Merge branch 'kvm'

slapos/recipe/addresiliency/renamer.py

-# -*- coding: utf-8 -*-
-
-import logging
-import time
-
-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 to replace the broken one.
-          later, slapgrid will change its software_type as well.
-
-        Then, after running slapgrid-cp a few times, the winner takes over and
-        a new cp is created to replace it as an importer.
-        """
-
-        slap = slapos.slap.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
-
-        cp_exporter_ref = self.namebase + '0'       # this is ok. the boss is always number zero.
-
-        # partition to be deactivated
-        cp_broken = cp_winner.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()))
-
-        log.debug("Renaming {}: {}".format(cp_broken.getId(), broken_new_ref))
-
-        cp_broken.rename(new_name=broken_new_ref)
-
-        log.debug("Renaming {}: {}".format(cp_winner.getId(), cp_exporter_ref))
-
-        # update name (and later, software type) for the partition that will take over
-
-        cp_winner.rename(new_name=cp_exporter_ref)
-        cp_winner.bang(message='partitions have been renamed!')
-
-
-
-    def failover(self):
-        try:
-            self._failover()
-            log.info('Renaming done')
-        except slapos.slap.ServerError:
-            log.info('Internal server error')
-

slapos/recipe/addresiliency/takeover.py

 # -*- coding: utf-8 -*-
+import logging
+import time
 
-import slapos.recipe.addresiliency.renamer
+import slapos
+
+log = logging.getLogger(__name__)
+logging.basicConfig(level=logging.DEBUG)
+
+def takeover(server_url, key_file, cert_file, computer_guid,
+             partition_id, software_release, namebase,
+             winner_instance_suffix = None):
+  """
+  This function 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 to replace the broken one.
+    later, slapgrid will change its software_type as well.
+
+  Then, after running slapgrid-cp a few times, the winner takes over and
+  a new cp is created to replace it as an importer.
+  """
+
+  slap = slapos.slap.slap()
+  slap.initializeConnection(server_url, key_file, cert_file)
+  current_partition = slap.registerComputerPartition(computer_guid=computer_guid,
+                                                     partition_id=partition_id)
+
+  # partition that will take over.
+  if winner_instance_suffix:
+    winner_instance_name = namebase + winner_instance_suffix
+    # XXX: we hardcode a lot of values here, because request is a settergetter, all at once.
+    cp_winner = current_partition.request(software_release=software_release,
+                                  software_type='%s-import' % namebase,
+                                  partition_reference=winner_instance_name)
+  else:
+    # This script is run in the winning partition: use this one as winner
+    cp_winner = current_partition
+  # XXX although we can already rename cp_winner, to change its software type we need to
+  # get hold of the root cp as well
+
+  cp_exporter_ref = namebase + '0'       # this is ok. the boss is always number zero.
+
+  # partition to be deactivated
+  cp_broken = cp_winner.request(software_release=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()))
+
+  log.debug("Renaming {}: {}".format(cp_broken.getId(), broken_new_ref))
+
+  cp_broken.rename(new_name=broken_new_ref)
+
+  log.debug("Renaming {}: {}".format(cp_winner.getId(), cp_exporter_ref))
+
+  # update name (and later, software type) for the partition that will take over
+
+  cp_winner.rename(new_name=cp_exporter_ref)
+  cp_winner.bang(message='partitions have been renamed!')
+  # Note: Root instance will reconfigure itself the winning instance (software_type
+  # and parameters.)
 
 def run(args):
-    renamer = slapos.recipe.addresiliency.renamer.Renamer(server_url = args.pop('server_url'),
+  slapos.recipe.addresiliency.takeover.takeover(server_url = args.pop('server_url'),
                                                 key_file = args.pop('key_file'),
                                                 cert_file = args.pop('cert_file'),
                                                 computer_guid = args.pop('computer_id'),
@@ -11,6 +73,3 @@ def run(args):
                                                 software_release = args.pop('software'),
                                                 namebase = args.pop('namebase'))
 
-    renamer.failover()
-
-

slapos/recipe/kvm/__init__.py

@@ -48,16 +48,16 @@ class Recipe(GenericBaseRecipe):
     if not self.isTrueValue(self.options.get('use-tap')):
       # XXX This could be done using Jinja.
       for port in self.options['nat-rules'].split():
-        ipv6_port = int(port) + 10000
+        tunnel_port = int(port) + 10000
         tunnel_path = self.createExecutable(
-            '%s-%sto%s' % (self.options['6tunnel-wrapper-path'], port, ipv6_port),
+            '%s-%s' % (self.options['6tunnel-wrapper-path'], tunnel_port),
             self.substituteTemplate(
                 self.getTemplateFilename('6to4.in'),
                 {
                     'ipv6': self.options['ipv6'],
-                    'ipv6_port': ipv6_port,
+                    'ipv6_port': tunnel_port,
                     'ipv4': self.options['ipv4'],
-                    'ipv4_port': port,
+                    'ipv4_port': tunnel_port,
                     'shell_path': self.options['shell-path'],
                     '6tunnel_path': self.options['6tunnel-path'],
                 },

slapos/recipe/kvm/template/kvm_run.in

@@ -18,7 +18,7 @@ nbd_list = (('%(nbd-host)s', %(nbd-port)s), ('%(nbd2-host)s', %(nbd2-port)s))
 default_disk_image = '%(default-disk-image)s'
 disk_path = '%(disk-path)s'
 virtual_hard_drive_url = '%(virtual-hard-drive-url)s'.strip()
-virtual_hard_drive_md5_url = '%(virtual-hard-drive-md5-url)s'.strip()
+virtual_hard_drive_md5sum = '%(virtual-hard-drive-md5sum)s'.strip()
 nat_rules = '%(nat-rules)s'.strip()
 use_tap = '%(use-tap)s'
 tap_interface = '%(tap-interface)s'
@@ -59,18 +59,27 @@ def getSocketStatus(host, port):
 
 # Download existing hard drive if needed at first boot
 if not os.path.exists(disk_path) and virtual_hard_drive_url != '':
+  print('Downloading virtual hard drive...')
   urllib.urlretrieve(virtual_hard_drive_url, disk_path)
+  md5sum = virtual_hard_drive_md5sum.strip()
+  if md5sum:
+    print('Checking MD5 checksum...')
     local_md5sum = md5Checksum(disk_path)
-  md5sum = urllib.urlopen(virtual_hard_drive_md5_url).read().strip()
     if local_md5sum != md5sum:
       os.remove(disk_path)
-    raise Exception('MD5 mismatch.')
+      raise Exception('MD5 mismatch. MD5 of local file is %%s, Specified MD5 is %%s.' %% (
+          local_md5sum, md5sum))
+    print('MD5sum check passed.')
+  else:
+    print('Warning: not checksum specified.')
 
 # Create disk if doesn't exist
 # XXX: move to Buildout profile
 if not os.path.exists(disk_path):
+  print('Creating virtual hard drive...')
   subprocess.Popen([qemu_img_path, 'create' ,'-f', 'qcow2',
       disk_path, '%%sG' %% disk_size])
+  print('Done.')
 
 # Generate network parameters
 # XXX: use_tap should be a boolean

software/kvm/common.cfg

@@ -42,6 +42,7 @@ eggs =
   websockify
   slapos.cookbook
   slapos.toolbox
+  erp5.util
 
 [http-proxy]
 # https://github.com/nodejitsu/node-http-proxy
@@ -78,7 +79,7 @@ command =
 [template]
 recipe = slapos.recipe.template
 url = ${:_profile_base_location_}/instance.cfg.in
-md5sum = 0e84223169661462f439c164d62c2a6a
+#md5sum = 461187060d12a6bea4d5b9699687c0e9
 output = ${buildout:directory}/template.cfg
 mode = 0644
 
@@ -91,9 +92,17 @@ mode = 0644
 
 [template-kvm-resilient]
 recipe = hexagonit.recipe.download
-url = ${:_profile_base_location_}/instance-kvm-resilient.cfg.jinja
+url = ${:_profile_base_location_}/instance-kvm-resilient.cfg.jinja2
 mode = 644
-md5sum = 3ee64c654aae503b93b39e9ccd6d3643
+md5sum = 6753004b582c0470bd028253ce1964ad
+download-only = true
+
+[template-kvm-resilient-test]
+recipe = slapos.recipe.template
+url = ${:_profile_base_location_}/instance-kvm-resilient-test.cfg.in
+#md5sum = 9dead7572072307b6c38cdf47052225b
+output = ${buildout:directory}/template-kvm-resilient-test.cfg
+mode = 0644
 
 [template-kvm-import]
 recipe = slapos.recipe.template
@@ -138,3 +147,4 @@ url = ${:_profile_base_location_}/instance-frontend.cfg.in
 md5sum = cdb690495e9eb007d2b7d2f8e12f5c59
 output = ${buildout:directory}/template-frontend.cfg
 mode = 0644
+

software/kvm/development.cfg

@@ -6,11 +6,13 @@ extends =
 parts +=
   slapos.cookbook-repository
   slapos.toolbox-repository
+  erp5.util-repository
   check-recipe
 
 develop =
   ${:parts-directory}/slapos.cookbook-repository
   ${:parts-directory}/slapos.toolbox-repository
+  ${:parts-directory}/erp5.util-repository
 
 [slapos.cookbook-repository]
 recipe = slapos.recipe.build:gitclone
@@ -24,13 +26,18 @@ repository = http://git.erp5.org/repos/slapos.toolbox.git
 branch = master
 git-executable = ${git:location}/bin/git
 
+[erp5.util-repository]
+recipe = slapos.recipe.build:gitclone
+repository = http://git.erp5.org/repos/erp5.git
+branch = scalability-master2
+git-executable = ${git:location}/bin/git
+
 [check-recipe]
 recipe = plone.recipe.command
 stop-on-error = true
 update-command = ${:command}
 command =
   grep parts ${buildout:develop-eggs-directory}/slapos.cookbook.egg-link &&
-  grep parts ${buildout:develop-eggs-directory}/slapos.toolbox.egg-link
+  grep parts ${buildout:develop-eggs-directory}/slapos.toolbox.egg-link &&
+  grep parts ${buildout:develop-eggs-directory}/erp5.util.egg-link
 
-[versions]
-websockify = 0.3.0
\ No newline at end of file

software/kvm/instance-kvm-input-schema.json

@@ -73,6 +73,12 @@
       "format": "uri",
       "type": "string",
     },
+    "virtual-hard-drive-md5sum": {
+      "title": "Checksum of virtual hard drive",
+      "description": "MD5 checksum of virtual hard drive, used if virtual-hard-drive-url is specified.",
+      "type": "string",
+    },
+virtual-hard-drive-md5sum
 
     "use-tap": {
       "title": "Use QEMU TAP network interface",

software/kvm/instance-kvm-resilient-test.cfg.in

+[buildout]
+
+eggs-directory = ${buildout:eggs-directory}
+develop-eggs-directory = ${buildout:develop-eggs-directory}
+offline = true
+
+parts =
+  deploy-resiliency-test
+  request-resilient-kvm
+
+[directory]
+recipe = slapos.cookbook:mkdirectory
+etc = $${buildout:directory}/etc/
+var = $${buildout:directory}/var/
+srv = $${buildout:directory}/srv/
+bin = $${buildout:directory}/bin/
+tmp = $${buildout:directory}/tmp/
+
+services = $${:etc}/service/
+scripts = $${:etc}/run/
+
+[deploy-resiliency-test]
+recipe = slapos.cookbook:wrapper
+#log-path = {directory:log}
+wrapper-path = $${directory:bin}/runKVMResiliencyTestSuite
+#output = $${binary-wrap-launcher:binary-path}
+#site-id = "{site_id}"
+# Suppose that there is a user zope family in the configuration
+#erp5-url = $${erp5-cluster:connection-family-user}{:site-id}
+#parameters-extra = $*
+command-line = ${buildout:bin-directory}/runKVMResiliencyTestSuite --server-url=$${slap-connection:server-url} --key-file=$${slap-connection:key-file} --cert-file=$${slap-connection:cert-file} --computer-id=$${slap-connection:computer-id} --partition-id=$${slap-connection:partition-id} --software=$${slap-connection:software-release-url} --namebase=kvm --kvm-rootinstance-name='$${request-resilient-kvm:name}'
+#return = url
+
+[request-resilient-kvm]
+<= slap-connection
+recipe = slapos.cookbook:request
+software-url = $${slap-connection:software-release-url}
+software-type = kvm-resilient
+name = Resilient KVM (Root Instance)
+config = virtual-hard-drive-url virtual-hard-drive-md5sum resiliency-backup-periodicity
+config-virtual-hard-drive-url = $${slap-parameter:virtual-hard-drive-url}
+config-virtual-hard-drive-md5sum = $${slap-parameter:virtual-hard-drive-md5sum}
+config-resiliency-backup-periodicity = */5
+# We don't use url parameter, but we want it to be there to make sure root instance is ready.
+return = url
+# XXX What to do?
+#sla = instance_guid
+#sla-instance_guid = $${slap-parameter:frontend-instance-guid}
+
+[slap-parameter]
+virtual-hard-drive-url = https://softinst43236.host.vifib.net/data/public/8e2138.php?dl=true
+virtual-hard-drive-md5sum = de0f10c7c6538e9928879332afd9be7a
+# XXX: what to do about pbs-kvm1-computer-guid and kvm1-computer-guid ?
+

software/kvm/instance-kvm-resilient.cfg.jinja2

@@ -10,10 +10,10 @@ offline = true
 
 # += because we need to take up parts (like instance-custom, slapmonitor etc) from the profile we extended
 parts +=
-  {{ parts.replicate("kvm", "2") }}
+  {{ parts.replicate("kvm", "3") }}
   publish-connection-informations
 
-{{ replicated.replicate("kvm", "2", "kvm-export", "kvm-import") }}
+{{ replicated.replicate("kvm", "3", "kvm-export", "kvm-import") }}
 
 # Bubble down the parameters of the requested instance to the user
 [request-kvm]
@@ -22,10 +22,10 @@ return =
 # Resilient related parameters
   url ssh-public-key ssh-url notification-id ip
 # KVM related parameters
-  backend-url url
+  backend-url url ipv6
 
 [publish-connection-informations]
 recipe = slapos.cookbook:publish
 backend-url = ${request-kvm:connection-backend-url}
 url = ${request-kvm:connection-url}
-
+ipv6 = ${request-kvm:connection-ipv6}

software/kvm/instance-kvm.cfg.in

@@ -88,7 +88,7 @@ nat-rules = $${slap-parameter:nat-rules}
 6tunnel-wrapper-path = $${directory:services}/6tunnel
 
 virtual-hard-drive-url = $${slap-parameter:virtual-hard-drive-url}
-virtual-hard-drive-md5-url = $${slap-parameter:virtual-hard-drive-md5-url}
+virtual-hard-drive-md5sum = $${slap-parameter:virtual-hard-drive-md5sum}
 
 shell-path = ${dash:location}/bin/dash
 qemu-path = ${kvm:location}/bin/qemu-system-x86_64
@@ -214,7 +214,7 @@ sla-instance_guid = $${slap-parameter:frontend-instance-guid}
 recipe = slapos.cookbook:publish
 backend-url = https://[$${novnc-instance:ip}]:$${novnc-instance:port}/vnc_auto.html?host=[$${novnc-instance:ip}]&port=$${novnc-instance:port}&encrypt=1&password=$${kvm-instance:vnc-passwd}
 url = $${request-slave-frontend:connection-url}/vnc_auto.html?host=$${request-slave-frontend:connection-domainname}&port=$${request-slave-frontend:connection-port}&encrypt=1&path=$${request-slave-frontend:connection-resource}&password=$${kvm-instance:vnc-passwd}
-
+ipv6 = $${slap-network-information:global-ipv6}
 
 [frontend-promise]
 recipe = slapos.cookbook:check_url_available
@@ -243,4 +243,4 @@ nat-rules = 22 80 443
 use-tap = False
 
 virtual-hard-drive-url =
-virtual-hard-drive-md5-url =
+virtual-hard-drive-md5sum =

software/kvm/instance.cfg.in

@@ -16,6 +16,9 @@ kvm-resilient = $${dynamic-template-kvm-resilient:rendered}
 kvm-import = ${template-kvm-import:output}
 kvm-export = ${template-kvm-export:output}
 
+# Used for the test of resiliency. The system wants a "test" software_type.
+test = ${template-kvm-resilient-test:output}
+
 frozen = ${instance-frozen:output}
 pull-backup = ${template-pull-backup:output}
 
@@ -29,8 +32,7 @@ cert = $${slap-connection:cert-file}
 
 [dynamic-template-kvm-resilient]
 recipe = slapos.recipe.template:jinja2
-template = ${:_profile_base_location_}/instance-kvm-resilient.cfg.jinja2
-md5sum = 1b7a2fcc884649b4d08f238e828899c1
+template = ${template-kvm-resilient:location}/instance-kvm-resilient.cfg.jinja2
 rendered = $${buildout:directory}/template-kvm-resilient.cfg
 context = key buildout buildout:bin-directory
           key develop_eggs_directory buildout:develop-eggs-directory

stack/resilient/buildout.cfg

@@ -48,7 +48,7 @@ mode = 0644
 recipe = slapos.recipe.template
 url = ${:_profile_base_location_}/pbsready-export.cfg.in
 output = ${buildout:directory}/pbsready-export.cfg
-md5sum = a0e22a5de727544c5767d6bee059a77a
+md5sum = ef3861861746d3574f39f1aa3200d74e
 mode = 0644
 
 [template-pull-backup]

stack/resilient/pbsready-export.cfg.in

@@ -39,5 +39,10 @@ notify = $${slap-parameter:notify}
 <= cron
 recipe = slapos.cookbook:cron.d
 name = backup
-frequency = 0 * * * *
+frequency = $${slap-parameter:resiliency-backup-periodicity} * * * *
 command = $${notifier-exporter:wrapper}
+
+[slap-parameter]
+# in minutes, modulo 60, in cron.d format (i.e */15 is accepted).
+resiliency-backup-periodicity = 0
+