crunch_scripts/run-command

   1 #!/usr/bin/env python
   2
   3 import arvados
   4 import re
   5 import os
   6 import subprocess
   7 import sys
   8 import shutil
   9 import subst
  10 import time
  11 import arvados.commands.put as put
  12 import signal
  13 import stat
  14 import copy
  15 import traceback
  16
  17 os.umask(0077)
  18
  19 t = arvados.current_task().tmpdir
  20
  21 api = arvados.api('v1')
  22
  23 os.chdir(arvados.current_task().tmpdir)
  24 os.mkdir("tmpdir")
  25 os.mkdir("output")
  26
  27 os.chdir("output")
  28
  29 taskp = None
  30 jobp = arvados.current_job()['script_parameters']
  31 if len(arvados.current_task()['parameters']) > 0:
  32     p = arvados.current_task()['parameters']
  33
  34 links = []
  35
  36 def sub_link(v):
  37     r = os.path.basename(v)
  38     os.symlink(os.path.join(os.environ['TASK_KEEPMOUNT'], v) , r)
  39     links.append(r)
  40     return r
  41
  42 def sub_tmpdir(v):
  43     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
  44
  45 def sub_cores(v):
  46      return os.environ['CRUNCH_NODE_SLOTS']
  47
  48 def sub_jobid(v):
  49      return os.environ['JOB_UUID']
  50
  51 def sub_taskid(v):
  52      return os.environ['TASK_UUID']
  53
  54 subst.default_subs["link "] = sub_link
  55 subst.default_subs["task.tmpdir"] = sub_tmpdir
  56 subst.default_subs["node.cores"] = sub_cores
  57 subst.default_subs["job.id"] = sub_jobid
  58 subst.default_subs["task.id"] = sub_taskid
  59
  60 rcode = 1
  61
  62 def machine_progress(bytes_written, bytes_expected):
  63     return "run-command: wrote {} total {}\n".format(
  64         bytes_written, -1 if (bytes_expected is None) else bytes_expected)
  65
  66 class SigHandler(object):
  67     def __init__(self):
  68         self.sig = None
  69
  70     def send_signal(self, sp, signum):
  71         sp.send_signal(signum)
  72         self.sig = signum
  73
  74 def expand_item(p, c):
  75     if isinstance(c, dict):
  76         if "foreach" in c and "command" in c:
  77             var = c["foreach"]
  78             items = get_items(p, p[var])
  79             r = []
  80             for i in items:
  81                 params = copy.copy(p)
  82                 params[var] = i
  83                 r.extend(expand_list(params, c["command"]))
  84             return r
  85     elif isinstance(c, list):
  86         return expand_list(p, c)
  87     elif isinstance(c, str):
  88         return [subst.do_substitution(p, c)]
  89
  90     return []
  91
  92 def expand_list(p, l):
  93     return [exp for arg in l for exp in expand_item(p, arg)]
  94
  95 def get_items(p, value):
  96     if isinstance(value, list):
  97         return expand_list(p, value)
  98
  99     fn = subst.do_substitution(p, value)
 100     mode = os.stat(fn).st_mode
 101     prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:]
 102     if mode != None:
 103         if stat.S_ISDIR(mode):
 104             items = ["$(dir %s/%s)" % (prefix, l) for l in os.listdir(fn)]
 105         elif stat.S_ISREG(mode):
 106             with open(fn) as f:
 107                 items = [line for line in f]
 108         return items
 109     else:
 110         return None
 111
 112 if "task.foreach" in jobp:
 113     if arvados.current_task()['sequence'] == 0:
 114         var = jobp["task.foreach"]
 115         items = get_items(jobp, jobp[var])
 116         if items != None:
 117             print("run-command: parallelizing on %s with items %s" % (var, items))
 118
 119             for i in items:
 120                 params = copy.copy(jobp)
 121                 params[var] = i
 122                 arvados.api().job_tasks().create(body={
 123                     'job_uuid': arvados.current_job()['uuid'],
 124                     'created_by_job_task_uuid': arvados.current_task()['uuid'],
 125                     'sequence': 1,
 126                     'parameters': params
 127                     }
 128                 ).execute()
 129             arvados.current_task().set_output(None)
 130             sys.exit(0)
 131         else:
 132             sys.exit(1)
 133 else:
 134     p = jobp
 135
 136 try:
 137     cmd = expand_list(p, p["command"])
 138
 139     stdoutname = None
 140     stdoutfile = None
 141     if "save.stdout" in p:
 142         stdoutname = subst.do_substitution(p, p["save.stdout"])
 143         stdoutfile = open(stdoutname, "wb")
 144
 145     print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
 146
 147     sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
 148     sig = SigHandler()
 149
 150     # forward signals to the process.
 151     signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(sp, signum))
 152     signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(sp, signum))
 153     signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(sp, signum))
 154
 155     # wait for process to complete.
 156     rcode = sp.wait()
 157
 158     if sig.sig != None:
 159         print("run-command: terminating on signal %s" % sig.sig)
 160         sys.exit(2)
 161     else:
 162         print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
 163
 164 except Exception as e:
 165     print("run-command: caught exception:")
 166     traceback.print_exc(file=sys.stdout)
 167
 168 # restore default signal handlers.
 169 signal.signal(signal.SIGINT, signal.SIG_DFL)
 170 signal.signal(signal.SIGTERM, signal.SIG_DFL)
 171 signal.signal(signal.SIGQUIT, signal.SIG_DFL)
 172
 173 for l in links:
 174     os.unlink(l)
 175
 176 print("run-command: the following output files will be saved to keep:")
 177
 178 subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
 179
 180 print("run-command: start writing output to keep")
 181
 182 done = False
 183 resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint"))
 184 reporter = put.progress_writer(machine_progress)
 185 bytes_expected = put.expected_bytes_for(".")
 186 while not done:
 187     try:
 188         out = put.ArvPutCollectionWriter.from_cache(resume_cache, reporter, bytes_expected)
 189         out.do_queued_work()
 190         out.write_directory_tree(".", max_manifest_depth=0)
 191         outuuid = out.finish()
 192         api.job_tasks().update(uuid=arvados.current_task()['uuid'],
 193                                              body={
 194                                                  'output':outuuid,
 195                                                  'success': (rcode == 0),
 196                                                  'progress':1.0
 197                                              }).execute()
 198         done = True
 199     except KeyboardInterrupt:
 200         print("run-command: terminating on signal 2")
 201         sys.exit(2)
 202     except Exception as e:
 203         print("run-command: caught exception:")
 204         traceback.print_exc(file=sys.stdout)
 205         time.sleep(5)
 206
 207 sys.exit(rcode)