Merge branch '2800-python-global-state' into 2800-pgs
[arvados.git] / crunch_scripts / run-command
index 7d77248674465d253fbd340330c226c3d76480fa..c624e3cadf7ec7fdaad169f04ffb8f34f8bcdd9f 100755 (executable)
@@ -1,12 +1,15 @@
 #!/usr/bin/env python
 
+import logging
+logging.basicConfig(level=logging.INFO, format="run-command: %(message)s")
+
 import arvados
 import re
 import os
 import subprocess
 import sys
 import shutil
-import subst
+import crunchutil.subst as subst
 import time
 import arvados.commands.put as put
 import signal
@@ -15,10 +18,10 @@ import copy
 import traceback
 import pprint
 import multiprocessing
-import logging
+import crunchutil.robust_put as robust_put
+import crunchutil.vwd as vwd
 
 os.umask(0077)
-logging.basicConfig(format="run-command: %(message)s")
 
 t = arvados.current_task().tmpdir
 
@@ -39,12 +42,6 @@ if len(arvados.current_task()['parameters']) > 0:
 
 links = []
 
-def sub_link(v):
-    r = os.path.join(outdir, os.path.basename(v))
-    os.symlink(v, r)
-    links.append(r)
-    return r
-
 def sub_tmpdir(v):
     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
 
@@ -63,7 +60,6 @@ def sub_taskid(v):
 def sub_jobsrc(v):
      return os.environ['CRUNCH_SRC']
 
-subst.default_subs["link "] = sub_link
 subst.default_subs["task.tmpdir"] = sub_tmpdir
 subst.default_subs["task.outdir"] = sub_outdir
 subst.default_subs["job.srcdir"] = sub_jobsrc
@@ -71,10 +67,6 @@ subst.default_subs["node.cores"] = sub_cores
 subst.default_subs["job.uuid"] = sub_jobid
 subst.default_subs["task.uuid"] = sub_taskid
 
-def machine_progress(bytes_written, bytes_expected):
-    return "run-command: wrote {} total {}\n".format(
-        bytes_written, -1 if (bytes_expected is None) else bytes_expected)
-
 class SigHandler(object):
     def __init__(self):
         self.sig = None
@@ -111,7 +103,7 @@ def get_items(p, value):
     fn = subst.do_substitution(p, value)
     mode = os.stat(fn).st_mode
     prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:]
-    if mode != None:
+    if mode is not None:
         if stat.S_ISDIR(mode):
             items = ["$(dir %s/%s/)" % (prefix, l) for l in os.listdir(fn)]
         elif stat.S_ISREG(mode):
@@ -123,6 +115,8 @@ def get_items(p, value):
 
 stdoutname = None
 stdoutfile = None
+stdinname = None
+stdinfile = None
 rcode = 1
 
 try:
@@ -131,7 +125,7 @@ try:
             var = jobp["task.foreach"]
             items = get_items(jobp, jobp[var])
             logging.info("parallelizing on %s with items %s" % (var, items))
-            if items != None:
+            if items is not None:
                 for i in items:
                     params = copy.copy(jobp)
                     params[var] = i
@@ -142,20 +136,36 @@ try:
                         'parameters': params
                         }
                     ).execute()
-                arvados.current_task().set_output(None)
+                if "task.vwd" in jobp:
+                    # Base vwd collection will be merged with output fragments from
+                    # the other tasks by crunch.
+                    arvados.current_task().set_output(subst.do_substitution(jobp, jobp["task.vwd"]))
+                else:
+                    arvados.current_task().set_output(None)
                 sys.exit(0)
             else:
                 sys.exit(1)
     else:
         taskp = jobp
 
+    if "task.vwd" in taskp:
+        # Populate output directory with symlinks to files in collection
+        vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir)
+
+    if "task.cwd" in taskp:
+        os.chdir(subst.do_substitution(taskp, taskp["task.cwd"]))
+
     cmd = expand_list(taskp, taskp["command"])
 
-    if "save.stdout" in taskp:
-        stdoutname = subst.do_substitution(taskp, taskp["save.stdout"])
+    if "task.stdin" in taskp:
+        stdinname = subst.do_substitution(taskp, taskp["task.stdin"])
+        stdinfile = open(stdinname, "rb")
+
+    if "task.stdout" in taskp:
+        stdoutname = subst.do_substitution(taskp, taskp["task.stdout"])
         stdoutfile = open(stdoutname, "wb")
 
-    logging.info("{}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
+    logging.info("{}{}{}".format(' '.join(cmd), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else ""))
 
 except Exception as e:
     logging.exception("caught exception")
@@ -164,7 +174,7 @@ except Exception as e:
     sys.exit(1)
 
 try:
-    sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
+    sp = subprocess.Popen(cmd, shell=False, stdin=stdinfile, stdout=stdoutfile)
     sig = SigHandler()
 
     # forward signals to the process.
@@ -175,7 +185,7 @@ try:
     # wait for process to complete.
     rcode = sp.wait()
 
-    if sig.sig != None:
+    if sig.sig is not None:
         logging.critical("terminating on signal %s" % sig.sig)
         sys.exit(2)
     else:
@@ -198,28 +208,21 @@ subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/
 
 logging.info("start writing output to keep")
 
-done = False
-resume_cache = put.ResumeCache(os.path.join(arvados.current_task().tmpdir, "upload-output-checkpoint"))
-reporter = put.progress_writer(machine_progress)
-bytes_expected = put.expected_bytes_for(".")
-while not done:
-    try:
-        out = put.ArvPutCollectionWriter.from_cache(resume_cache, reporter, bytes_expected)
-        out.do_queued_work()
-        out.write_directory_tree(".", max_manifest_depth=0)
-        outuuid = out.finish()
-        api.job_tasks().update(uuid=arvados.current_task()['uuid'],
-                                             body={
-                                                 'output':outuuid,
-                                                 'success': (rcode == 0),
-                                                 'progress':1.0
-                                             }).execute()
-        done = True
-    except KeyboardInterrupt:
-        logging.critical("terminating on signal 2")
-        sys.exit(2)
-    except Exception as e:
-        logging.exception("caught exception:")
-        time.sleep(5)
+if "task.vwd" in taskp:
+    if "task.foreach" in jobp:
+        # This is a subtask, so don't merge with the original collection, that will happen at the end
+        outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=False).manifest_text()
+    else:
+        # Just a single task, so do merge with the original collection
+        outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=True).manifest_text()
+else:
+    outcollection = robust_put.upload(outdir)
+
+api.job_tasks().update(uuid=arvados.current_task()['uuid'],
+                                     body={
+                                         'output': outcollection,
+                                         'success': (rcode == 0),
+                                         'progress':1.0
+                                     }).execute()
 
 sys.exit(rcode)