5 logger = logging.getLogger('run-command')
6 log_handler = logging.StreamHandler()
7 log_handler.setFormatter(logging.Formatter("run-command: %(message)s"))
8 logger.addHandler(log_handler)
9 logger.setLevel(logging.INFO)
17 import crunchutil.subst as subst
19 import arvados.commands.put as put
25 import multiprocessing
26 import crunchutil.robust_put as robust_put
27 import crunchutil.vwd as vwd
33 parser = argparse.ArgumentParser()
34 parser.add_argument('--dry-run', action='store_true')
35 parser.add_argument('--script-parameters', type=str, default="{}")
36 args = parser.parse_args()
41 api = arvados.api('v1')
42 t = arvados.current_task().tmpdir
43 os.chdir(arvados.current_task().tmpdir)
52 jobp = arvados.current_job()['script_parameters']
53 if len(arvados.current_task()['parameters']) > 0:
54 taskp = arvados.current_task()['parameters']
57 jobp = json.loads(args.script_parameters)
58 os.environ['JOB_UUID'] = 'zzzzz-8i9sb-1234567890abcde'
59 os.environ['TASK_UUID'] = 'zzzzz-ot0gb-1234567890abcde'
60 os.environ['CRUNCH_SRC'] = '/tmp/crunch-src'
61 if 'TASK_KEEPMOUNT' not in os.environ:
62 os.environ['TASK_KEEPMOUNT'] = '/keep'
67 return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
73 return str(multiprocessing.cpu_count())
76 return os.environ['JOB_UUID']
79 return os.environ['TASK_UUID']
82 return os.environ['CRUNCH_SRC']
84 subst.default_subs["task.tmpdir"] = sub_tmpdir
85 subst.default_subs["task.outdir"] = sub_outdir
86 subst.default_subs["job.srcdir"] = sub_jobsrc
87 subst.default_subs["node.cores"] = sub_cores
88 subst.default_subs["job.uuid"] = sub_jobid
89 subst.default_subs["task.uuid"] = sub_taskid
91 class SigHandler(object):
95 def send_signal(self, subprocesses, signum):
96 for sp in subprocesses:
97 sp.send_signal(signum)
100 # http://rightfootin.blogspot.com/2006/09/more-on-python-flatten.html
101 def flatten(l, ltypes=(list, tuple)):
106 while isinstance(l[i], ltypes):
116 def add_to_group(gr, match):
120 gr[m].append(match.group(0))
122 class EvaluationError(Exception):
125 # Return the name of variable ('var') that will take on each value in 'items'
126 # when performing an inner substitution
127 def var_items(p, c, key):
129 raise EvaluationError("'%s' was expected in 'p' but is missing" % key)
132 if not isinstance(c["var"], basestring):
133 raise EvaluationError("Value of 'var' must be a string")
134 # Var specifies the variable name for inner parameter substitution
135 return (c["var"], get_items(p, c[key]))
137 # The component function ('key') value is a list, so return the list
138 # directly with no parameter selected.
139 if isinstance(c[key], list):
140 return (None, get_items(p, c[key]))
141 elif isinstance(c[key], basestring):
142 # check if c[key] is a string that looks like a parameter
143 m = re.match("^\$\((.*)\)$", c[key])
144 if m and m.group(1) in p:
145 return (m.group(1), get_items(p, c[key]))
147 # backwards compatible, foreach specifies bare parameter name to use
148 return (c[key], get_items(p, p[c[key]]))
150 raise EvaluationError("Value of '%s' must be a string or list" % key)
152 # "p" is the parameter scope, "c" is the item to be expanded.
153 # If "c" is a dict, apply function expansion.
154 # If "c" is a list, recursively expand each item and return a new list.
155 # If "c" is a string, apply parameter substitution
156 def expand_item(p, c):
157 if isinstance(c, dict):
158 if "foreach" in c and "command" in c:
159 # Expand a command template for each item in the specified user
161 var, items = var_items(p, c, "foreach")
163 raise EvaluationError("Must specify 'var' in foreach")
166 params = copy.copy(p)
168 r.append(expand_item(params, c["command"]))
170 elif "list" in c and "index" in c and "command" in c:
171 # extract a single item from a list
172 var, items = var_items(p, c, "list")
174 raise EvaluationError("Must specify 'var' in list")
175 params = copy.copy(p)
176 params[var] = items[int(c["index"])]
177 return expand_item(params, c["command"])
179 pattern = re.compile(c["regex"])
181 # filter list so that it only includes items that match a
183 _, items = var_items(p, c, "filter")
184 return [i for i in items if pattern.match(i)]
186 # generate a list of lists, where items are grouped on common
187 # subexpression match
188 _, items = var_items(p, c, "group")
191 match = pattern.match(i)
193 add_to_group(groups, match)
194 return [groups[k] for k in groups]
196 # generate a list of lists, where items are split by
197 # subexpression match
198 _, items = var_items(p, c, "extract")
201 match = pattern.match(i)
203 r.append(list(match.groups()))
205 elif "batch" in c and "size" in c:
206 # generate a list of lists, where items are split into a batch size
207 _, items = var_items(p, c, "batch")
210 for j in xrange(0, len(items), sz):
211 r.append(items[j:j+sz])
213 raise EvaluationError("Missing valid list context function")
214 elif isinstance(c, list):
215 return [expand_item(p, arg) for arg in c]
216 elif isinstance(c, basestring):
217 m = re.match("^\$\((.*)\)$", c)
218 if m and m.group(1) in p:
219 return expand_item(p, p[m.group(1)])
221 return subst.do_substitution(p, c)
223 raise EvaluationError("expand_item() unexpected parameter type %s" % type(c))
225 # Evaluate in a list context
226 # "p" is the parameter scope, "value" will be evaluated
227 # if "value" is a list after expansion, return that
228 # if "value" is a path to a directory, return a list consisting of each entry in the directory
229 # if "value" is a path to a file, return a list consisting of each line of the file
230 def get_items(p, value):
231 value = expand_item(p, value)
232 if isinstance(value, list):
234 elif isinstance(value, basestring):
235 mode = os.stat(value).st_mode
236 prefix = value[len(os.environ['TASK_KEEPMOUNT'])+1:]
238 if stat.S_ISDIR(mode):
239 items = [os.path.join(value, l) for l in os.listdir(value)]
240 elif stat.S_ISREG(mode):
241 with open(value) as f:
242 items = [line.rstrip("\r\n") for line in f]
244 raise EvaluationError("get_items did not yield a list")
251 # Construct the cross product of all values of each variable listed in fvars
252 def recursive_foreach(params, fvars):
255 items = get_items(params, params[var])
256 logger.info("parallelizing on %s with items %s" % (var, items))
257 if items is not None:
259 params = copy.copy(params)
262 recursive_foreach(params, fvars)
265 arvados.api().job_tasks().create(body={
266 'job_uuid': arvados.current_job()['uuid'],
267 'created_by_job_task_uuid': arvados.current_task()['uuid'],
272 if isinstance(params["command"][0], list):
273 for c in params["command"]:
274 logger.info(flatten(expand_item(params, c)))
276 logger.info(flatten(expand_item(params, params["command"])))
278 logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var]))
282 if "task.foreach" in jobp:
283 if args.dry_run or arvados.current_task()['sequence'] == 0:
284 # This is the first task to start the other tasks and exit
285 fvars = jobp["task.foreach"]
286 if isinstance(fvars, basestring):
288 if not isinstance(fvars, list) or len(fvars) == 0:
289 logger.error("value of task.foreach must be a string or non-empty list")
291 recursive_foreach(jobp, jobp["task.foreach"])
293 if "task.vwd" in jobp:
294 # Set output of the first task to the base vwd collection so it
295 # will be merged with output fragments from the other tasks by
297 arvados.current_task().set_output(subst.do_substitution(jobp, jobp["task.vwd"]))
299 arvados.current_task().set_output(None)
302 # This is the only task so taskp/jobp are the same
304 except Exception as e:
305 logger.exception("caught exception")
306 logger.error("job parameters were:")
307 logger.error(pprint.pformat(jobp))
312 if "task.vwd" in taskp:
313 # Populate output directory with symlinks to files in collection
314 vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir)
316 if "task.cwd" in taskp:
317 os.chdir(subst.do_substitution(taskp, taskp["task.cwd"]))
320 if isinstance(taskp["command"][0], list):
321 for c in taskp["command"]:
322 cmd.append(flatten(expand_item(taskp, c)))
324 cmd.append(flatten(expand_item(taskp, taskp["command"])))
326 if "task.stdin" in taskp:
327 stdinname = subst.do_substitution(taskp, taskp["task.stdin"])
329 stdinfile = open(stdinname, "rb")
331 if "task.stdout" in taskp:
332 stdoutname = subst.do_substitution(taskp, taskp["task.stdout"])
334 stdoutfile = open(stdoutname, "wb")
336 logger.info("{}{}{}".format(' | '.join([' '.join(c) for c in cmd]), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else ""))
340 except subst.SubstitutionError as e:
342 logger.error("task parameters were:")
343 logger.error(pprint.pformat(taskp))
345 except Exception as e:
346 logger.exception("caught exception")
347 logger.error("task parameters were:")
348 logger.error(pprint.pformat(taskp))
351 # rcode holds the return codes produced by each subprocess
357 close_streams.append(stdinfile)
358 next_stdin = stdinfile
360 for i in xrange(len(cmd)):
362 # this is the last command in the pipeline, so its stdout should go to stdoutfile
363 next_stdout = stdoutfile
365 # this is an intermediate command in the pipeline, so its stdout should go to a pipe
366 next_stdout = subprocess.PIPE
368 sp = subprocess.Popen(cmd[i], shell=False, stdin=next_stdin, stdout=next_stdout)
370 # Need to close the FDs on our side so that subcommands will get SIGPIPE if the
371 # consuming process ends prematurely.
373 close_streams.append(sp.stdout)
375 # Send this processes's stdout to to the next process's stdin
376 next_stdin = sp.stdout
378 subprocesses.append(sp)
380 # File descriptors have been handed off to the subprocesses, so close them here.
381 for s in close_streams:
384 # Set up signal handling
387 # Forward terminate signals to the subprocesses.
388 signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(subprocesses, signum))
389 signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(subprocesses, signum))
390 signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(subprocesses, signum))
393 pids = set([s.pid for s in subprocesses])
395 (pid, status) = os.wait()
397 if not taskp.get("task.ignore_rcode"):
398 rcode[pid] = (status >> 8)
402 if sig.sig is not None:
403 logger.critical("terminating on signal %s" % sig.sig)
406 for i in xrange(len(cmd)):
407 r = rcode[subprocesses[i].pid]
408 logger.info("%s completed with exit code %i (%s)" % (cmd[i][0], r, "success" if r == 0 else "failed"))
410 except Exception as e:
411 logger.exception("caught exception")
413 # restore default signal handlers.
414 signal.signal(signal.SIGINT, signal.SIG_DFL)
415 signal.signal(signal.SIGTERM, signal.SIG_DFL)
416 signal.signal(signal.SIGQUIT, signal.SIG_DFL)
421 logger.info("the following output files will be saved to keep:")
423 subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"], stdout=sys.stderr)
425 logger.info("start writing output to keep")
427 if "task.vwd" in taskp:
428 if "task.foreach" in jobp:
429 # This is a subtask, so don't merge with the original collection, that will happen at the end
430 outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=False).manifest_text()
432 # Just a single task, so do merge with the original collection
433 outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=True).manifest_text()
435 outcollection = robust_put.upload(outdir, logger)
437 # Success if we ran any subprocess, and they all exited 0.
438 success = rcode and all(status == 0 for status in rcode.itervalues())
440 api.job_tasks().update(uuid=arvados.current_task()['uuid'],
442 'output': outcollection,
447 sys.exit(0 if success else 1)