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 import pprint
  17
  18 os.umask(0077)
  19
  20 t = arvados.current_task().tmpdir
  21
  22 api = arvados.api('v1')
  23
  24 os.chdir(arvados.current_task().tmpdir)
  25 os.mkdir("tmpdir")
  26 os.mkdir("output")
  27
  28 os.chdir("output")
  29
  30 taskp = None
  31 jobp = arvados.current_job()['script_parameters']
  32 if len(arvados.current_task()['parameters']) > 0:
  33     p = arvados.current_task()['parameters']
  34
  35 links = []
  36
  37 def sub_link(v):
  38     r = os.path.basename(v)
  39     os.symlink(os.path.join(os.environ['TASK_KEEPMOUNT'], v) , r)
  40     links.append(r)
  41     return r
  42
  43 def sub_tmpdir(v):
  44     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
  45
  46 def sub_cores(v):
  47      return os.environ['CRUNCH_NODE_SLOTS']
  48
  49 def sub_jobid(v):
  50      return os.environ['JOB_UUID']
  51
  52 def sub_taskid(v):
  53      return os.environ['TASK_UUID']
  54
  55 subst.default_subs["link "] = sub_link
  56 subst.default_subs["task.tmpdir"] = sub_tmpdir
  57 subst.default_subs["node.cores"] = sub_cores
  58 subst.default_subs["job.id"] = sub_jobid
  59 subst.default_subs["task.id"] = sub_taskid
  60
  61 rcode = 1
  62
  63 def machine_progress(bytes_written, bytes_expected):
  64     return "run-command: wrote {} total {}\n".format(
  65         bytes_written, -1 if (bytes_expected is None) else bytes_expected)
  66
  67 class SigHandler(object):
  68     def __init__(self):
  69         self.sig = None
  70
  71     def send_signal(self, sp, signum):
  72         sp.send_signal(signum)
  73         self.sig = signum
  74
  75 def expand_item(p, c):
  76     if isinstance(c, dict):
  77         if "foreach" in c and "command" in c:
  78             var = c["foreach"]
  79             items = get_items(p, p[var])
  80             r = []
  81             for i in items:
  82                 params = copy.copy(p)
  83                 params[var] = i
  84                 r.extend(expand_list(params, c["command"]))
  85             return r
  86     elif isinstance(c, list):
  87         return expand_list(p, c)
  88     elif isinstance(c, str):
  89         return [subst.do_substitution(p, c)]
  90
  91     return []
  92
  93 def expand_list(p, l):
  94     return [exp for arg in l for exp in expand_item(p, arg)]
  95
  96 def get_items(p, value):
  97     if isinstance(value, list):
  98         return expand_list(p, value)
  99
 100     fn = subst.do_substitution(p, value)
 101     mode = os.stat(fn).st_mode
 102     prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:]
 103     if mode != None:
 104         if stat.S_ISDIR(mode):
 105             items = ["$(dir %s/%s/)" % (prefix, l) for l in os.listdir(fn)]
 106         elif stat.S_ISREG(mode):
 107             with open(fn) as f:
 108                 items = [line for line in f]
 109         return items
 110     else:
 111         return None
 112
 113 stdoutname = None
 114 stdoutfile = None
 115
 116 try:
 117     if "task.foreach" in jobp:
 118         if arvados.current_task()['sequence'] == 0:
 119             var = jobp["task.foreach"]
 120             items = get_items(jobp, jobp[var])
 121             print("run-command: parallelizing on %s with items %s" % (var, items))
 122             if items != None:
 123                 for i in items:
 124                     params = copy.copy(jobp)
 125                     params[var] = i
 126                     arvados.api().job_tasks().create(body={
 127                         'job_uuid': arvados.current_job()['uuid'],
 128                         'created_by_job_task_uuid': arvados.current_task()['uuid'],
 129                         'sequence': 1,
 130                         'parameters': params
 131                         }
 132                     ).execute()
 133                 arvados.current_task().set_output(None)
 134                 sys.exit(0)
 135             else:
 136                 sys.exit(1)
 137     else:
 138         p = jobp
 139
 140     cmd = expand_list(p, p["command"])
 141
 142     if "save.stdout" in p:
 143         stdoutname = subst.do_substitution(p, p["save.stdout"])
 144         stdoutfile = open(stdoutname, "wb")
 145
 146     print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
 147
 148 except Exception as e:
 149     print("run-command: caught exception:")
 150     traceback.print_exc(file=sys.stdout)
 151     print("run-command: task parameters was:")
 152     pprint.pprint(p)
 153     sys.exit(1)
 154
 155 try:
 156     sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
 157     sig = SigHandler()
 158
 159     # forward signals to the process.
 160     signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(sp, signum))
 161     signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(sp, signum))
 162     signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(sp, signum))
 163
 164     # wait for process to complete.
 165     rcode = sp.wait()
 166
 167     if sig.sig != None:
 168         print("run-command: terminating on signal %s" % sig.sig)
 169         sys.exit(2)
 170     else:
 171         print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
 172
 173 except Exception as e:
 174     print("run-command: caught exception:")
 175     traceback.print_exc(file=sys.stdout)
 176
 177 # restore default signal handlers.
 178 signal.signal(signal.SIGINT, signal.SIG_DFL)
 179 signal.signal(signal.SIGTERM, signal.SIG_DFL)
 180 signal.signal(signal.SIGQUIT, signal.SIG_DFL)
 181
 182 for l in links:
 183     os.unlink(l)
 184
 185 print("run-command: the following output files will be saved to keep:")
 186
 187 subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
 188
 189 print("run-command: start writing output to keep")
 190
 191 done = False
 192 resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint"))
 193 reporter = put.progress_writer(machine_progress)
 194 bytes_expected = put.expected_bytes_for(".")
 195 while not done:
 196     try:
 197         out = put.ArvPutCollectionWriter.from_cache(resume_cache, reporter, bytes_expected)
 198         out.do_queued_work()
 199         out.write_directory_tree(".", max_manifest_depth=0)
 200         outuuid = out.finish()
 201         api.job_tasks().update(uuid=arvados.current_task()['uuid'],
 202                                              body={
 203                                                  'output':outuuid,
 204                                                  'success': (rcode == 0),
 205                                                  'progress':1.0
 206                                              }).execute()
 207         done = True
 208     except KeyboardInterrupt:
 209         print("run-command: terminating on signal 2")
 210         sys.exit(2)
 211     except Exception as e:
 212         print("run-command: caught exception:")
 213         traceback.print_exc(file=sys.stdout)
 214         time.sleep(5)
 215
 216 sys.exit(rcode)