bin/takeover script

slapos/recipe/addresiliency/__init__.py

@@ -31,7 +31,10 @@ import os
 
 class Recipe(GenericSlapRecipe):
     """ This class provides the installation of the resilience
-        script on the partition.
+        scripts on the partition.
+
+        bin/takeover will perform a rename (must be run manually).
+        bin/bully will monitor, run elections and perform renames when needed.
     """
 
     def _install(self):
@@ -55,12 +58,12 @@ class Recipe(GenericSlapRecipe):
         slap_connection = self.buildout['slap-connection']
 
         if self.optionIsTrue('enable-bully-service', default=False):
-            wrapper_dir = self.options['services']
+            bully_dir = self.options['services']
         else:
-            wrapper_dir = self.options['bin']
+            bully_dir = self.options['bin']
 
-        wrapper = self.createPythonScript(
-            name=os.path.join(wrapper_dir, self.parameter_dict['wrapper']),
+        bully_wrapper = self.createPythonScript(
+            name=os.path.join(bully_dir, self.options['wrapper-bully']),
             absolute_function='slapos.recipe.addresiliency.bully.run',
             arguments={
                 'confpath': confpath,
@@ -73,7 +76,22 @@ class Recipe(GenericSlapRecipe):
                 'namebase': self.parameter_dict['namebase'],
             })
 
-        path_list.append(wrapper)
+        path_list.append(bully_wrapper)
+
+        takeover_wrapper = self.createPythonScript(
+            name=os.path.join(self.options['bin'], self.options['wrapper-takeover']),
+            absolute_function='slapos.recipe.addresiliency.takeover.run',
+            arguments={
+                'server_url': slap_connection['server-url'],
+                'key_file': slap_connection.get('key-file'),
+                'cert_file': slap_connection.get('cert-file'),
+                'computer_id': slap_connection['computer-id'],
+                'partition_id': slap_connection['partition-id'],
+                'software': slap_connection['software-release-url'],
+                'namebase': self.parameter_dict['namebase'],
+            })
+
+        path_list.append(takeover_wrapper)
 
         return path_list
 

slapos/recipe/addresiliency/bully.py

@@ -6,7 +6,7 @@ import socket
 import thread
 import time
 
-from slapos import slap as slapmodule
+import slapos.recipe.addresiliency.renamer
 import slapos
 
 log = logging.getLogger(__name__)
@@ -29,86 +29,6 @@ STATE_REORGANIZATION = 'reorganization'
 
 
 
-
-class Renamer(object):
-    def __init__(self, server_url, key_file, cert_file, computer_guid,
-                 partition_id, software_release, namebase):
-        self.server_url = server_url
-        self.key_file = key_file
-        self.cert_file = cert_file
-        self.computer_guid = computer_guid
-        self.partition_id = partition_id
-        self.software_release = software_release
-        self.namebase = namebase
-
-
-    def _failover(self):
-        """\
-        This method does
-
-        - retrieve the broken computer partition
-        - change its reference to 'broken-...' and its software type to 'frozen'
-        - retrieve the winner computer partition (attached to this process)
-        - change its reference and software type to replace the broken one
-
-        Then, after running slapgrid-cp a few times, the winner takes over and
-        a new cp is created to replace it as an importer.
-        """
-
-        # TODO: replace hardcoded strings with values from the API
-
-        slap = slapmodule.slap()
-        slap.initializeConnection(self.server_url, self.key_file, self.cert_file)
-
-        # partition that will take over.
-        cp_winner = slap.registerComputerPartition(computer_guid=self.computer_guid,
-                                                   partition_id=self.partition_id)
-        # XXX although we can already rename cp_winner, to change its software type we need to
-        # get hold of the root cp as well
-
-        root_partition_id = 'slappart0'       # XXX hardcoded. what's the API for this?
-
-        cp_root = slap.registerComputerPartition(computer_guid=self.computer_guid,
-                                                 partition_id=root_partition_id)
-
-        cp_exporter_ref = self.namebase + '0'       # this is ok. the boss is always number zero.
-
-        # partition to be deactivated
-        cp_broken = cp_root.request(software_release=self.software_release,
-                                    software_type='frozen',
-                                    state='stopped',
-                                    partition_reference=cp_exporter_ref)
-
-        broken_new_ref = 'broken-{}'.format(time.strftime("%d-%b_%H:%M:%S", time.gmtime()))
-
-        # XXX can we retrieve and log the old reference name?
-        log.debug("Renaming {}: {}".format(cp_broken.getId(), broken_new_ref))
-
-        cp_broken.rename(new_name=broken_new_ref)
-
-        cp_broken.stopped()
-
-        log.debug("Renaming {}: {}".format(cp_broken.getId(), cp_exporter_ref))
-
-        # update software type and name for the partition that will take over
-
-        cp_winner_ref = self.namebase + '2'     # XXX hardcoded. what's the API for this?
-
-        cp_root.request(software_release=self.software_release,
-                        software_type=self.namebase+'-export',
-                        partition_reference=cp_winner_ref).rename(new_name=cp_exporter_ref)
-
-
-
-    def failover(self):
-        try:
-            self._failover()
-            log.info('Renaming done')
-        except slapos.slap.slap.ServerError:
-            log.info('Internal server error')
-
-
-
 ## Leader is always number 0
 
 class ResilientInstance(object):
@@ -291,7 +211,7 @@ class Wrapper(object):
 def run(args):
     confpath = args.pop('confpath')
 
-    renamer = Renamer(server_url = args.pop('server_url'),
+    renamer = slapos.recipe.addresiliency.renamer.Renamer(server_url = args.pop('server_url'),
                                                           key_file = args.pop('key_file'),
                                                           cert_file = args.pop('cert_file'),
                                                           computer_guid = args.pop('computer_id'),

slapos/recipe/addresiliency/renamer.py

+# -*- coding: utf-8 -*-
+
+import logging
+import time
+
+from slapos import slap as slapmodule
+import slapos
+
+log = logging.getLogger(__name__)
+logging.basicConfig(level=logging.DEBUG)
+
+
+
+class Renamer(object):
+    def __init__(self, server_url, key_file, cert_file, computer_guid,
+                 partition_id, software_release, namebase):
+        self.server_url = server_url
+        self.key_file = key_file
+        self.cert_file = cert_file
+        self.computer_guid = computer_guid
+        self.partition_id = partition_id
+        self.software_release = software_release
+        self.namebase = namebase
+
+
+    def _failover(self):
+        """\
+        This method does
+
+        - retrieve the broken computer partition
+        - change its reference to 'broken-...' and its software type to 'frozen'
+        - retrieve the winner computer partition (attached to this process)
+        - change its reference and software type to replace the broken one
+
+        Then, after running slapgrid-cp a few times, the winner takes over and
+        a new cp is created to replace it as an importer.
+        """
+
+        # TODO: replace hardcoded strings with values from the API
+
+        slap = slapmodule.slap()
+        slap.initializeConnection(self.server_url, self.key_file, self.cert_file)
+
+        # partition that will take over.
+        cp_winner = slap.registerComputerPartition(computer_guid=self.computer_guid,
+                                                   partition_id=self.partition_id)
+        # XXX although we can already rename cp_winner, to change its software type we need to
+        # get hold of the root cp as well
+
+        root_partition_id = 'slappart0'       # XXX hardcoded. what's the API for this?
+
+        cp_root = slap.registerComputerPartition(computer_guid=self.computer_guid,
+                                                 partition_id=root_partition_id)
+
+        cp_exporter_ref = self.namebase + '0'       # this is ok. the boss is always number zero.
+
+        # partition to be deactivated
+        cp_broken = cp_root.request(software_release=self.software_release,
+                                    software_type='frozen',
+                                    state='stopped',
+                                    partition_reference=cp_exporter_ref)
+
+        broken_new_ref = 'broken-{}'.format(time.strftime("%d-%b_%H:%M:%S", time.gmtime()))
+
+        # XXX can we retrieve and log the old reference name?
+        log.debug("Renaming {}: {}".format(cp_broken.getId(), broken_new_ref))
+
+        cp_broken.rename(new_name=broken_new_ref)
+
+        cp_broken.stopped()
+
+        log.debug("Renaming {}: {}".format(cp_broken.getId(), cp_exporter_ref))
+
+        # update software type and name for the partition that will take over
+
+        cp_winner_ref = self.namebase + '2'     # XXX hardcoded. what's the API for this?
+
+        cp_root.request(software_release=self.software_release,
+                        software_type=self.namebase+'-export',
+                        partition_reference=cp_winner_ref).rename(new_name=cp_exporter_ref)
+
+
+
+    def failover(self):
+        try:
+            self._failover()
+            log.info('Renaming done')
+        except slapos.slap.slap.ServerError:
+            log.info('Internal server error')
+

slapos/recipe/addresiliency/takeover.py

+# -*- coding: utf-8 -*-
+
+import slapos.recipe.addresiliency.renamer
+
+def run(args):
+    renamer = slapos.recipe.addresiliency.renamer.Renamer(server_url = args.pop('server_url'),
+                                                          key_file = args.pop('key_file'),
+                                                          cert_file = args.pop('cert_file'),
+                                                          computer_guid = args.pop('computer_id'),
+                                                          partition_id = args.pop('partition_id'),
+                                                          software_release = args.pop('software'),
+                                                          namebase = args.pop('namebase'))
+
+    renamer.failover()
+
+

stack/resilient/buildout.cfg

@@ -22,7 +22,7 @@ parts =
 recipe = slapos.recipe.template
 url = ${:_profile_base_location_}/pbsready.cfg.in
 output = ${buildout:directory}/pbsready.cfg
-md5sum = b6102416d000cae81dd2b06268946ea9
+md5sum = af02afc439530a6cef9f5a21a25f3363
 mode = 0644
 
 [pbsready-import]
@@ -53,7 +53,7 @@ mode = 0644
 [template-replicated]
 recipe = slapos.recipe.download
 url = ${:_profile_base_location_}/template-replicated.cfg.in
-md5sum = 1017d919dbf41904f04f5c17dcb574fa
+md5sum = 63b5649f3cf1c9a77315382793d9593f
 mode = 0644
 destination = ${buildout:directory}/template-replicated.cfg.in
 

stack/resilient/pbsready.cfg.in

@@ -45,6 +45,8 @@ notifier-callbacks = $${basedirectory:notifier}/callbacks
 # If false, they can be run with bin/bullly for all the PBSReady instances.
 enable-bully-service = False
 recipe = slapos.cookbook:addresiliency
+wrapper-bully = bully
+wrapper-takeover = takeover
 services = $${basedirectory:services}
 bin = $${rootdirectory:bin}
 etc = $${rootdirectory:etc}

stack/resilient/template-replicated.cfg.in

@@ -2,8 +2,6 @@
 
 ## Tells the Backupable recipe that we want a backup
 [resilient]
-config-script = bully.py
-config-wrapper = bully
 config-namebase = {{namebase}}
 
 ## Every request is double to provide the 3 IPs.
@@ -15,7 +13,7 @@ software-type = {{typeexport}}
 name = {{namebase}}0
 return = url ssh-public-key ssh-url notification-id ip
 
-config = number script wrapper authorized-key notify ip-list namebase
+config = number authorized-key notify ip-list namebase
 config-number = 0
 config-authorized-key = {% for id in range(1,nbbackup|int) %} ${request-pbs-{{namebase}}-{{id}}:connection-ssh-key}{% endfor %}
 config-notify = {% for id in range(1,nbbackup|int) %} ${request-pbs-{{namebase}}-{{id}}:connection-notification-url}{% endfor %}
@@ -36,7 +34,7 @@ return = url ssh-public-key ssh-url notification-url ip
 
 pbs-notification-id = ${slap-connection:computer-id}-${slap-connection:partition-id}-{{namebase}}-push
 
-config = number script wrapper authorized-key on-notification ip-list namebase
+config = number authorized-key on-notification ip-list namebase
 config-number = {{id}}
 config-authorized-key = ${request-pbs-{{namebase}}-{{id}}:connection-ssh-key}
 config-on-notification = ${request-pbs-{{namebase}}-{{id}}:connection-feeds-url}${:pbs-notification-id}
@@ -60,7 +58,7 @@ software-url = ${slap-connection:software-release-url}
 software-type = {{typeexport}}
 return = url ssh-public-key ssh-url notification-id ip
 
-config = number script wrapper authorized-key notify ip-list namebase
+config = number authorized-key notify ip-list namebase
 config-number = 0
 config-authorized-key = {% for id in range(1,nbbackup|int) %} ${request-pbs-{{namebase}}-{{id}}:connection-ssh-key}{% endfor %}
 config-notify = {% for id in range(1,nbbackup|int) %} ${request-pbs-{{namebase}}-{{id}}:connection-notification-url}{% endfor %}
@@ -81,7 +79,7 @@ return = url ssh-public-key ssh-url notification-url
 
 pbs-notification-id = ${slap-connection:computer-id}-${slap-connection:partition-id}-{{namebase}}-push
 
-config = number script wrapper authorized-key on-notification ip-list namebase
+config = number authorized-key on-notification ip-list namebase
 config-number = {{id}}
 config-authorized-key = ${request-pbs-{{namebase}}-{{id}}:connection-ssh-key}
 config-on-notification = ${request-pbs-{{namebase}}-{{id}}:connection-feeds-url}${:pbs-notification-id}