdir') @@ -97,36 +98,95 @@ class SigHandler(object): 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)) -def expand_item(p, c, flatten=True): +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_item(params, c["command"])) + r.append(expand_item(params, c["command"])) return r - if "list" in c and "index" in c and "command" in c: - var = c["list"] - items = get_items(p, p[var]) + 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_list(params, c["command"]) - if "regex" in c: + return expand_item(params, c["command"]) + elif "regex" in c: pattern = re.compile(c["regex"]) if "filter" in c: - items = get_items(p, p[c["filter"]]) + # 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: - items = get_items(p, p[c["group"]]) + # 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) @@ -134,60 +194,66 @@ def expand_item(p, c, flatten=True): add_to_group(groups, match) return [groups[k] for k in groups] elif "extract" in c: - items = get_items(p, p[c["extract"]]) + # 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) + return [expand_item(p, arg) for arg in c] elif isinstance(c, basestring): - if flatten: - return [subst.do_substitution(p, c)] + 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) - - return [] - -def expand_list(p, l, flatten=True): - if isinstance(l, basestring): - return expand_item(p, l) - elif flatten: - return [exp for arg in l for exp in expand_item(p, arg, flatten)] else: - return [expand_item(p, arg, flatten) for arg in l] - -def get_items(p, value, flatten=True): - if isinstance(value, dict): - return expand_item(p, value) - + 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, flatten) - - 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 = [os.path.join(fn, l) for l in os.listdir(fn)] - elif stat.S_ISREG(mode): - with open(fn) as f: - items = [line.rstrip("\r\n") for line in f] - return items - else: - return None + 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], False) + 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: @@ -205,9 +271,10 @@ def recursive_foreach(params, fvars): }).execute() else: if isinstance(params["command"][0], list): - logger.info(expand_list(params, params["command"], False)) + for c in params["command"]: + logger.info(flatten(expand_item(params, c))) else: - logger.info(expand_list(params, params["command"], True)) + logger.info(flatten(expand_item(params, params["command"]))) else: logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var])) sys.exit(1) @@ -252,9 +319,10 @@ try: cmd = [] if isinstance(taskp["command"][0], list): - cmd.append(expand_list(taskp, taskp["command"], False)) + for c in taskp["command"]: + cmd.append(flatten(expand_item(taskp, c))) else: - cmd.append(expand_list(taskp, taskp["command"], True)) + cmd.append(flatten(expand_item(taskp, taskp["command"]))) if "task.stdin" in taskp: stdinname = subst.do_substitution(taskp, taskp["task.stdin"]) @@ -266,6 +334,13 @@ try: if not args.dry_run: stdoutfile = open(stdoutname, "wb") + if "task.env" in taskp: + env = copy.copy(os.environ) + for k,v in taskp["task.env"].items(): + env[k] = subst.do_substitution(taskp, v) + else: + env = None + 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: @@ -281,6 +356,8 @@ except Exception as e: logger.error(pprint.pformat(taskp)) sys.exit(1) +# rcode holds the return codes produced by each subprocess +rcode = {} try: subprocesses = [] close_streams = [] @@ -296,7 +373,7 @@ try: # this is an intermediate command in the pipeline, so its stdout should go to a pipe next_stdout = subprocess.PIPE - sp = subprocess.Popen(cmd[i], shell=False, stdin=next_stdin, stdout=next_stdout) + sp = subprocess.Popen(cmd[i], shell=False, stdin=next_stdin, stdout=next_stdout, env=env) # Need to close the FDs on our side so that subcommands will get SIGPIPE if the # consuming process ends prematurely. @@ -322,11 +399,20 @@ try: active = 1 pids = set([s.pid for s in subprocesses]) - rcode = {} while len(pids) > 0: - (pid, status) = os.wait() - pids.discard(pid) - rcode[pid] = (status >> 8) + try: + (pid, status) = os.wait() + except OSError as e: + if e.errno == errno.EINTR: + pass + else: + raise + else: + pids.discard(pid) + if not taskp.get("task.ignore_rcode"): + rcode[pid] = (status >> 8) + else: + rcode[pid] = 0 if sig.sig is not None: logger.critical("terminating on signal %s" % sig.sig) @@ -344,32 +430,29 @@ 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) +subprocess.call(["find", "-L", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"], stdout=sys.stderr, cwd=outdir) 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) +if "task.vwd" in taskp and "task.foreach" in jobp: + for root, dirs, files in os.walk(outdir): + for f in files: + s = os.lstat(os.path.join(root, f)) + if stat.S_ISLNK(s.st_mode): + os.unlink(os.path.join(root, f)) + +(outcollection, checkin_error) = vwd.checkin(outdir) -success = reduce(lambda x, y: x & (y == 0), [True]+rcode.values()) +# Success if we ran any subprocess, and they all exited 0. +success = rcode and all(status == 0 for status in rcode.itervalues()) and not checkin_error api.job_tasks().update(uuid=arvados.current_task()['uuid'], body={ - 'output': outcollection, + 'output': outcollection.manifest_text(), 'success': success, 'progress':1.0 }).execute() -sys.exit(rcode) +sys.exit(0 if success else 1)