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
  14 os.umask(0077)
  15
  16 t = arvados.current_task().tmpdir
  17
  18 api = arvados.api('v1')
  19
  20 os.chdir(arvados.current_task().tmpdir)
  21 os.mkdir("tmpdir")
  22 os.mkdir("output")
  23
  24 os.chdir("output")
  25
  26 if len(arvados.current_task()['parameters']) > 0:
  27     p = arvados.current_task()['parameters']
  28 else:
  29     p = arvados.current_job()['script_parameters']
  30
  31 links = []
  32
  33 def sub_link(v):
  34     r = os.path.basename(v)
  35     os.symlink(os.path.join(os.environ['TASK_KEEPMOUNT'], v) , r)
  36     links.append(r)
  37     return r
  38
  39 def sub_tmpdir(v):
  40     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
  41
  42 def sub_cores(v):
  43      return os.environ['CRUNCH_NODE_SLOTS']
  44
  45 subst.default_subs["link "] = sub_link
  46 subst.default_subs["tmpdir"] = sub_tmpdir
  47 subst.default_subs["node.cores"] = sub_cores
  48
  49 rcode = 1
  50
  51 def machine_progress(bytes_written, bytes_expected):
  52     return "run-command: {} written {} total\n".format(
  53         bytes_written, -1 if (bytes_expected is None) else bytes_expected)
  54
  55 class SigHandler(object):
  56     def __init__(self):
  57         self.sig = None
  58
  59     def send_signal(self, sp, signum):
  60         sp.send_signal(signum)
  61         self.sig = signum
  62         print("We get signal! %s" % signum)
  63
  64 try:
  65     cmd = []
  66     for c in p["command"]:
  67         cmd.append(subst.do_substitution(p, c))
  68
  69     stdoutname = None
  70     stdoutfile = None
  71     if "stdout" in p:
  72         stdoutname = subst.do_substitution(p, p["stdout"])
  73         stdoutfile = open(stdoutname, "wb")
  74
  75     print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
  76
  77     sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
  78     sig = SigHandler()
  79
  80     # forward signals to the process.
  81     signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(sp, signum))
  82     signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(sp, signum))
  83     signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(sp, signum))
  84
  85     # wait for process to complete.
  86     rcode = sp.wait()
  87
  88     print("Sig is %s" % sig)
  89     print("Sig.sig is %s" % sig.sig)
  90
  91     if sig.sig != None:
  92         print("run-command: terminating on signal %s" % sig.sig)
  93         sys.exit(rcode)
  94     else:
  95         print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
  96
  97 except Exception as e:
  98     print("run-command: caught exception: {}".format(e))
  99
 100 finally:
 101     # restore default signal handlers.
 102     signal.signal(signal.SIGINT, signal.SIG_DFL)
 103     signal.signal(signal.SIGTERM, signal.SIG_DFL)
 104     signal.signal(signal.SIGQUIT, signal.SIG_DFL)
 105
 106     for l in links:
 107         os.unlink(l)
 108
 109     print("run-command: the follow output files will be saved to keep:")
 110
 111     subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
 112
 113     print("run-command: start writing output to keep")
 114
 115     done = False
 116     resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint"))
 117     reporter = put.progress_writer(machine_progress)
 118     bytes_expected = put.expected_bytes_for(".")
 119     while not done:
 120         try:
 121             out = put.ArvPutCollectionWriter(resume_cache, reporter, bytes_expected)
 122             out.do_queued_work()
 123             out.write_directory_tree(".", max_manifest_depth=0)
 124             outuuid = out.finish()
 125             api.job_tasks().update(uuid=arvados.current_task()['uuid'],
 126                                                  body={
 127                                                      'output':outuuid,
 128                                                      'success': (rcode == 0),
 129                                                      'progress':1.0
 130                                                  }).execute()
 131             done = True
 132         except KeyboardInterrupt:
 133             print("run-command: terminating on signal SIGINT")
 134             done = True
 135         except Exception as e:
 136             print("run-command: caught exception: {}".format(e))
 137             time.sleep(5)
 138
 139 sys.exit(rcode)