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
  13 os.umask(0077)
  14
  15 t = arvados.current_task().tmpdir
  16
  17 api = arvados.api('v1')
  18
  19 os.chdir(arvados.current_task().tmpdir)
  20 os.mkdir("tmpdir")
  21 os.mkdir("output")
  22
  23 os.chdir("output")
  24
  25 if len(arvados.current_task()['parameters']) > 0:
  26     p = arvados.current_task()['parameters']
  27 else:
  28     p = arvados.current_job()['script_parameters']
  29
  30 links = []
  31
  32 def sub_link(v):
  33     r = os.path.basename(v)
  34     os.symlink(os.path.join(os.environ['TASK_KEEPMOUNT'], v) , r)
  35     links.append(r)
  36     return r
  37
  38 def sub_tmpdir(v):
  39     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
  40
  41 def sub_cores(v):
  42      return os.environ['CRUNCH_NODE_SLOTS']
  43
  44 subst.default_subs["link "] = sub_link
  45 subst.default_subs["tmpdir"] = sub_tmpdir
  46 subst.default_subs["node.cores"] = sub_cores
  47
  48 rcode = 1
  49
  50 def machine_progress(bytes_written, bytes_expected):
  51     return "run-command: {} written {} total\n".format(
  52         bytes_written, -1 if (bytes_expected is None) else bytes_expected)
  53
  54 try:
  55     cmd = []
  56     for c in p["command"]:
  57         cmd.append(subst.do_substitution(p, c))
  58
  59     stdoutname = None
  60     stdoutfile = None
  61     if "stdout" in p:
  62         stdoutname = subst.do_substitution(p, p["stdout"])
  63         stdoutfile = open(stdoutname, "wb")
  64
  65     print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
  66
  67     rcode = subprocess.call(cmd, stdout=stdoutfile)
  68
  69     print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
  70
  71 except Exception as e:
  72     print("run-command: caught exception: {}".format(e))
  73
  74 finally:
  75     for l in links:
  76         os.unlink(l)
  77
  78     print("run-command: the follow output files will be saved to keep:")
  79
  80     subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
  81
  82     print("run-command: start writing output to keep")
  83
  84     done = False
  85     resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint"))
  86     reporter = put.progress_writer(machine_progress)
  87     bytes_expected = put.expected_bytes_for(".")
  88     while not done:
  89         try:
  90             out = put.ArvPutCollectionWriter(resume_cache, reporter, bytes_expected)
  91             out.do_queued_work()
  92             out.write_directory_tree(".", max_manifest_depth=0)
  93             outuuid = out.finish()
  94             api.job_tasks().update(uuid=arvados.current_task()['uuid'],
  95                                                  body={
  96                                                      'output':outuuid,
  97                                                      'success': (rcode == 0),
  98                                                      'progress':1.0
  99                                                  }).execute()
 100             done = True
 101         except Exception as e:
 102             print("run-command: caught exception: {}".format(e))
 103             time.sleep(5)
 104
 105 sys.exit(rcode)