Merge branch 'bcbio-nextgen' refs #2991

[arvados.git] / crunch_scripts / run-command

diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command
index 1351f3bfadbac51d94879046e5b0017774f72230..2eb558947109fc7afe25bea6e3efba8ac6631395 100755 (executable)
--- a/crunch_scripts/run-command
+++ b/crunch_scripts/run-command
@@ -7,11 +7,14 @@ import subprocess
 import sys
 import shutil
 import subst
+import time
 
 os.umask(0077)
 
 t = arvados.current_task().tmpdir
 
+api = arvados.api('v1')
+
 os.chdir(arvados.current_task().tmpdir)
 os.mkdir("tmpdir")
 os.mkdir("output")
@@ -34,8 +37,12 @@ def sub_link(v):
 def sub_tmpdir(v):
     return os.path.join(arvados.current_task().tmpdir, 'tmpdir')
 
+def sub_cores(v):
+     return os.environ['CRUNCH_NODE_SLOTS']
+
 subst.default_subs["link "] = sub_link
 subst.default_subs["tmpdir"] = sub_tmpdir
+subst.default_subs["node.cores"] = sub_cores
 
 rcode = 1
 
@@ -50,41 +57,48 @@ try:
         stdoutname = subst.do_substitution(p, p["stdout"])
         stdoutfile = open(stdoutname, "wb")
 
-    print("run-command {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
+    print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
 
     rcode = subprocess.call(cmd, stdout=stdoutfile)
 
-    print("run-command completed with exit code %i" % rcode)
+    print("run-command: completed with exit code %i" % rcode)
 
 except Exception as e:
-    print("run-command exception: {}".format(e))
+    print("run-command: caught exception: {}".format(e))
 
 finally:
     for l in links:
         os.unlink(l)
 
-    print("run-command output:")
+    print("run-command: the follow output files will be saved to keep:")
 
-    subprocess.call(["find", "."])
+    subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
 
-    print("run-command writing to keep")
+    print("run-command: start writing output to keep")
 
     out = arvados.CollectionWriter()
     out.write_directory_tree(".", max_manifest_depth=0)
-    outuuid = out.finish()
-    arvados.api('v1').job_tasks().update(uuid=arvados.current_task()['uuid'],
-                                         body={
-                                             'output':outuuid,
-                                             'success': (rcode == 0),
-                                             'progress':1.0
-                                         }).execute()
+    done = False
+    while not done:
+        try:
+            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 Exception as e:
+            print("run-command: caught exception: {}".format(e))
+            time.sleep(5)
 
 if rcode == 0:
     os.chdir("..")
     shutil.rmtree("tmpdir")
     shutil.rmtree("output")
-    print("run-command success")
+    print("run-command: success")
 else:
-    print("run-command failed")
+    print("run-command: failed")
 
 sys.exit(rcode)