From 2e07567c16bdd339f2305ee67e23ede60ba9a3ce Mon Sep 17 00:00:00 2001
From: Hugh Saunders <hugh@wherenow.org>
Date: Fri, 27 Mar 2015 18:24:33 +0000
Subject: [PATCH] Retry exec command via ssh_retry

This PR adds the option to retry failed ssh executions, if the failure
is caused by ssh itself, not the remote command. This can be helpful if
there are transient network issues. Retries are only implemented in the
openssh connection plugin and are disabled by default. Retries are
enabled by setting ssh_connection > retries to an integer greater
than 0.

Running a long series of playbooks, or a short playbook against a large
cluster may result in transient ssh failures, some examples logged
[here](https://trello.com/c/1yh6csEQ/13-ssh-errors).

Ansible should be able to retry an ssh connection in order to survive
transient failures.

Ansible marks a host as failed the first time it fails to contact it.
---
 lib/ansible/constants.py                    |  2 +
 v1/ansible/runner/connection_plugins/ssh.py | 67 +++++++++++++++++----
 2 files changed, 58 insertions(+), 11 deletions(-)

diff --git a/lib/ansible/constants.py b/lib/ansible/constants.py
index d24dc311a79..9c1c820421a 100644
--- a/lib/ansible/constants.py
+++ b/lib/ansible/constants.py
@@ -195,7 +195,9 @@ RETRY_FILES_SAVE_PATH          = get_config(p, DEFAULTS, 'retry_files_save_path'
 ANSIBLE_SSH_ARGS               = get_config(p, 'ssh_connection', 'ssh_args', 'ANSIBLE_SSH_ARGS', None)
 ANSIBLE_SSH_CONTROL_PATH       = get_config(p, 'ssh_connection', 'control_path', 'ANSIBLE_SSH_CONTROL_PATH', "%(directory)s/ansible-ssh-%%h-%%p-%%r")
 ANSIBLE_SSH_PIPELINING         = get_config(p, 'ssh_connection', 'pipelining', 'ANSIBLE_SSH_PIPELINING', False, boolean=True)
+ANSIBLE_SSH_RETRIES            = get_config(p, 'ssh_connection', 'retries', 'ANSIBLE_SSH_RETRIES', 0, integer=True)
 PARAMIKO_RECORD_HOST_KEYS      = get_config(p, 'paramiko_connection', 'record_host_keys', 'ANSIBLE_PARAMIKO_RECORD_HOST_KEYS', True, boolean=True)
+
 # obsolete -- will be formally removed
 ZEROMQ_PORT                    = get_config(p, 'fireball_connection', 'zeromq_port', 'ANSIBLE_ZEROMQ_PORT', 5099, integer=True)
 ACCELERATE_PORT                = get_config(p, 'accelerate', 'accelerate_port', 'ACCELERATE_PORT', 5099, integer=True)
diff --git a/v1/ansible/runner/connection_plugins/ssh.py b/v1/ansible/runner/connection_plugins/ssh.py
index 036175f6a9c..ff7e8e03c87 100644
--- a/v1/ansible/runner/connection_plugins/ssh.py
+++ b/v1/ansible/runner/connection_plugins/ssh.py
@@ -16,21 +16,22 @@
 # along with Ansible.  If not, see <http://www.gnu.org/licenses/>.
 #
 
-import os
-import re
-import subprocess
-import shlex
-import pipes
-import random
-import select
 import fcntl
-import hmac
-import pwd
 import gettext
+import hmac
+import os
+import pipes
 import pty
+import pwd
+import random
+import re
+import select
+import shlex
+import subprocess
+import time
 from hashlib import sha1
 import ansible.constants as C
-from ansible.callbacks import vvv
+from ansible.callbacks import vvv, vv
 from ansible import errors
 from ansible import utils
 
@@ -256,7 +257,51 @@ class Connection(object):
             vvv("EXEC previous known host file not found for %s" % host)
         return True
 
-    def exec_command(self, cmd, tmp_path, become_user=None, sudoable=False, executable='/bin/sh', in_data=None):
+    def exec_command(self, *args, **kwargs):
+        """ Wrapper around _exec_command to retry in the case of an ssh
+            failure
+
+            Will retry if:
+            * an exception is caught
+            * ssh returns 255
+
+            Will not retry if
+            * remaining_tries is <2
+            * retries limit reached
+            """
+        remaining_tries = C.get_config(
+            C.p, 'ssh_connection', 'retries',
+            'ANSIBLE_SSH_RETRIES', 3, integer=True) + 1
+        cmd_summary = "%s %s..." % (args[0], str(kwargs)[:200])
+        for attempt in xrange(remaining_tries):
+            pause = 2 ** attempt - 1
+            if pause > 30:
+                pause = 30
+            time.sleep(pause)
+            try:
+                return_tuple = self._exec_command(*args, **kwargs)
+            except Exception as e:
+                msg = ("ssh_retry: attempt: %d, caught exception(%s) from cmd "
+                       "(%s).") % (attempt, e, cmd_summary)
+                vv(msg)
+                if attempt == remaining_tries - 1:
+                    raise e
+                else:
+                    continue
+            # 0 = success
+            # 1-254 = remote command return code
+            # 255 = failure from the ssh command itself
+            if return_tuple[0] != 255:
+                break
+            else:
+                msg = ('ssh_retry: attempt: %d, ssh return code is 255. cmd '
+                       '(%s).') % (attempt, cmd_summary)
+                vv(msg)
+
+        return return_tuple
+
+
+    def _exec_command(self, cmd, tmp_path, become_user=None, sudoable=False, executable='/bin/sh', in_data=None):
         ''' run a command on the remote host '''
 
         if sudoable and self.runner.become and self.runner.become_method not in self.become_methods_supported: