X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/bcff555f9786f24ded842f1cf45e162f1a744c32..1cf946bc2ac7370431c193463ab865d71bc32ed3:/crunch_scripts/run-command

diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command
index 9d4f361893..e6ec889bbc 100755
--- a/crunch_scripts/run-command
+++ b/crunch_scripts/run-command
@@ -7,11 +7,16 @@ import subprocess
 import sys
 import shutil
 import subst
+import time
+import arvados.commands.put as put
+import signal
 
 os.umask(0077)
 
 t = arvados.current_task().tmpdir
 
+api = arvados.api('v1')
+
 os.chdir(arvados.current_task().tmpdir)
 os.mkdir("tmpdir")
 os.mkdir("output")
@@ -34,11 +39,27 @@ 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
 
+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
+
+    def send_signal(self, sp, signum):
+        sp.send_signal(signum)
+        self.sig = signum
+
 try:
     cmd = []
     for c in p["command"]:
@@ -50,41 +71,64 @@ try:
         stdoutname = subst.do_substitution(p, p["stdout"])
         stdoutfile = open(stdoutname, "wb")
 
-    print("run-command {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
-
-    rcode = subprocess.call(cmd, stdout=stdoutfile)
+    print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
 
-    print("run-command completed with exit code %i" % rcode)
+    sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
+    sig = SigHandler()
 
-except Exception as e:
-    print("run-command exception: {}".format(e))
-
-finally:
-    for l in links:
-        os.unlink(l)
-
-    print("run-command output:")
+    # forward signals to the process.
+    signal.signal(signal.SIGINT, lambda signum, frame: sig.send_signal(sp, signum))
+    signal.signal(signal.SIGTERM, lambda signum, frame: sig.send_signal(sp, signum))
+    signal.signal(signal.SIGQUIT, lambda signum, frame: sig.send_signal(sp, signum))
 
-    subprocess.call(["find", ".", "-type", "f", "-printf", "run-command %12.12s %h/%f\\n"])
+    # wait for process to complete.
+    rcode = sp.wait()
 
-    print("run-command writing to keep")
+    if sig.sig != None:
+        print("run-command: terminating on signal %s" % sig.sig)
+        sys.exit(2)
+    else:
+        print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
 
-    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()
-
-if rcode == 0:
-    os.chdir("..")
-    shutil.rmtree("tmpdir")
-    shutil.rmtree("output")
-    print("run-command success")
-else:
-    print("run-command failed")
+except Exception as e:
+    print("run-command: caught exception: {}".format(e))
+
+# restore default signal handlers.
+signal.signal(signal.SIGINT, signal.SIG_DFL)
+signal.signal(signal.SIGTERM, signal.SIG_DFL)
+signal.signal(signal.SIGQUIT, signal.SIG_DFL)
+
+for l in links:
+    os.unlink(l)
+
+print("run-command: the follow output files will be saved to keep:")
+
+subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"])
+
+print("run-command: 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:
+        print("run-command: terminating on signal 2")
+        sys.exit(2)
+    except Exception as e:
+        print("run-command: caught exception: {}".format(e))
+        time.sleep(5)
 
 sys.exit(rcode)