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'
65 return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
71 return str(multiprocessing.cpu_count())
74 return os.environ['JOB_UUID']
77 return os.environ['TASK_UUID']
80 return os.environ['CRUNCH_SRC']
82 subst.default_subs["task.tmpdir"] = sub_tmpdir
83 subst.default_subs["task.outdir"] = sub_outdir
84 subst.default_subs["job.srcdir"] = sub_jobsrc
85 subst.default_subs["node.cores"] = sub_cores
86 subst.default_subs["job.uuid"] = sub_jobid
87 subst.default_subs["task.uuid"] = sub_taskid
89 class SigHandler(object):
93 def send_signal(self, subprocesses, signum):
94 for sp in subprocesses:
95 sp.send_signal(signum)
98 # http://rightfootin.blogspot.com/2006/09/more-on-python-flatten.html
99 def flatten(l, ltypes=(list, tuple)):
104 while isinstance(l[i], ltypes):
114 def add_to_group(gr, match):
118 gr[m].append(match.group(0))
120 class EvaluationError(Exception):
123 # Return the name of variable ('var') that will take on each value in 'items'
124 # when performing an inner substitution
125 def var_items(p, c, key):
127 raise EvaluationError("'%s' was expected in 'p' but is missing" % key)
130 if not isinstance(c["var"], basestring):
131 raise EvaluationError("Value of 'var' must be a string")
132 # Var specifies the variable name for inner parameter substitution
133 return (c["var"], get_items(p, c[key]))
135 # The component function ('key') value is a list, so return the list
136 # directly with no parameter selected.
137 if isinstance(c[key], list):
138 return (None, get_items(p, c[key]))
139 elif isinstance(c[key], basestring):
140 # check if c[key] is a string that looks like a parameter
141 m = re.match("^\$\((.*)\)$", c[key])
142 if m and m.group(1) in p:
143 return (m.group(1), get_items(p, c[key]))
145 # backwards compatible, foreach specifies bare parameter name to use
146 return (c[key], get_items(p, p[c[key]]))
148 raise EvaluationError("Value of '%s' must be a string or list" % key)
150 # "p" is the parameter scope, "c" is the item to be expanded.
151 # If "c" is a dict, apply function expansion.
152 # If "c" is a list, recursively expand each item and return a new list.
153 # If "c" is a string, apply parameter substitution
154 def expand_item(p, c):
155 if isinstance(c, dict):
156 if "foreach" in c and "command" in c:
157 # Expand a command template for each item in the specified user
159 var, items = var_items(p, c, "foreach")
161 raise EvaluationError("Must specify 'var' in foreach")
164 params = copy.copy(p)
166 r.append(expand_item(params, c["command"]))
168 elif "list" in c and "index" in c and "command" in c:
169 # extract a single item from a list
170 var, items = var_items(p, c, "list")
172 raise EvaluationError("Must specify 'var' in list")
173 params = copy.copy(p)
174 params[var] = items[int(c["index"])]
175 return expand_item(params, c["command"])
177 pattern = re.compile(c["regex"])
179 # filter list so that it only includes items that match a
181 _, items = var_items(p, c, "filter")
182 return [i for i in items if pattern.match(i)]
184 # generate a list of lists, where items are grouped on common
185 # subexpression match
186 _, items = var_items(p, c, "group")
189 match = pattern.match(i)
191 add_to_group(groups, match)
192 return [groups[k] for k in groups]
194 # generate a list of lists, where items are split by
195 # subexpression match
196 _, items = var_items(p, c, "extract")
199 match = pattern.match(i)
201 r.append(list(match.groups()))
203 elif "batch" in c and "size" in c:
204 # generate a list of lists, where items are split into a batch size
205 _, items = var_items(p, c, "batch")
208 for j in xrange(0, len(items), sz):
209 r.append(items[j:j+sz])
211 raise EvaluationError("Missing valid list context function")
212 elif isinstance(c, list):
213 return [expand_item(p, arg) for arg in c]
214 elif isinstance(c, basestring):
215 m = re.match("^\$\((.*)\)$", c)
216 if m and m.group(1) in p:
217 return expand_item(p, p[m.group(1)])
219 return subst.do_substitution(p, c)
221 raise EvaluationError("expand_item() unexpected parameter type %s" % type(c))
223 # Evaluate in a list context
224 # "p" is the parameter scope, "value" will be evaluated
225 # if "value" is a list after expansion, return that
226 # if "value" is a path to a directory, return a list consisting of each entry in the directory
227 # if "value" is a path to a file, return a list consisting of each line of the file
228 def get_items(p, value):
229 value = expand_item(p, value)
230 if isinstance(value, list):
232 elif isinstance(value, basestring):
233 mode = os.stat(value).st_mode
234 prefix = value[len(os.environ['TASK_KEEPMOUNT'])+1:]
236 if stat.S_ISDIR(mode):
237 items = [os.path.join(value, l) for l in os.listdir(value)]
238 elif stat.S_ISREG(mode):
239 with open(value) as f:
240 items = [line.rstrip("\r\n") for line in f]
242 raise EvaluationError("get_items did not yield a list")
249 # Construct the cross product of all values of each variable listed in fvars
250 def recursive_foreach(params, fvars):
253 items = get_items(params, params[var])
254 logger.info("parallelizing on %s with items %s" % (var, items))
255 if items is not None:
257 params = copy.copy(params)
260 recursive_foreach(params, fvars)
263 arvados.api().job_tasks().create(body={
264 'job_uuid': arvados.current_job()['uuid'],
265 'created_by_job_task_uuid': arvados.current_task()['uuid'],
270 if isinstance(params["command"][0], list):
271 for c in params["command"]:
272 logger.info(flatten(expand_item(params, c)))
274 logger.info(flatten(expand_item(params, params["command"])))
276 logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var]))
280 if "task.foreach" in jobp:
281 if args.dry_run or arvados.current_task()['sequence'] == 0:
282 # This is the first task to start the other tasks and exit
283 fvars = jobp["task.foreach"]
284 if isinstance(fvars, basestring):
286 if not isinstance(fvars, list) or len(fvars) == 0:
287 logger.error("value of task.foreach must be a string or non-empty list")
289 recursive_foreach(jobp, jobp["task.foreach"])
291 if "task.vwd" in jobp:
292 # Set output of the first task to the base vwd collection so it
293 # will be merged with output fragments from the other tasks by
295 arvados.current_task().set_output(subst.do_substitution(jobp, jobp["task.vwd"]))
297 arvados.current_task().set_output(None)
300 # This is the only task so taskp/jobp are the same
302 except Exception as e:
303 logger.exception("caught exception")
304 logger.error("job parameters were:")
305 logger.error(pprint.pformat(jobp))
310 if "task.vwd" in taskp:
311 # Populate output directory with symlinks to files in collection
312 vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir)
314 if "task.cwd" in taskp:
315 os.chdir(subst.do_substitution(taskp, taskp["task.cwd"]))
318 if isinstance(taskp["command"][0], list):
319 for c in taskp["command"]:
320 cmd.append(flatten(expand_item(taskp, c)))
322 cmd.append(flatten(expand_item(taskp, taskp["command"])))
324 if "task.stdin" in taskp:
325 stdinname = subst.do_substitution(taskp, taskp["task.stdin"])
327 stdinfile = open(stdinname, "rb")
329 if "task.stdout" in taskp:
330 stdoutname = subst.do_substitution(taskp, taskp["task.stdout"])
332 stdoutfile = open(stdoutname, "wb")
334 logger.info("{}{}{}".format(' | '.join([' '.join(c) for c in cmd]), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else ""))
338 except subst.SubstitutionError as e:
340 logger.error("task parameters were:")
341 logger.error(pprint.pformat(taskp))
343 except Exception as e:
344 logger.exception("caught exception")
345 logger.error("task parameters were:")
346 logger.error(pprint.pformat(taskp))
349 # rcode holds the return codes produced by each subprocess
355 close_streams.append(stdinfile)
356 next_stdin = stdinfile
358 for i in xrange(len(cmd)):
360 # this is the last command in the pipeline, so its stdout should go to stdoutfile
361 next_stdout = stdoutfile
363 # this is an intermediate command in the pipeline, so its stdout should go to a pipe
364 next_stdout = subprocess.PIPE
366 sp = subprocess.Popen(cmd[i], shell=False, stdin=next_stdin, stdout=next_stdout)
368 # Need to close the FDs on our side so that subcommands will get SIGPIPE if the
369 # consuming process ends prematurely.
371 close_streams.append(sp.stdout)
373 # Send this processes's stdout to to the next process's stdin
374 next_stdin = sp.stdout
376 subprocesses.append(sp)
378 # File descriptors have been handed off to the subprocesses, so close them here.
379 for s in close_streams:
382 # Set up signal handling
385 # Forward terminate signals to the subprocesses.
386 signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(subprocesses, signum))
387 signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(subprocesses, signum))
388 signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(subprocesses, signum))
391 pids = set([s.pid for s in subprocesses])
393 (pid, status) = os.wait()
395 if not taskp.get("task.ignore_rcode"):
396 rcode[pid] = (status >> 8)
400 if sig.sig is not None:
401 logger.critical("terminating on signal %s" % sig.sig)
404 for i in xrange(len(cmd)):
405 r = rcode[subprocesses[i].pid]
406 logger.info("%s completed with exit code %i (%s)" % (cmd[i][0], r, "success" if r == 0 else "failed"))
408 except Exception as e:
409 logger.exception("caught exception")
411 # restore default signal handlers.
412 signal.signal(signal.SIGINT, signal.SIG_DFL)
413 signal.signal(signal.SIGTERM, signal.SIG_DFL)
414 signal.signal(signal.SIGQUIT, signal.SIG_DFL)
416 logger.info("the following output files will be saved to keep:")
418 subprocess.call(["find", ".", "-printf", "run-command: %12.12s %h/%f\\n"], stdout=sys.stderr, cwd=outdir)
420 logger.info("start writing output to keep")
422 if "task.vwd" in taskp and "task.foreach" in jobp:
423 for root, dirs, files in os.walk(outdir):
425 s = os.lstat(os.path.join(root, f))
426 if stat.S_ISLNK(s.st_mode):
427 os.unlink(os.path.join(root, f))
429 outcollection = vwd.checkin(outdir).manifest_text()
431 # Success if we ran any subprocess, and they all exited 0.
432 success = rcode and all(status == 0 for status in rcode.itervalues())
434 api.job_tasks().update(uuid=arvados.current_task()['uuid'],
436 'output': outcollection,
441 sys.exit(0 if success else 1)