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