Mistral Main Features

Task result / Data flow

Mistral supports transferring data from one task to another. In other words, if taskA produces a value then taskB which follows taskA can use it. In order to use this data Mistral relies on a query language called YAQL. YAQL is a powerful yet simple tool that allows the user to filter information, transform data and call functions. Find more information about it in the YAQL official documentation . This mechanism for transferring data plays a central role in the workflow concept and is referred to as Data Flow.

Below is a simple example of how Mistral Data Flow looks like from the Mistral Workflow Language perspective:

version: '2.0'

my_workflow:
  input:
    - host
    - username
    - password

  tasks:
    task1:
      action: std.ssh host=<% $.host %> username=<% $.username %> \
              password=<% $.password %>
      input:
        cmd: "cd ~ && ls"
      on-complete: task2

    task2:
      action: do_something data=<% task(task1).result %>

The task called “task1” produces a result that contains a list of the files in a user’s home folder on a host(both username and host are provided as workflow input) and the task “task2” uses this data using the YAQLexpression “task(task1).result”. “task()” here is a function registered in YAQL by Mistral to get information about a task by its name.

Task affinity

Task affinity is a feature which could be useful for executing particular tasks on specific Mistral executors. In fact, there are 2 cases:

  1. You need to execute the task on a single executor.

  2. You need to execute the task on any executor within a named group.

To enable the task affinity feature, edit the “host” property in the “executor” section of the configuration file:

[executor]
host = my_favorite_executor

Then start (restart) the executor. Use the “target” task property to specify this executor in Mistral Workflow Language:

... Workflow YAML ...
task1:
  ...
  target: my_favorite_executor
... Workflow YAML ...

Task policies

Any Mistral task regardless of its workflow type can optionally have configured policies. Policies control the flow of the task - for example, a policy can delay task execution before the task starts or after the task completes.

YAML example

my_task:
  action: my_action
  pause-before: true
  wait-before: 2
  wait-after: 4
  fail-on: <% $.some_value < 4 %>
  timeout: 30
  retry:
    count: 10
    delay: 20
    break-on: <% $.my_var = true %>

There are different types of policies in Mistral.

  1. pause-before

Specifies whether Mistral Engine should put the workflow on pause or not before starting a task.

  1. wait-before

Specifies a delay in seconds that Mistral Engine should wait before starting a task.

  1. wait-after

Specifies a delay in seconds that Mistral Engine should wait after a task has completed before starting the tasks specified in ‘on-success’, ‘on-error’ or ‘on-complete’.

  1. fail-on

Specifies a condition under which the task will fail, even if the action was completed successfully.

  1. timeout

Specifies a period of time in seconds after which a task will be failed automatically by the engine if it hasn’t completed.

  1. retry

Specifies a pattern for how the task should be repeated.

  • count - Specifies a maximum number of times that a task can be repeated.

  • delay - Specifies a delay in seconds between subsequent task iterations.

  • break-on - Specifies a YAQL expression that will break the iteration loop if it evaluates to ‘true’. If it fires then the task is considered to have experienced an error.

  • continue-on - Specifies a YAQL expression that will continue the iteration loop if it evaluates to ‘true’. If it fires then the task is considered successful.

A retry policy can also be configured on a single line, as follows

task1:
  action: my_action
  retry: count=10 delay=5 break-on=<% $.foo = 'bar' %>

All parameter values for any policy can be defined as YAQL expressions.

NOTE: It would be rare to use both break-on and continue-on in the same retry block. break-on should be used when one expects the action to be in an ERROR state for some amount of tries, but may eventually go to a SUCCESS state, thereby stopping the loop. But if break-on is ‘true’ then the retries will stop and the task will be in ERROR. continue-on should be used if the action will usually return SUCCESS, but the action has other results that can be used to signal whether to continue the loop or not.

Join

Join flow control allows to synchronize multiple parallel workflow branches and aggregate their data.

Full join (join: all).

YAML example

register_vm_in_load_balancer:
  ...
  on-success:
    - wait_for_all_registrations

register_vm_in_dns:
  ...
  on-success:
    - wait_for_all_registrations

try_to_do_something_without_registration:
  ...
  on-error:
    - wait_for_all_registrations

wait_for_all_registrations:
  join: all
  action: send_email

When a task has property “join” assigned with value “all” the task will run only if all upstream tasks (ones that lead to this task) are completed and corresponding conditions have triggered. Task A is considered an upstream task of Task B if Task A has Task B mentioned in any of its “on-success”, “on-error” and “on-complete” clauses regardless of YAQL guard expressions.

Partial join (join: 2)

YAML example

register_vm_in_load_balancer:
  ...
  on-success:
    - wait_for_all_registrations

register_vm_in_dns:
  ...
  on-success:
    - wait_for_all_registrations

register_vm_in_zabbix:
  ...
  on-success:
    - wait_for_all_registrations

wait_for_two_registrations:
  join: 2
  action: send_email

When a task has a numeric value assigned to the property “join”, then the task will run once at least this number of upstream tasks are completed and the corresponding conditions have triggered. In the example above, the task “wait_for_two_registrations” will run if two any of the “register_vm_xxx” tasks are complete.

Discriminator (join: one)

Discriminator is the special case of Partial Join where the “join” property has the value 1. In this case instead of 1 it is possible to specify the special string value “one” which is introduced for symmetry with “all”. However, it’s up to the user whether to use “1” or “one”.

Processing collections (with-items)

YAML example

---
version: '2.0'

create_vms:
  description: Creating multiple virtual servers using "with-items".
  input:
    - vm_names
    - image_ref
    - flavor_ref
  output:
    vm_ids: <% $.vm_ids %>

  tasks:
    create_servers:
      with-items: vm_name in <% $.vm_names %>
      action: nova.servers_create name=<% $.vm_name %> \
              image=<% $.image_ref %> flavor=<% $.flavor_ref %>
      publish:
        vm_ids: <% task().result.id %>
      on-success:
        - wait_for_servers

    wait_for_servers:
      with-items: vm_id in <% $.vm_ids %>
      action: nova.servers_find id=<% $.vm_id %> status='ACTIVE'
      retry:
        delay: 5
        count: <% $.vm_names.len() * 10 %>

The workflow “create_vms” in this example creates as many virtual servers as we provide in the “vm_names” input parameter. E.g., if we specify vm_names=[“vm1”, “vm2”] then it’ll create servers with these names based on the same image and flavor. This is possible because we are using the “with-items” keyword that associates an action or a workflow with a task run multiple times. The value of the “with-items” task property contains an expression in the form: <variable_name> in <% YAQL_expression %>.

The most common form is

with-items:
  - var1 in <% YAQL_expression_1 %>
  - var2 in <% YAQL_expression_2 %>
  ...
  - varN in <% YAQL_expression_N %>

where collections expressed as YAQL_expression_1, YAQL_expression_2, YAQL_expression_N must have equal sizes. When a task gets started Mistral will iterate over all collections in parallel, i.e. the number of iterations will be equal to the length of any of the collections.

Note that in the “with-items” case, the task result (accessible in workflow context as <% $.task_name %>) will be a list containing results of corresponding action/workflow calls. If at least one action/workflow call has failed then the whole task will get into ERROR state. It’s also possible to apply retry policy for tasks with a “with-items” property. In this case the retry policy will relaunch all action/workflow calls according to the “with-items” configuration. Other policies can also be used in the same way as with regular non-“with-items” tasks.

Execution expiration policy

When Mistral is used in production it can be difficult to control the number of completed workflow executions. By default Mistral will store all executions indefinitely and over time the number stored will accumulate. This can be resolved by setting an expiration policy.

By default this feature is disabled.

This policy defines the maximum age of an execution since the last updated time (in minutes) and the maximum number of finished executions. Each evaluation will satisfy these conditions, so the expired executions (older than specified) will be deleted, and the number of execution in finished state (regardless of expiration) will be limited to max_finished_executions.

To enable the policy, edit the Mistral configuration file and specify evaluation_interval and at least one of the older_than or evaluation_interval options.

[execution_expiration_policy]
evaluation_interval = 120  # 2 hours
older_than = 10080  # 1 week
max_finished_executions = 500
  • evaluation_interval

The evaluation interval defines how frequently Mistral will check and ensure the above mentioned constraints. In the above example it is set to two hours, so every two hours Mistral will remove executions older than 1 week, and keep only the 500 latest finished executions.

  • older_than

Defines the maximum age of an execution in minutes since it was last updated. It must be greater or equal to 1.

  • max_finished_executions

Defines the maximum number of finished executions. It must be greater or equal to 1.

Workflow namespaces

Mistral allows creating workflows within a namespace. So it is possible to create many workflows with the same name as long as they are in different namespaces.

See more at Workflow namespaces

Task skip

Mistral has an ability to skip tasks in ERROR state. The task moves from ERROR state to SKIPPED state, publish variables from publish-on-skip section, and the workflow continues from tasks specified in on-skip section. To configure task’s behavior on skip, fill the following attributes in the task definition:

  • on-skip - Optional. This parameter specifies which tasks should be started after skipping this task.

  • publish-on-skip - Optional. This parameter specifies which variables should be published after skipping this task.

It is also possible to skip task which does not have predefined parameters described above, in this case task will not publish anything and will continue by on-success branch. It could be not safe for next tasks, because they probably would not have some inputs, so think twice before skipping such tasks.

Task skip could be performed by following request:

PUT /v2/tasks

{
  "id": "<task-id>",
  "state": "SKIPPED"
}