MANY dunder variable fixes for proxies + proxy keepalive from @mirceaulinic (#38829)

* Add proxy keep alive options

* Documentation for the alive function

* MANY dunder variable fixes for proxies.

* Lint.

* Lint.

* More Lint.

doc/topics/proxyminion/index.rst

@@ -351,6 +351,13 @@ the keyword ``pass`` if there is no shutdown logic required.
 be defined in the proxymodule. The code for ``ping`` should contact the
 controlled device and make sure it is really available.
 
+``alive(opts)``: Another optional function, it is used together with the
+``proxy_keep_alive`` option (default: ``True``). This function should
+return a boolean value corresponding to the state of the connection.
+If the connection is down, will try to restart (``shutdown``
+followed by ``init``). The polling frequency is controlled using
+the ``proxy_keep_alive_interval`` option, in minutes.
+
 ``grains()``: Rather than including grains in /srv/salt/_grains or in
 the standard install directories for grains, grains can be computed and
 returned by this function.  This function will be called automatically
@@ -404,6 +411,9 @@ and status; "package" installation, and a ping.
         return True
 
 
+    def _complicated_function_that_determines_if_alive():
+        return True
+
     # Every proxy module needs an 'init', though you can
     # just put DETAILS['initialized'] = True here if nothing
     # else needs to be done.
@@ -419,6 +429,16 @@ and status; "package" installation, and a ping.
         if not DETAILS['url'].endswith('/'):
             DETAILS['url'] += '/'
 
+    def alive(opts):
+        '''
+        This function returns a flag with the connection state.
+        It is very useful when the proxy minion establishes the communication
+        via a channel that requires a more elaborated keep-alive mechanism, e.g.
+        NETCONF over SSH.
+        '''
+        log.debug('rest_sample proxy alive() called...')
+        return _complicated_function_that_determines_if_alive()
+
 
     def initialized():
         '''

salt/config/__init__.py

@@ -548,6 +548,14 @@ VALID_OPTS = {
     # but this was the default pre 2015.8.2.  This should default to
     # False in 2016.3.0
     'add_proxymodule_to_opts': bool,
+
+    # Poll the connection state with the proxy minion
+    # If enabled, this option requires the function `alive`
+    # to be implemented in the proxy module
+    'proxy_keep_alive': bool,
+
+    # Frequency of the proxy_keep_alive, in minutes
+    'proxy_keep_alive_interval': int,
     'git_pillar_base': str,
     'git_pillar_branch': str,
     'git_pillar_env': str,
@@ -1501,6 +1509,8 @@ DEFAULT_PROXY_MINION_OPTS = {
     'proxy_merge_grains_in_module': True,
     'append_minionid_config_dirs': ['cachedir', 'pidfile'],
     'default_include': 'proxy.d/*.conf',
+    'proxy_keep_alive': True,  # by default will try to keep alive the connection
+    'proxy_keep_alive_interval': 1  # frequency of the proxy keepalive in minutes
 }
 
 # ----- Salt Cloud Configuration Defaults ----------------------------------->

salt/engines/__init__.py

@@ -21,12 +21,12 @@ def start_engines(opts, proc_mgr, proxy=None):
     '''
     Fire up the configured engines!
     '''
-    utils = salt.loader.utils(opts)
+    utils = salt.loader.utils(opts, proxy=proxy)
     if opts['__role'] == 'master':
         runners = salt.loader.runner(opts, utils=utils)
     else:
         runners = []
-    funcs = salt.loader.minion_mods(opts, utils=utils)
+    funcs = salt.loader.minion_mods(opts, utils=utils, proxy=proxy)
     engines = salt.loader.engines(opts, funcs, runners, proxy=proxy)
 
     engines_opt = opts.get('engines', [])

salt/loader.py

@@ -497,8 +497,8 @@ def beacons(opts, functions, context=None, proxy=None):
         _module_dirs(opts, 'beacons'),
         opts,
         tag='beacons',
-        virtual_funcs=['__validate__'],
         pack={'__context__': context, '__salt__': functions, '__proxy__': proxy or {}},
+        virtual_funcs=['__validate__'],
     )
 
 

salt/minion.py

@@ -3221,15 +3221,17 @@ class ProxyMinion(Minion):
         # we do a sync_all here in case proxy code was installed by
         # SPM or was manually placed in /srv/salt/_modules etc.
         self.functions['saltutil.sync_all'](saltenv=self.opts['environment'])
+        # Pull in the utils
+        self.utils = salt.loader.utils(self.opts)
 
         # Pull in the utils
         self.utils = salt.loader.utils(self.opts)
 
         # Then load the proxy module
-        self.proxy = salt.loader.proxy(self.opts)
+        self.proxy = salt.loader.proxy(self.opts, utils=self.utils)
 
         # And re-load the modules so the __proxy__ variable gets injected
-        self.functions, self.returners, self.function_errors, self.executors = self._load_modules()
+        self._setup_core()
         self.functions.pack['__proxy__'] = self.proxy
         self.proxy.pack['__salt__'] = self.functions
         self.proxy.pack['__ret__'] = self.returners
@@ -3258,17 +3260,23 @@ class ProxyMinion(Minion):
 
         self.opts['grains'] = salt.loader.grains(self.opts, proxy=self.proxy)
 
-        self.serial = salt.payload.Serial(self.opts)
-        self.mod_opts = self._prep_mod_opts()
-        self.matcher = Matcher(self.opts, self.functions)
-        self.beacons = salt.beacons.Beacon(self.opts, self.functions)
-        uid = salt.utils.get_uid(user=self.opts.get('user', None))
-        self.proc_dir = get_proc_dir(self.opts['cachedir'], uid=uid)
+        self.setup_beacons()
 
+        if self.connected and self.opts['pillar']:
+            # The pillar has changed due to the connection to the master.
+            # Reload the functions so that they can use the new pillar data.
+            self.functions, self.returners, self.function_errors, self.executors = self._load_modules()
+            if hasattr(self, 'schedule'):
+                self.schedule.functions = self.functions
+                self.schedule.returners = self.returners
+
+        if not hasattr(self, 'schedule'):
             self.schedule = salt.utils.schedule.Schedule(
                 self.opts,
                 self.functions,
-            self.returners)
+                self.returners,
+                cleanup=[master_event(type='alive')],
+                proxy=self.proxy)
 
         # add default scheduling jobs to the minions scheduler
         if self.opts['mine_enabled'] and 'mine.update' in self.functions:
@@ -3288,7 +3296,8 @@ class ProxyMinion(Minion):
 
         # add master_alive job if enabled
         if (self.opts['transport'] != 'tcp' and
-                self.opts['master_alive_interval'] > 0):
+                self.opts['master_alive_interval'] > 0 and
+                self.connected):
             self.schedule.add_job({
                 master_event(type='alive', master=self.opts['master']):
                 {
@@ -3301,6 +3310,7 @@ class ProxyMinion(Minion):
                                'connected': True}
                 }
             }, persist=True)
+
             if self.opts['master_failback'] and \
                     'master_list' in self.opts and \
                     self.opts['master'] != self.opts['master_list'][0]:
@@ -3315,12 +3325,35 @@ class ProxyMinion(Minion):
                         'kwargs': {'master': self.opts['master_list'][0]}
                     }
                 }, persist=True)
+
             else:
                 self.schedule.delete_job(master_event(type='failback'), persist=True)
         else:
             self.schedule.delete_job(master_event(type='alive', master=self.opts['master']), persist=True)
             self.schedule.delete_job(master_event(type='failback'), persist=True)
 
+        # proxy keepalive
+        proxy_alive_fn = fq_proxyname+'.alive'
+        if proxy_alive_fn in self.proxy and 'status.proxy_reconnect' in self.functions and \
+           ('proxy_keep_alive' not in self.opts or ('proxy_keep_alive' in self.opts and self.opts['proxy_keep_alive'])):
+            # if `proxy_keep_alive` is either not specified, either set to False does not retry reconnecting
+            self.schedule.add_job({
+                '__proxy_keepalive':
+                {
+                    'function': 'status.proxy_reconnect',
+                    'minutes': self.opts.get('proxy_keep_alive_interval', 1),  # by default, check once per minute
+                    'jid_include': True,
+                    'maxrunning': 1,
+                    'return_job': False,
+                    'kwargs': {
+                        'proxy_name': fq_proxyname
+                    }
+                }
+            }, persist=True)
+            self.schedule.enable_schedule()
+        else:
+            self.schedule.delete_job('__proxy_keepalive', persist=True)
+
         #  Sync the grains here so the proxy can communicate them to the master
         self.functions['saltutil.sync_grains'](saltenv=self.opts['environment'])
         self.grains_cache = self.opts['grains']

salt/modules/rest_sample_utils.py

@@ -30,6 +30,5 @@ def get_test_string():
     .. code-block:: bash
 
         salt 'rest-sample-proxy' rest_sample.get_test_string
-
     '''
     return __proxy__['rest_sample.test_from_state']()

salt/modules/status.py

@@ -14,6 +14,7 @@ import fnmatch
 import collections
 import copy
 import time
+import logging
 
 # Import 3rd-party libs
 import salt.ext.six as six
@@ -29,6 +30,8 @@ from salt.utils.network import host_to_ips as _host_to_ips
 from salt.ext.six.moves import zip
 from salt.exceptions import CommandExecutionError
 
+log = logging.getLogger(__file__)
+
 __virtualname__ = 'status'
 __opts__ = {}
 
@@ -1115,6 +1118,46 @@ def ping_master(master):
     return result
 
 
+def proxy_reconnect(proxy_name, opts=None):
+    '''
+    Forces proxy minion reconnection when not alive.
+
+    proxy_name
+        The virtual name of the proxy module.
+
+    opts: None
+        Opts dictionary.
+    '''
+
+    if not opts:
+        opts = __opts__
+
+    if 'proxy' not in opts:
+        return False  # fail
+
+    proxy_keepalive_fn = proxy_name+'.alive'
+    if proxy_keepalive_fn not in __proxy__:
+        return False  # fail
+
+    is_alive = __proxy__[proxy_keepalive_fn](opts)
+    if not is_alive:
+        minion_id = opts.get('proxyid', '') or opts.get('id', '')
+        log.info('{minion_id} ({proxy_name} proxy) is down. Restarting.'.format(
+                minion_id=minion_id,
+                proxy_name=proxy_name
+            )
+        )
+        __proxy__[proxy_name+'.shutdown'](opts)  # safely close connection
+        __proxy__[proxy_name+'.init'](opts)  # reopen connection
+        log.debug('Restarted {minion_id} ({proxy_name} proxy)!'.format(
+                minion_id=minion_id,
+                proxy_name=proxy_name
+            )
+        )
+
+    return True  # success
+
+
 def time_(format='%A, %d. %B %Y %I:%M%p'):
     '''
     .. versionadded:: 2016.3.0

salt/proxy/rest_sample.py

@@ -33,11 +33,11 @@ def __virtual__():
     log.debug('rest_sample proxy __virtual__() called...')
     return True
 
-
 # Every proxy module needs an 'init', though you can
 # just put DETAILS['initialized'] = True here if nothing
 # else needs to be done.
 
+
 def init(opts):
     log.debug('rest_sample proxy init() called...')
     DETAILS['initialized'] = True
@@ -59,6 +59,13 @@ def initialized():
     return DETAILS.get('initialized', False)
 
 
+def alive(opts):
+    log.debug('=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-')
+    log.debug('proxys alive() fn called')
+    log.debug('=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-')
+    return False
+
+
 def id(opts):
     '''
     Return a unique ID for this proxy minion.  This ID MUST NOT CHANGE.

salt/state.py

@@ -861,7 +861,7 @@ class State(object):
         Read the state loader value and loadup the correct states subsystem
         '''
         if self.states_loader == 'thorium':
-            self.states = salt.loader.thorium(self.opts, self.functions, {})  # TODO: Add runners
+            self.states = salt.loader.thorium(self.opts, self.functions, {})  # TODO: Add runners, proxy?
         else:
             self.states = salt.loader.states(self.opts, self.functions, self.utils,
                                              self.serializers, proxy=self.proxy)

salt/utils/schedule.py

@@ -378,7 +378,7 @@ class Schedule(object):
     '''
     instance = None
 
-    def __new__(cls, opts, functions, returners=None, intervals=None, cleanup=None):
+    def __new__(cls, opts, functions, returners=None, intervals=None, cleanup=None, proxy=None):
         '''
         Only create one instance of Schedule
         '''
@@ -388,18 +388,19 @@ class Schedule(object):
             # it in a WeakValueDictionary-- which will remove the item if no one
             # references it-- this forces a reference while we return to the caller
             cls.instance = object.__new__(cls)
-            cls.instance.__singleton_init__(opts, functions, returners, intervals, cleanup)
+            cls.instance.__singleton_init__(opts, functions, returners, intervals, cleanup, proxy)
         else:
             log.debug('Re-using Schedule')
         return cls.instance
 
     # has to remain empty for singletons, since __init__ will *always* be called
-    def __init__(self, opts, functions, returners=None, intervals=None, cleanup=None):
+    def __init__(self, opts, functions, returners=None, intervals=None, cleanup=None, proxy=None):
         pass
 
     # an init for the singleton instance to call
-    def __singleton_init__(self, opts, functions, returners=None, intervals=None, cleanup=None):
+    def __singleton_init__(self, opts, functions, returners=None, intervals=None, cleanup=None, proxy=None):
         self.opts = opts
+        self.proxy = proxy
         self.functions = functions
         if isinstance(intervals, dict):
             self.intervals = intervals
@@ -742,8 +743,8 @@ class Schedule(object):
             # This also needed for ZeroMQ transport to reset all functions
             # context data that could keep paretns connections. ZeroMQ will
             # hang on polling parents connections from the child process.
-            self.functions = salt.loader.minion_mods(self.opts)
-            self.returners = salt.loader.returners(self.opts, self.functions)
+            self.functions = salt.loader.minion_mods(self.opts, proxy=self.proxy)
+            self.returners = salt.loader.returners(self.opts, self.functions, proxy=self.proxy)
         ret = {'id': self.opts.get('id', 'master'),
                'fun': func,
                'fun_args': [],