Orchestration as a Data Management Challenge-Part 2

Tim Coats
Last Updated on November 16, 2023 by Editorial Staff
November 16, 2023
Reading time < 5 minutes

In Part 1 of this blog series, I proposed that closed-loop orchestration (CLO) is a data management problem. Data extends beyond your medical history, social media, and other PII in the digital world. Configurations, inventories, and monitoring systems are based on managing, interpreting, and actioning data. CLO is a workflow about manipulating the data around the target environments.  Even the policies and constraints need to be expressed as data.


The figure from the first blog is shown below again as a reminder of our framework. We have identified four key data types: user inputs, operational data, observability data, and Automation Packs. The data support the atomic actions of the workflow, which we grouped into five steps: initiation, planning, execution, confirmation, and updates. Automation packs provide API integration into the components of the infrastructure. Remember, these workflows are stateless without this data. 

Figure 1 – Generic Data Sets Required to Orchestrate a Workflow

This blog will dig into each data type and how it supports the workflow. Our focus is on the initiation phase and how the data gathered during this phase provides a foundation for planning the execution models.

Laying the Foundation: Initiation Phase

The initiation phase is about gathering the data needed to conduct the workflow. The data must be codified with specific data points, as machines are not good at figuring out intent. Data can be text, tables, tuples, or other forms.  Data must be accessible by each element, and there should only be one source of truth for the workflow to eliminate potential conflicts. 

  • The process starts by capturing the Objectives as the desired end state of the systems. The methods may be modifying infrastructure, updating software, or rolling out new applications or capabilities. These changes and additions will be instantiated in the configuration and running state of the environments.
  • Inventory information can come from external CMDBs or, more often, directly from the environment using the Automation APIs (see Figure 2). These APIs require the credentials to be available as part of the user inputs or be accessible from a credential management system. The inventory becomes part of the data management structure to ensure the correct data is used for the subsequent activities.
  • The running Inventory of the systems becomes the working CMDB. The inventory includes deployed assets, available allocations of resources, and details about the current capacities of those environments. Resource allocation targets can be direct user input or based on SLA targets with predefined tables.
  • And, of course, there are the governing Policies that not only guide the decisions made within the workflow but provide the constraints for those actions. These policies also include the requirements for ticketing and workflow management.
  • Each step will also send back telemetry as part of the Observability Data that is used to capture the results of the actions and make decisions during the processes.

Once gathered and organized, this information provides the needed information to perform the workflow, update the systems of record, and provide the logs of the actions taken. The data show the impacts of any changes and identify the root causes of any failures.

Figure 2 – Automation APIs

Data Management Techniques

Data not only needs to be managed, but it also needs context. Context is often overlooked when merging data from multiple sources into a single repository. The centralized data mechanism coordinated with the workflow to maintain the data context with respect to the actions and included systems.

The task of gathering this data is like that of advanced analytics. Data currency, completeness, and quality checks to support the workflow’s actions. Essential pieces that the data management system needs to understand are:

  • The data needed for the specific actions.  For example, the network information for new systems: IPs, subnets, VLAN, connection speed requirements, etc.,
  • The types and range of data are expected in the context of each system and action. Continuing our examples, the specific octets for the IP, the VLAN options, port speed and availability of the systems, etc.,
  • How the requirements, policies, and constraints impact the decisions, like placements based on SLA, routing requirements, firewall configurations, access to specific NFVs, etc.

The resulting information represents a real-time CMDB specific to the environment with the context needed to accurately model the workflow and assess the impact of its execution.

Orchestral.ai Tackles Data Gathering

Critical development criteria for Orchestral.ai’s Composer include data management capabilities. It was designed with the understanding that data management is key to CLO.

Some of these aspects include:

  • Access to automation packs for over 250 vendors that provide a low/no-code API approach to collecting data from the infrastructure.
  • Centralized data repository that all actions can access in real-time.
  • Telemetry from the systems to provide visibility of response at runtime.

In addition to the system logs, Composer maintains its own logs to provide information concerning the workflow’s actions.

The focus on data management has elevated Composer to one of the top functioning CLO engines on the market

But, it is only the beginning for Orchestral.ai.

Maestro – Leveraging AI to Contextual Digital Twins

As we alluded to earlier, when the data is collected as described, the completeness and coherency enable developing and deploying more advanced workflows using advanced analytics and AI.

And that is just Maestro’s capabilities to provide to the Orchestral suite.


CLOs of today work from predefined logic, static workflows, and simple decision trees. Maestro will provide more optimized processes that are responsive to the observed reactions. It will also be able to anticipate the impacts of change performed and compare results to the expected values. All this helps refine the system models and drives continuous improvement.In short, a digital twin of the environment can be created within the context of the specific environment. But to achieve this, a standard lexicon must be maintained across all sources to ensure the integrity of the data.  And it is into that discussion our third installation of this series will venture.

Learn more at: Orchestral.ai