Introduction

• The Agent Cluster brings fail-over capabilities and horizontal scalability to Agents.
• FEATURE AVAILABILITY STARTING FROM RELEASE 2.2.0

Architecture

Operational View

The architecture applies to the clustering of Agent Directors and the clustering of Subagents for high-availability and scalability.

• The components involved in an Agent Cluster are represented as one Agent to the Controller.
• A Controller can manage any number of Agents either from different Agent Clusters or from a number of Standalone Agents.

Agent Director

The Agent Director ships with an Integrated Subagent that is running inside the Director.

• The Subagent is instructed by the active Director to execute jobs. The Director instance holding the active role can be the Subagent's Director or a second Director. 
• It is an option to disable the Director's Subagent in order to have the Director perform orchestration only and not to execute jobs with its Subagent.

The Agent Director is contacted by the Controller for deployment of scheduling objects, for commands that change an order's state such as suspend, resume and cancel, and to report back execution results of JS7 - Workflow Instructions.

Director Journal

The Agent Director holds a journal to store

• scheduling objects such as workflows and jobs that are deployed via a Controller,
• events about JS7 - Order State Transitions and to store log output reported from Subagents executing jobs.

The journal is essential for restart capabilities of the Agent Director.

Director Cluster

The Agent Director can be operated as a Standalone instance and as a member in a Director Cluster of two Agent Directors.

In a Director Cluster one Director holds the active role and the second Director holds the passive role.

If the Agent Director is operated as a cluster then the active Director synchronizes its journal with the standby Director. If both journals of active and standby Directors are in sync then a fail-over (automatically) or switch-over (caused by user intervention) can occur and Directors switch active and standby roles.

Subagent

Subagents come in two shapes: they ship with an Agent Director and they can be operated as Standalone instances.

• Subagents are lightweight and do not operate a journal.
• Subagents are deployed to any number of servers in a network.
• Subagents are closely monitored by Agent Directors using bi-directional heartbeats.

Job Execution

Subagents execute jobs and report back execution results.

• Subagents execute jobs on behalf of the active Agent Director.
• They report back log output and execution results of jobs to the active Agent Director.
• If a Subagent fails then the active Agent Director will hand-over the job execution request to a next Subagent.
• A Subagent kills running jobs if the connection from the active Agent Director is lost or if it is instructed by the active Agent Director to suspend/kill or to cancel/kill an order.
  • This behavior is intended to prevent double job execution by more than one Subagent.
  • If the connection from the active Agent Director can be re-established within a given timeout then the Subagent continues to execute jobs.

Subagent Cluster

Subagents can be grouped to clusters.

• A Subagent Cluster is specified by a Selection and Scheduling Mode:
  • The cluster can include a single Subagent, a number of Subagents or all Subagents.
  • The Scheduling Mode is one of fixed-priority or round-robin.
• Any number of Subagent Clusters can be configured.

Functional View

Jobs in workflows are assigned a Subagent Cluster that includes a Selection and Scheduling Mode of Subagents: 

• Subagent Clusters present a logical view on the way how a given number of Subagents co-operate for job execution. 
  • Any number of Subagent Clusters can be configured using the same Subagents.
• The Selection makes use of one or more Subagents.
  • A Subagent can be a member in one or more Subagent Clusters.
  • The configuration of Subagent Clusters is performed with JOC Cockpit and is forwarded to the Controller and to the active Agent Director.
• The Scheduling Mode is one of
  • fixed-priority: execute jobs with the first Subagent and switch to the next Subagent only if the first Subagent becomes unavailable (passive clustering).
  • round-robin: execute each next job on the next Subagent (active clustering).

Network Connections

Network connections use the HTTP protocol and can be secured by TLS/SSL certificates.

Connections are unidirectional and are established

• from the Controller to the Agent Director,
• from the Agent Director to the Subagent.

Workflow Execution

Workflows are deployed from JOC Cockpit to a Controller and are forwarded to the Agent Director. Similarly orders are submitted to a Controller and to the respective Agent Director.

Orders are scheduled for a given date and time that the Agent Director will start the workflow for.

• The Agent Director will execute a number of JS7 - Workflow Instructions, for example Retry, Try/Catch, Fork, that are in scope of the Agent Director.
• The Agent Director will request execution of jobs as available from the JS7 - Job Instruction with a Subagent. When choosing the Subagent the Agent Director considers the Selection of Subagents and Scheduling Mode of the Subagent Cluster that is assigned the job. For example, a specific Subagent will be selected for a fixed-priority scheduling mode.
• The Agent Director will hand back the order to the Controller if it meets an instruction that is out of scope of a single Agent, for example in case of a JS7 - Lock Instruction.