X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/2ec515c8ef7f9cae426a6490d1317333718e1d5e..ff723f5f08e286df4c4c58a27a9db574ca604a6f:/crunch_scripts/run-command diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command index e21089ee80..74793d4fce 100755 --- a/crunch_scripts/run-command +++ b/crunch_scripts/run-command @@ -57,12 +57,10 @@ 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') @@ -119,42 +117,73 @@ def add_to_group(gr, match): 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: - # Var specifies + 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])) - m = re.match("^\$\((.*)\)$", c[key]) - if m and m.group(1) in p: - return (m.group(1), 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: - # backwards compatible, foreach specifies bare parameter name to use - return (c[key], get_items(p, p[c[key]])) + 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: + # 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.append(expand_item(params, c["command"])) return r - if "list" in c and "index" in c and "command" in c: + 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"]) - if "regex" in c: + elif "regex" in c: pattern = re.compile(c["regex"]) if "filter" in c: - var, items = var_items(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: - var, items = var_items(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) @@ -162,20 +191,24 @@ def expand_item(p, c): add_to_group(groups, match) return [groups[k] for k in groups] elif "extract" in c: - var, items = var_items(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 - if "batch" in c and "size" in c: - var, items = var_items(p, c, "batch") + 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_item(p, arg) for arg in c] elif isinstance(c, basestring): @@ -184,9 +217,14 @@ def expand_item(p, c): 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)) - return [] - +# 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): @@ -201,13 +239,14 @@ def get_items(p, value): with open(value) as f: items = [line.rstrip("\r\n") for line in f] return items - raise Exception("get_items did not yield a list") + 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:] @@ -292,6 +331,13 @@ 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: @@ -307,6 +353,8 @@ 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 = [] @@ -322,7 +370,7 @@ 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. @@ -348,11 +396,20 @@ 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) @@ -370,32 +427,29 @@ 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)