The public API for TripleO uses the OpenStack Workflow service, Mistral to provide its interface. This allows external systems to consume and use the same Workflows used by python-tripleoclient and tripleo-ui.
TripleO functionality can be accessed via Mistral Workflows and Actions. Workflows define a set number of steps and typically use a number of actions. There is a set of actions which are intended to be used directly. When actions are called directly, Mistral executes them synchronously which is quicker for simple actions.
Mistral can be used with the CLI interface or Python bindings which are both provided by python-mistralclient or via the REST API directly. This guide will use the Mistral CLI interface for brevity.
When using the CLI, all of the TripleO workflows can be viewed with the
command openstack workflow list
. All of the workflows provided by TripleO
will have a name starting with tripleo.
$ openstack workflow list
+--------------------------------------+-----------------------------------------+------------------------------+
| ID | Name | Input |
+--------------------------------------+-----------------------------------------+------------------------------+
| 1ae040b6-d330-4181-acb9-8638dc486b79 | tripleo.baremetal.v1.set_node_state | node_uuid, state_action, ... |
| 2ef20a58-b380-4b6b-a6cd-270352d0f3d2 | tripleo.deployment.v1.deploy_on_servers | server_name, config_name,... |
+--------------------------------------+-----------------------------------------+------------------------------+
To view the individual workflows in more detail and see the inputs they
accept use the openstack workflow show
command. This command will also
show the default values for input parameters. If no default is given, then it
is required.
$ openstack workflow show tripleo.plan_management.v1.create_default_deployment_plan
+------------+-----------------------------------------------------------+
| Field | Value |
+------------+-----------------------------------------------------------+
| ID | fa8256ec-b585-476f-a83e-e800beb26684 |
| Name | tripleo.plan_management.v1.create_default_deployment_plan |
| Project ID | 65c5259b7a96436f898fd518815c42c1 |
| Tags | <none> |
| Input | container, queue_name=tripleo |
| Created at | 2016-08-19 10:07:10 |
| Updated at | None |
+------------+-----------------------------------------------------------+
This workflow can then be executed with the openstack workflow execution
create
command.
$ openstack workflow execution create tripleo.plan_management.v1.create_default_deployment_plan \
'{"container": "my_cloud"}'
+-------------------+-----------------------------------------------------------+
| Field | Value |
+-------------------+-----------------------------------------------------------+
| ID | 824a8cf7-3306-4ef2-8efd-a2715dd0dbce |
| Workflow ID | fa8256ec-b585-476f-a83e-e800beb26684 |
| Workflow name | tripleo.plan_management.v1.create_default_deployment_plan |
| Description | |
| Task Execution ID | <none> |
| State | RUNNING |
| State info | None |
| Created at | 2016-08-22 12:33:35.493135 |
| Updated at | 2016-08-22 12:33:35.495764 |
+-------------------+-----------------------------------------------------------+
After a Workflow execution is created it will be scheduled up by Mistral and
executed asynchronously. Mistral can either be polled until it is finished or
you can subscribe to the Zaqar queue for messages from the running
Workflow. By default the TripleO workflows will send messages to a Zaqar queue
with the name tripleo
, the workflows all accept a queue_name
parameter
which allows a user defined queue name to be used. It can be useful to use
different queue names if you plan to execute multiple workflows and want the
messages to be handled individually.
Actions can be used in a similar way to workflows, but the CLI commands are
openstack action definition list
, openstack action definition show
and openstack action execution run
.
API reference documentation is available for all TripleO Workflows.
Deployment plans consist of a Swift container and a Mistral Environment. The TripleO Heat Templates are stored in Swift and then user defined parameters are stored in the Mistral environment.
When the undercloud is installed, it will create a default plan with the name
overcloud
. To create a new plan from the packaged version of
tripleo-heat-templates on the undercloud use the workflow
tripleo.plan_management.v1.create_default_deployment_plan
. This workflow
accepts a name which will be used for the Swift container and Mistral
environment.
The following command creates a plan called my_cloud
.
$ openstack workflow execution create tripleo.plan_management.v1.create_default_deployment_plan \
'{"container": "my_cloud"}'
+-------------------+-----------------------------------------------------------+
| Field | Value |
+-------------------+-----------------------------------------------------------+
| ID | dc4800ef-8d0a-436e-9564-a7ee81ba93d5 |
| Workflow ID | fa8256ec-b585-476f-a83e-e800beb26684 |
| Workflow name | tripleo.plan_management.v1.create_default_deployment_plan |
| Description | |
| Task Execution ID | <none> |
| State | RUNNING |
| State info | None |
| Created at | 2016-08-23 10:06:45.372767 |
| Updated at | 2016-08-23 10:06:45.376122 |
+-------------------+-----------------------------------------------------------+
Note
When updating the packages on the undercloud with yum the TripleO Heat Templates will be updated in /usr/share/.. but any plans that were previously created will not be updated automatically. At the moment this is a manual process.
Manually creating a plan with custom templates is a three stage process. Each
step must use the same name for the container, we are using my_cloud
, but
it can be changed if they are all consistent. This will be the plan name.
Create the Swift container.
openstack action execution run tripleo.plan.create_container \
'{"container":"my_cloud"}'
Note
Creating a swift container directly isn’t sufficient, as this Mistral action also sets metadata on the container and may include further steps in the future.
Upload the files to Swift.
swift upload my_cloud path/to/tripleo/templates
Trigger the plan create Workflow, which will create the Mistral environment for the uploaded templates, do some initial template processing and generate the passwords.
openstack workflow execution create tripleo.plan_management.v1.create_deployment_plan \
'{"container":"my_cloud"}'
Some functionality for dealing with bare metal nodes is provided by the
tripleo.baremetal
workflows.
Baremetal nodes can be registered with Ironic via Mistral. The input for this workflow is a bit larger, so this time we will store it in a file and pass it in, rather than working inline.
$ cat nodes.json
{
"remove": false,
"ramdisk_name": "bm-deploy-ramdisk",
"kernel_name": "bm-deploy-kernel",
"instance_boot_option": "local",
"nodes_json": [
{
"pm_password": "$RSA_PRIVATE_KEY",
"pm_type": "pxe_ssh",
"pm_addr": "192.168.122.1",
"mac": [
"00:8f:61:0d:6a:e1"
],
"memory": "8192",
"disk": "40",
"arch": "x86_64",
"cpu": "4",
"pm_user": "root"
}
]
}
remove
is set to true, any nodes that are not passed to the workflow
will be removed.ramdisk_name
and kernel_name
are the Glance names for the kernel and
ramdisk to use for the nodes.instance_boot_option
defines whether to set instances for booting from
the local hard drive (local) or network (netboot).$ openstack workflow execution create tripleo.baremetal.v1.register_or_update \
nodes.json
The result of this workflow can be seen with the following command.
$ mistral execution-get-output $EXECUTION_ID
{
"status": "SUCCESS",
"new_nodes": [],
"message": "Nodes set to managed.",
"__task_execution": {
"id": "001892c5-4197-4c04-af74-aff95f6d584f",
"name": "send_message"
},
"registered_nodes": [
{
"uuid": "93feecfb-8a4d-418c-9f2c-5ef8db7aff2e",
...
},
]
}
The above information is accessible like this, or via the zaqar queue. The registered_nodes property will contain each of the nodes registered with all their properties from Ironic, including the UUID which is useful for introspection.
To introspect the nodes, we need to either use the Ironic UUID’s returned by the register_or_update workflow or retrieve them directly from Ironic. Then those UUID’s can be passed to the introspection workflow. The workflow expects nodes to be in the “manageable” state.
$ openstack workflow execution create tripleo.baremetal.v1.introspect \
'{"nodes_uuids": ["UUID1", "UUID2"]}'
It is recommended to clean previous information from all disks on the bare
metal nodes before new deployments. As TripleO disables automated cleaning, it
has to be done manually via the manual_clean
workflow. A node has to be in
the manageable
state for it to work.
Note
See Ironic cleaning documentation for more details.
To remove partitions from all disks on a given node, use the following command:
$ openstack workflow execution create tripleo.baremetal.v1.manual_cleaning \
'{"node_uuid": "UUID", "clean_steps": [{"step": "erase_devices_metadata", "interface": "deploy"}]}'
To remove all data from all disks (either by ATA secure erase or by shredding them), use the following command:
$ openstack workflow execution create tripleo.baremetal.v1.manual_cleaning \
'{"node_uuid": "UUID", "clean_steps": [{"step": "erase_devices", "interface": "deploy"}]}'
The node state is set back to manageable
after successful cleaning and to
clean failed
after a failure. Inspect node’s last_error
field for the
cause of the failure.
Warning
Shredding disks can take really long, up to several hours.
After the nodes have been introspected they will still be in the manageable state. To make them available for a deployment, use the provide workflow, which has the same interface as introspection.
$ openstack workflow execution create tripleo.baremetal.v1.provide \
'{"nodes_uuids": ["UUID1", "UUID2"]}'
A number of parameters will need to be provided for a deployment to be
successful. These required parameters will depend on the Heat templates that
are being used. Parameters can be set with the Mistral Action
tripleo.parameters.update
.
Note
This action will merge the passed parameters with those already set on the
plan. To set the parameters first use tripleo.parameters.reset
to
remove any old parameters first.
In the following example we set the ComputeCount
parameter to 2
on the
my_cloud
plan. This only sets one parameter, but any number can be provided.
$ openstack action execution run tripleo.parameters.update \
'{"container":"my_cloud", "parameters":{"ComputeCount":2}}'
After the plan has been configured it should be ready to be deployed.
$ openstack workflow execution create tripleo.deployment.v1.deploy_plan \
'{"container": "my_cloud"}'
Once the deployment is triggered, the templates will be processed and sent to Heat. This workflow will complete when the Heat action has started, or if there are any errors.
Deployment progress can be tracked via the Heat API. It is possible to either follow the Heat events or simply wait for the Heat stack status to change.
Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.