Using Hooks¶

The Hook subsystem enables the execution of custom scripts tied to a change in state in a particular resource, or API call. This opens a wide area of automation for system administrators to tailor their cloud infrastructures. It also features a logging mechanism that allows a convenient way to query the execution history or to retry the execution of a given hook.

Overview¶

The hook subsystem has two main components:

Hook Manager Driver: it receives information about every event triggered by OpenNebula service and publishes it to an event queue. Custom components can also use this queue to subscribe to OpenNebula events, more information here.
Hook Execution Manager (HEM): It registers itself to the events that triggers the hooks defined in the system. When an event is received it takes care of executing the corresponding hook command.

hook-subsystem

Both components are started together with the OpenNebula service. Note that, provided the network communication is secure, you can grant network access to the event queue and hence deploy HEM in a different server.

Configuration¶

Hook Manager¶

Hook Manager configuration is set in HM_MAD section in /etc/one/oned.conf. The configuration attributes are described below:

Parameter

Description

Executable

Path of the hook driver executable, can be an absolute path or relative to $ONE_LOCATION/lib/mads (or /usr/lib/one/mads/ if OpenNebula was installed in /)

Arguments

Arguments for the driver executable, the following values are supported:

--publisher-port, -p: The port where the Hook Manager will publish the events reported by oned.
--logger-port, -l: The port where the Hook Manager will receive information about hook executions.
--hwm, -h: The HWM value for the publisher socket, more information can be found here.
--bind, -b: Address to bind the publisher socket.

Hook Execution Manager¶

Hook Execution Manager configuration is set in /etc/one/onehem-server.conf:

Parameter	Description
debug_level	Set the log debug level shown in `/var/log/one/onehem.log`
hook_base_path	Base location to look for hook scripts when commands use a relative path (default value `/var/lib/one/remotes/hooks`)
remote_hook_base_path	Base location to look for hook scripts when commands use a relative path and `REMOTE="yes"` is specified (default value `'/var/tmp/one/hooks'`)
subscriber_endpoint	To subscribe for OpenNebula events, must match those in `HM_MAD` section of `oned.conf`.
replier_endpoint	To send hook execution results (reply to events) to oned, it must match those in `HM_MAD` section of `oned.conf`.
concurrency	Number of hooks executed simultaneously.

Hook Log¶

You can check the execution results of a hook using the hook log. The hook log stores a record for each execution including the standard output and error of the command, as well as the arguments used. This information is used to retry the execution of a given record.

The number of execution log records stored in the database for each hook can be configured in oned.conf. For example, to keep only the last 10 execution records of each hook use LOG_RETENTION = 10. This value can be tuned taking into account the number of hooks and how many times hooks are executed:

HOOK_LOG_CONF = [
    LOG_RETENTION = 10 ]

Hook Types¶

There are two types of hooks, API hooks (triggered on API calls) and State hooks (triggered on state change). The main attributes for both types are described in the table below:

Option	Mandatory \| Description
NAME	YES	The name of the hook.
COMMAND	YES	The command that will be executed when hook is triggered. Typically a path to a script.
ARGUMENTS	NO	The arguments that will be passed to the script when the hook is triggered.
ARGUMENTS_STDIN	NO	If `yes` the ARGUMENTS will be passed through STDIN instead of as command line arguments.
TYPE	YES	The hook type either `api` or `state`.
TIMEOUT	NO	Hook execution timeout. If not used the timeout is infinite.

API Hooks¶

The API hooks are triggered after the execution of an API call. The specific attributes for API hooks are listed below:

Option	Mandatory	Description
CALL	YES	The API call which trigger the hook. For more info about API calls please check XML-RPC API section.

For API hooks, the $API keyword can be used in the ARGUMENTS attribute to get the information about the API call. If $API is used, a base64-encoded XML document containing the API call arguments and result will be passed to the hook command. The schema of the XML is defined here.

Note

If the API method defined in CALL is an allocate or delete, the $API document will also include the template of the corresponding resource.

The following example defines an API hook that executes the command /var/lib/one/remotes/hooks/log_new_user.rb whenever a new user is created:

NAME      = hook-API
TYPE      = api
COMMAND   = "log_new_user.rb"
ARGUMENTS = $API
CALL      = "one.user.allocate"
ARGUMENTS_STDIN = yes

State Hooks¶

The state hooks are only available for Hosts and Virtual Machines and they are triggered on specific state transitions. The specific attributes to define state hooks are:

Option	Mandatory	Description
RESOURCE	YES	Type of the resource, supported values are `IMAGE`, `HOST` and `VM`.
REMOTE	NO	If `yes` the hook will be executed in the host that triggered the hook (for Host hooks) or in the host where the VM is running (for VM hooks). Not used for Image hooks
STATE	YES	The state that triggers the hook.
LCM_STATE	YES	The LCM state that triggers the hook (Only for VM hooks)
ON	YES	For `RESOURCE=VM`, shortcut to define common `STATE`/`LCM_STATE` pairs. Supported values are: PROLOG, RUNNING, SHUTDOWN, STOP, DONE, UNKNOWN, CUSTOM

Warning

Note that ON is mandatory for VM hooks, use ON=CUSTOM with STATE and LCM_STATE to define hooks on specific state transitions.

For state hooks, $TEMPLATE can be used in the ARGUMENTS attribute to get the template (in XML format, base64 encoded) of the resource which triggered the hook. The XSD schema files for the XML document of each object are available here

The following examples define two state hooks for VMs, hosts and images:

# VM
NAME = hook-vm
TYPE = state
COMMAND = new_vm.rb
ARGUMENTS = $TEMPLATE
ON = PROLOG
RESOURCE = VM

# HOST
NAME = hook-host
TYPE = state
COMMAND = host-disabled.rb
STATE = DISABLED
RESOURCE = HOST
REMOTE = yes

# IMAGE
NAME = hook-image
TYPE = state
COMMAND = image-ready.rb
STATE = READY
RESOURCE = IMAGE

Note

Check each resource guide for info about Image states, VM states and Host states

Managing Hooks¶

Hooks can be managed via the CLI through the onehook command and the API. This section describes the common operations to control the life-cycle of a hook using the CLI.

Creating Hooks¶

In order to create a new hook you need to create a hook template:

$ cat > hook.tmpl << EOF
     NAME      = hook-vm
     TYPE      = state
     COMMAND   = vm-pending.rb
     ARGUMENTS = "\$TEMPLATE pending"
     ON        = CUSTOM
     RESOURCE  = VM
     STATE     = PENDING
     LCM_STATE = LCM_INIT
 EOF

Then, simply create the hook by running the following command:

$ onehook create hook.tmpl
  ID: 0

We have just created a hook which will be triggered each time a VM switch to PENDING state.

Note

Note that just one hook can be created for each trigger event.

Checking Hook Execution¶

We can check the execution records of a hook by accessing its detailed information. For example, to get the execution history of the previous hook use onehook show 0:

$ onevm create --cpu 1 --memory 2 --name test
  ID: 0
$ onehook show 0
  HOOK 0 INFORMATION
  ID                : 0
  NAME              : hook-vm
  TYPE              : state
  LOCK              : None

  HOOK TEMPLATE
  ARGUMENTS="$TEMPLATE pending"
  COMMAND="vm-pending.rb"
  LCM_STATE="LCM_INIT"
  REMOTE="NO"
  RESOURCE="VM"
  STATE="PENDING"

  EXECUTION LOG
    ID    TIMESTAMP    EXECUTION
    0     09/23 15:10  ERROR (255)

We can see that there is an execution which have failed with error code 255. To get more information about a specific execution use the -e option:

$ onehook show 0 -e 0
  HOOK 0 INFORMATION
  ID                : 0
  NAME              : hook-vm
  TYPE              : state
  LOCK              : None

  HOOK EXECUTION RECORD
  EXECUTION ID      : 0
  TIMESTAMP         : 09/23 15:10:38
  COMMAND           : /var/lib/one/remotes/hooks/vm-pending.rb PFZNPgogIDxJR...8+CjwvVk0+ pending
  ARGUMENTS         :
  <VM>
  <ID>0</ID>
  <UID>0</UID>
  <GID>0</GID>
  <UNAME>oneadmin</UNAME>
  <GNAME>oneadmin</GNAME>
  <NAME>test</NAME>
  <PERMISSIONS>
      <OWNER_U>1</OWNER_U>
      <OWNER_M>1</OWNER_M>
      <OWNER_A>0</OWNER_A>
      <GROUP_U>0</GROUP_U>
      <GROUP_M>0</GROUP_M>
      <GROUP_A>0</GROUP_A>
      <OTHER_U>0</OTHER_U>
      <OTHER_M>0</OTHER_M>
      <OTHER_A>0</OTHER_A>
  </PERMISSIONS>
  <LAST_POLL>0</LAST_POLL>
  <STATE>1</STATE>
  <LCM_STATE>0</LCM_STATE>
  <PREV_STATE>1</PREV_STATE>
  <PREV_LCM_STATE>0</PREV_LCM_STATE>
  <RESCHED>0</RESCHED>
  <STIME>1569244238</STIME>
  <ETIME>0</ETIME>
  <DEPLOY_ID/>
  <MONITORING/>
  <TEMPLATE>
      <AUTOMATIC_REQUIREMENTS><![CDATA[!(PUBLIC_CLOUD = YES) & !(PIN_POLICY = PINNED)]]></AUTOMATIC_REQUIREMENTS>
      <CPU><![CDATA[1]]></CPU>
      <MEMORY><![CDATA[2]]></MEMORY>
      <VMID><![CDATA[0]]></VMID>
  </TEMPLATE>
  <USER_TEMPLATE/>
  <HISTORY_RECORDS/>
  </VM> pending
  EXIT CODE         : 255

  EXECUTION STDOUT


  EXECUTION STDERR
  ERROR MESSAGE --8<------
  Internal error No such file or directory - /var/lib/one/remotes/hooks/vm-pending.rb
  ERROR MESSAGE ------>8--

The EXECUTION STDERR message shows the reason for the hook execution failure, the script does not exists:

Internal error No such file or directory - /var/lib/one/remotes/hooks/vm-pending.rb

Important

The hook log can be queried and filtered by several criteria using onehook log. More info about onehook log command can be found running onehook log --help.

Retrying Hook Executions¶

We are going to fix the previous error, let’s first create the vm-pending.rb script, and then retry the hook execution.

Note

Note that any hook execution can be retried regardless of it result.

$ vim /var/lib/one/remotes/hooks/vm-pending.rb
  #!/usr/bin/ruby
  puts "Executed!"

$ chmod 760 /var/lib/one/remotes/hooks/vm-pending.rb
$ onehook retry 0 0
$ onehook show 0
  HOOK 0 INFORMATION
  ID                : 0
  NAME              : hook-vm
  TYPE              : state
  LOCK              : None

  HOOK TEMPLATE
  ARGUMENTS="$TEMPLATE pending"
  COMMAND="vm-pending.rb"
  LCM_STATE="LCM_INIT"
  REMOTE="NO"
  RESOURCE="VM"
  STATE="PENDING"

  EXECUTION LOG
  ID       TIMESTAMP    RC    EXECUTION
  0        09/23 15:10  255   ERROR
  1        09/23 15:59    0   SUCCESS

Note the last successful execution record!

Important

When a hook execution is retried the same execution context is used, i.e. arguments and $TEMPLATE/$API values.

Developing Hooks¶

First thing you need to decide is the type of hook you are interested in, being it API or STATE hooks. Each type of hook is triggered by a different event and requires/provides different runtime information.

In this section you’ll find two simple script hooks (in ruby) for each type. This examples are good starting points for developing custom hooks.

Warning

Note that the scripts below import some dependencies that need to be installed properly in the front-end node. As most of the dependencies are OpenNebula dependencies these dependencies can be used by adding the snippet below to your scripts.

ONE_LOCATION = ENV['ONE_LOCATION']

if !ONE_LOCATION
    RUBY_LIB_LOCATION = '/usr/lib/one/ruby'
    GEMS_LOCATION     = '/usr/share/one/gems'
else
    RUBY_LIB_LOCATION = ONE_LOCATION + '/lib/ruby'
    GEMS_LOCATION     = ONE_LOCATION + '/share/gems'
end

if File.directory?(GEMS_LOCATION)
    Gem.use_paths(GEMS_LOCATION)
end

$LOAD_PATH << RUBY_LIB_LOCATION

API Hook example¶

This script prints to stdout the result of one.user.create API call and the username of new user.

# Hook template
#
# NAME = user-create
# TYPE = api
# COMMAND = "user_create.rb"
# ARGUMENTS = "$API"
# CALL = "one.user.allocate"

#!/usr/bin/ruby

require 'base64'
require 'nokogiri'

#api_info= Nokogiri::XML(Base64::decode64(STDIN.gets.chomp)) for reading from STDIN
api_info = Nokogiri::XML(Base64::decode64(ARGV[0]))

success = api_info.xpath("/CALL_INFO/RESULT").text.to_i == 1
uname   = api_info.xpath('//PARAMETER[TYPE="IN" and POSITION=2]/VALUE').text

if !success
    puts "Failing to create user"
    exit -1
end

puts "User #{uname} successfully created"

State Hook example (HOST)¶

This script prints to stdout when an host is in error state.

# Hook template
#
#NAME = host-error
#TYPE = state
#COMMAND = host_error.rb
#ARGUMENTS="$TEMPLATE"
#STATE = ERROR
#RESOURCE = HOST

#!/usr/bin/ruby

require 'base64'
require 'nokogiri'

#host_template = Nokogiri::XML(Base64::decode64(STDIN.gets.chomp)) for reading from STDIN
host_template = Nokogiri::XML(Base64::decode64(ARGV[0]))

host_id = host_template.xpath("//ID").text.to_i

puts "Host #{host_id} is in error state!!"

State Hook example (VM)¶

This script prints to stdout if when a VM is in prolog state.

# Hook template
#
#NAME = vm-prolog
#TYPE = state
#COMMAND = vm_prolog.rb
#ARGUMENTS = $TEMPLATE
#ON = PROLOG
#RESOURCE = VM

#!/usr/bin/ruby

require 'base64'
require 'nokogiri'

#vm_template = Nokogiri::XML(Base64::decode64(STDIN.gets.chomp)) for reading from STDIN
vm_template = Nokogiri::XML(Base64::decode64(ARGV[0]))

vm_id = vm_template.xpath("//ID").text.to_i

puts "VM #{vm_id} is in PROLOG state"

State Hook example (IMAGE)¶

This script prints to stdout the ID of image when it is ready to use.

# Hook template
#
# NAME = image-ready
# TYPE = state
# COMMAND = image_ready.rb
# STATE = READY
# RESOURCE = IMAGE
# ARGUMENTS = "$TEMPLATE"

#!/usr/bin/ruby

require 'base64'
require 'nokogiri'

#img_template = Nokogiri::XML(Base64::decode64(STDIN.gets.chomp)) for reading from STDIN
img_template = Nokogiri::XML(Base64::decode64(ARGV[0]))

img_id = img_template.xpath("//ID").text.to_i

puts "Image #{img_id} ready to use!!"

A Complete Example: Autostart Hooks for KVM¶

OpenNebula creates transient KVM domains, i.e. they only exist while the domain is running. Therefore, the KVM autostart feature cannot be activated via libvirt configuration. This example shows how to implement VM autostart for the KVM hypervisor in OpenNebula using state hooks.

The hooks track Hosts reboots and resume VMs allocated to the Host that includes the AUTOSTART attribute in their template. If AUTOSTART=yes the VM will be automatically restarted if it was running before the host was rebooted, if AUTOSTART=always the VM will be automatically restarted always regardless its previous state. This functionality is implemented by two state hooks: one for the HOST and one for the VM resource. The code for both hooks can be found in /var/lib/one/remotes/hooks/autostart/ folder.

To install the hooks create a definition file for each one. The autostart-host.tmpl definition file for the Host hook will be as follows:

NAME = autostart-host
TYPE = state
COMMAND = autostart/host
ARGUMENTS = \$TEMPLATE
ARGUMENTS_STDIN = yes
RESOURCE = HOST
STATE = MONITORED

Similarly for the VM hook, the autostart-vm.tmpl definition file will be as follows:

NAME = autostart-vm
TYPE = state
COMMAND = autostart/vm
ARGUMENTS = \$TEMPLATE
ARGUMENTS_STDIN = yes
RESOURCE = VM
STATE = POWEROFF
LCM_STATE = LCM_INIT
ON = CUSTOM

Now you can create the hooks using these two files:

$ onehook create autostart-host.tmpl
$ onehook create autostart-vm.tmpl

Note

Note that any command can be specified in COMMAND, for debugging (COMMAND="/usr/bin/echo") can be very helpful.

Hook Manager Events¶

The Hook Manager publish the OpenNebula events over a ZeroMQ publisher socket. This can be used for developing custom components that subscribe to specific events to perform custom actions.

Hook Manager Messages¶

The hook manager sends two different types of messages:

EVENT messages: represent an OpenNebula event, which potentially could trigger a hook if the Hook Execution Manager is subscribed to it.

RETRY messages: represent a retry action of a given hook execution. These event messages are specific to the Hook Execution Manager.

Both messages have a key that can be used by the subscriber to receive messages related to an specific event class; and a content that contains the information related to the event encoded in base64.

Event Messages Format¶

There are two different types of EVENT messages, representing API and state events. The key format for both types are listed below:

EVENT	Key format
API	`EVENT API <API_CALL> <SUCCESS>` (e.g. EVENT API one.vm.allocate 1` or `Key: EVENT API one.hook.info 0)
STATE (HOST)	`EVENT STATE HOST/<STATE>/` (e.g. `EVENT STATE HOST/INIT/`) `EVENT HOST <HOST_ID>/<STATE>/` (e.g. `EVENT HOST 0/INIT/`)
STATE (VM)	`EVENT STATE VM/<STATE>/<LCM_STATE>` (e.g. `EVENT STATE VM/PENDING/LCM_INIT`) `EVENT VM <VM_ID>/<STATE>/<LCM_STATE>` (e.g. `EVENT VM 0/PENDING/LCM_INIT`)
CHANGE (SERVICE)	`EVENT SERVICE SERVICE_ID` (e.g. `EVENT SERVICE 0`)

Keys are used to subscribe to specific events. Note also that you do not need to specify the whole key, form example EVENT STATE HOST/ERROR/ will subscribe for state changes to ERROR on the hosts, while EVENT STATE HOST/ will subscribe for all state changes of the hosts.

The content of an event message is an XML document containing information related to the event, the XSD representing this content are available here.

Retry Messages Format¶

The key format of the retry messages is just the word RETRY, no more info is needed in the key. The retry messages content is an XML file with information about the execution to retry. The XSD representing this content is available here.

Subscriber script example¶

Below there is an example of script which retrieve all the event messages and prints their key and their content.

#!/usr/bin/ruby

require 'ffi-rzmq'
require 'base64'

@context    = ZMQ::Context.new(1)
@subscriber = @context.socket(ZMQ::SUB)

@subscriber.setsockopt(ZMQ::SUBSCRIBE, "EVENT")
@subscriber.connect("tcp://localhost:2101")

key = ''
content = ''

loop do
    @subscriber.recv_string(key)
    @subscriber.recv_string(content)

    puts "Key: #{key}\nContent: #{Base64.decode64(content)}"
    puts "\n===================================================================\n"
end

Note

Note that the key and the content as read separately as ZeroMQ envelopes are used.