Make run_command process communication smarter

The run_command module function will now poll stdout/stderr for
data rather than using the builtin command communicate(), which can
hang under certain circumstances.

Fixes #7452
Fixes #7748
Fixes #8163

lib/ansible/module_utils/basic.py

@@ -53,6 +53,7 @@ import sys
 import syslog
 import types
 import time
+import select
 import shutil
 import stat
 import tempfile
@@ -1163,7 +1164,7 @@ class AnsibleModule(object):
             # rename might not preserve context
             self.set_context_if_different(dest, context, False)
 
-    def run_command(self, args, check_rc=False, close_fds=False, executable=None, data=None, binary_data=False, path_prefix=None, cwd=None, use_unsafe_shell=False):
+    def run_command(self, args, check_rc=False, close_fds=True, executable=None, data=None, binary_data=False, path_prefix=None, cwd=None, use_unsafe_shell=False):
         '''
         Execute a command, returns rc, stdout, and stderr.
         args is the command to run
@@ -1237,7 +1238,7 @@ class AnsibleModule(object):
             executable=executable,
             shell=shell,
             close_fds=close_fds,
-            stdin= st_in,
+            stdin=st_in,
             stdout=subprocess.PIPE,
             stderr=subprocess.PIPE 
         )
@@ -1260,10 +1261,48 @@ class AnsibleModule(object):
         try:
             cmd = subprocess.Popen(args, **kwargs)
 
+            # the communication logic here is essentially taken from that
+            # of the _communicate() function in ssh.py
+
+            stdout = ''
+            stderr = ''
+            rpipes = [cmd.stdout, cmd.stderr]
+
             if data:
                 if not binary_data:
                     data += '\n'
-            out, err = cmd.communicate(input=data)
+                cmd.stdin.write(data)
+                cmd.stdin.close()
+
+            while True:
+                rfd, wfd, efd = select.select(rpipes, [], rpipes, 1)
+                if cmd.stdout in rfd:
+                    dat = os.read(cmd.stdout.fileno(), 9000)
+                    stdout += dat
+                    if dat == '':
+                        rpipes.remove(cmd.stdout)
+                if cmd.stderr in rfd:
+                    dat = os.read(cmd.stderr.fileno(), 9000)
+                    stderr += dat
+                    if dat == '':
+                        rpipes.remove(cmd.stderr)
+                # only break out if no pipes are left to read or
+                # the pipes are completely read and
+                # the process is terminated
+                if (not rpipes or not rfd) and cmd.poll() is not None:
+                    break
+                # No pipes are left to read but process is not yet terminated
+                # Only then it is safe to wait for the process to be finished
+                # NOTE: Actually cmd.poll() is always None here if rpipes is empty
+                elif not rpipes and cmd.poll() == None:
+                    cmd.wait()
+                    # The process is terminated. Since no pipes to read from are
+                    # left, there is no need to call select() again.
+                    break
+
+            cmd.stdout.close()
+            cmd.stderr.close()
+
             rc = cmd.returncode
         except (OSError, IOError), e:
             self.fail_json(rc=e.errno, msg=str(e), cmd=clean_args)
@@ -1271,13 +1310,13 @@ class AnsibleModule(object):
             self.fail_json(rc=257, msg=traceback.format_exc(), cmd=clean_args)
 
         if rc != 0 and check_rc:
-            msg = err.rstrip()
+            msg = stderr.rstrip()
-            self.fail_json(cmd=clean_args, rc=rc, stdout=out, stderr=err, msg=msg)
+            self.fail_json(cmd=clean_args, rc=rc, stdout=stdout, stderr=stderr, msg=msg)
 
         # reset the pwd
         os.chdir(prev_dir)
 
-        return (rc, out, err)
+        return (rc, stdout, stderr)
 
     def append_to_file(self, filename, str):
         filename = os.path.expandvars(os.path.expanduser(filename))