More work on signal handling

[arvados.git] / crunch_scripts / run-command

diff --git a/crunch_scripts/run-command b/crunch_scripts/run-command
index a8aae24f16aae813adc8083684787c054a1a97be..2eba372357fe95ab5228bd285ee3264a2fe0427c 100755 (executable)
--- a/crunch_scripts/run-command
+++ b/crunch_scripts/run-command
@@ -8,6 +8,8 @@ import sys
 import shutil
 import subst
 import time
+import arvados.commands.put as put
+import signal
 
 os.umask(0077)
 
@@ -37,11 +39,23 @@ 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: {} written {} total\n".format(
+        bytes_written, -1 if (bytes_expected is None) else bytes_expected)
+
+def send_signal(sp, signum, sig):
+    sp.send_signal(signum)
+    sig.append(signum)
+
 try:
     cmd = []
     for c in p["command"]:
@@ -55,14 +69,32 @@ try:
 
     print("run-command: {}{}".format(' '.join(cmd), (" > " + stdoutname) if stdoutname != None else ""))
 
-    rcode = subprocess.call(cmd, stdout=stdoutfile)
+    sp = subprocess.Popen(cmd, shell=False, stdout=stdoutfile)
+    sig = []
+
+    # forward signals to the process.
+    signal.signal(signal.SIGINT, lambda signum, frame: send_signal(sp, signum, sig))
+    signal.signal(signal.SIGTERM, lambda signum, frame: send_signal(sp, signum, sig))
+    signal.signal(signal.SIGQUIT, lambda signum, frame: send_signal(sp, signum, sig))
+
+    # wait for process to complete.
+    rcode = sp.wait()
 
-    print("run-command: completed with exit code %i" % rcode)
+    if len(sig) > 0:
+        print("run-command: terminating on signal %s" % sig[0])
+        sys.exit(rcode)
+    else:
+        print("run-command: completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
 
 except Exception as e:
     print("run-command: caught exception: {}".format(e))
 
 finally:
+    # 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)
 
@@ -72,11 +104,15 @@ finally:
 
     print("run-command: start writing output to keep")
 
-    out = arvados.CollectionWriter()
-    out.write_directory_tree(".", max_manifest_depth=0)
     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(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={
@@ -85,16 +121,11 @@ finally:
                                                      'progress':1.0
                                                  }).execute()
             done = True
+        except KeyboardInterrupt:
+            print("run-command: terminating on signal SIGINT")
+            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")
-else:
-    print("run-command: failed")
-
 sys.exit(rcode)