Skip to main content

Orkes Conductor Glossary

Here are some common technical terms used in the Orkes Conductor documentation.

AI Agent​

An autonomous software entity that perceives its environment, reasons about its goals, and takes actions. An AI agent is typically an LLM equipped with a set of tools (e.g., function calling via APIs) and memory (e.g., context window via database access).

The LLM should be capable of instruction-driven reasoning, which can be prompted with techniques such as Reason & Act (ReAct), Chain-of-Thought (CoT), or Tree-of-thoughts (ToT).

Agentic Workflow​

A multi-step, dynamic process that orchestrates multiple services, tasks, AI agents, and even humans-in-the-loop within a dynamic control graph. Unlike a typical workflow, an agentic workflow involves an LLM that dynamically chooses execution paths at runtime.

An agentic workflow is flexible and can be used for many scenarios:

  • Introduce agentic elements (LLM-driven decision-making) in workflows for dynamism
  • Implement an AI agent as a workflow for improved governance
  • Orchestrate multi-agent collaboration

Application​

(In the context of Conductor) An access layer for external systems to interact with a Conductor server via APIs or SDKs. It is similar to a service account, where a system’s ability to access Conductor resources is governed by permissions.

An application access layer can be used in a variety of ways:

  • A worker application that polls the Conductor server for tasks to complete.
  • A client application that runs a Conductor workflow.
  • An application that manages other Conductor applications.

Asynchronous Execution​

Starts a workflow and returns the workflow (execution) ID to the system immediately. This ID can be used to check on the workflow progress and result.

Related: Synchronous Execution

Change Data Capture (CDC)​

A method for identifying changes in a source database and replicating them in real time to other systems. Change Data Capture (CDC) follows a three-step extract-transform-load process, where the relevant data is extracted, transformed into the required format for the target system, and then loaded into the system.

CDC provides a real-time stream of data changes, which ensures consistent replication across data-dependent systems.

Correlation ID​

A unique identifier that links together a related set of workflow executions. A correlation ID enables you to trace a single transaction or user across different workflows. For example, it can be used to link an order in a purchase workflow with its subsequent refund in the refundOrder workflow.

Directed Acyclic Graph (DAG)​

A type of graph where its vertices are connected in one direction and without any repeating cycles. All Conductor workflows are directed acyclic graphs (DAGs), providing a clear map of the workflow and its progress through different states.

Event-Driven​

An architectural design where the system communicates via event messages to drive the flow of subsequent tasks. For example, in event-driven microservices, each microservice takes action in response to specific events, and sends events to other microservices based on its own state change. This could look like a payment service that triggers credit card authorization upon receiving an order confirmation event and thereafter sends out a payment authorization event for other services to be notified of.

Event-driven architecture facilitates loose coupling between services and responsiveness.

Event Handler​

A module that either listens for specific events in an event-driven system. This can be a piece of code, as simple as a callback function, or in Conductor, a JSON configuration. As implied by its name, the event handler reacts to particular events and triggers subsequent actions like starting a workflow or updating a workflow variable.

Idempotency​

A property where an operation executed multiple times will not affect the result beyond the initial execution. Idempotency prevents any unwanted side effects or duplicate results. For example, if an API call to delete a resource occurs repeatedly, it does not create any side effects. Or if the same event message is sent multiple times, subsequent messages will not trigger the service again.

In Conductor, workflows can be made idempotent to prevent duplicate workflow executions. Upon duplicate requests, the Conductor server can return the initial workflow execution ID or fail the request.

JSON Web Token (JWT)​

An open standard for securely transmitting information between two systems, often used for web-based authentication. The JSON web token (JWT) is stateless and self-contained, carrying all the necessary information for authentication, like the user’s claims and permissions, in a JSON format.

Each JWT is also signed using cryptography to ensure that a third party cannot alter its contents. A signed JWT can be used for client-to-server or server-to-server communication.

JDBC (Java Database Connectivity)​

An API that enables connectivity to any database management system (DBMS) like PostgreSQL. JDBC enables applications to easily query, update, add, or delete data.

LLM (Large Language Model)​

An AI model that is trained on vast amounts of data and can comprehend and generate natural language as a result. An LLM (large language model) is built on deep learning architectures. It can be used in chatbots or as virtual assistants for a variety of tasks like text generation, translation, question answering, and code generation.

Operator​

A built-in task type in Conductor that provides control flow logic, like parallel tasks, iterative loops, and if-else switches.

Orchestration​

The process of coordinating distributed software components, systems, agents, and humans so that they execute seamlessly as an automated, repeatable process. For example, an order management process may involve coordinating across an inventory system, a payment service, multiple notification channels, and even the logistics and shipping partner.

Orchestration facilitates state tracking and reliable execution of interdependent tasks.

Retrieval Augmented Generation (RAG)​

An AI technique that supplements an LLM with external knowledge, enhancing its final output (grounded generation). With RAG, relevant information is retrieved from a database based on the user query and supplemented to the LLM.

Rate Limits​

A method to control the number of network requests made to a server at any given time.

By rate limiting workflows or their tasks, you can prevent task workers from being overloaded by requests and also gracefully handle external limits set by a service provider.

Server​

A computer program that provides a service to other programs or users.

The Conductor server stores all workflow metadata resources, like the workflow definition, task definition, and integrations, and manages workflow executions, enabling users to build workflows and applications to run workflows.

Sub-Workflow​

A workflow that executes within another workflow (usually referred to as the parent workflow).

Sub-workflows promote modularity and reusability of common processes across different workflows. For example, a notification workflow can be used as a sub-workflow across CI/CD pipelines or system monitoring workflows.

Synchronous Execution​

Starts a workflow and returns the workflow output either once the workflow has reached a terminal state or when a certain duration has elapsed.

Related: Asynchronous Execution

System Task​

A built-in task type in Conductor that provides common logic without coding required, like calling an HTTP endpoint, publishing an event message, or calling an LLM.

Task​

A unit in a Conductor workflow that carries out some kind of business logic or control flow logic. Each task has a well-defined input and output, and contains a status (like IN PROGRESS, COMPLETED, FAILED) for state tracking.

Related: Task Configuration, Task Definition

Task Configuration​

A workflow-specific configuration, specifying the task type, reference name, input parameters, and other details about the task to be executed. The task configuration is part of the workflow definition and it is an ordered array that describes the order and control flow of tasks, the data flow between tasks, and task behaviour like optionality, caching, and schema enforcement.

Related: Task, Task Definition

Task Definition​

A configuration that is general to a specific task type or name, and applies to all instances of the task across workflows. It contains general implementation details, like rate limits, expected input and output keys, and failure-handling behavior, like retries and timeouts.

Related: Task, Task Configuration

Task Routing​

A strategy that distributes task executions for a particular task to different worker pools. Task routing optimizes resource allocation and usage across different scenarios. In Conductor, task routing is accomplished using a task-to-domain approach.

Related: Task-to-Domain

Task Queue​

A system design where task executions are grouped into queues according to their task type or name (HTTP, Event, some_custom_task) and domain (if applicable), and get completed by task workers in a first-in-first-out manner.

Task-to-Domain​

An approach that routes task executions specified under a domain name to its corresponding worker pool with the same domain name. To do so, you must map each worker pool to a specific domain beforehand, then specify the task’s domain mapping when triggering the workflow.

Task-to-domain can be used to route traffic from different origins (web, desktop, mobile) to different worker pools, or to customize the same task with domain-specific business logic.

Related: Developer Guide for Routing Tasks

Vector Database​

Stores data in the form of vectors, which are arrays of numbers that represent text, images, or audio. These vectors are stored in a multi-dimensional vector space and are clustered together based on similarity. With this form of storage, vector databases are suited for low-latency storage and retrieval of unstructured data.

Webhook​

An event-driven integration technique where a source system sends an HTTP request to a destination system through a registered endpoint every time a specified event in the source system occurs. This HTTP request is also known as a webhook payload, which often contains data about the event. The webhook endpoint is configured in the destination system and registered to the source system.

Webhooks enable real-time updates from source systems without the need for constant polling by the destination system.

Worker​

A process or program that accomplishes some task logic. Workers poll the Conductor server for work to be done in its corresponding task queue.

System workers are responsible for executing system tasks and are managed by Conductor. External workers can be created to handle custom logic beyond the scope of system tasks, and can run on any environment.

Related: Worker Task

Worker Task​

(Also known as a Simple Task) A Conductor task type for custom logic handled by external workers.

Related: Worker

Workflow​

A sequence of tasks with a defined order and execution flow, aimed at carrying out a specific process in an automated fashion, such as claims processing or cloud upgrades.

Related: Workflow Definition, Workflow Execution

Workflow Definition​

The JSON structure that specifies how the workflow should be executed at runtime, such as the workflow inputs, caching behavior, the series of tasks involved, and so on.

Related: Workflow, Workflow Execution

Workflow Execution​

The execution instance of a given workflow definition. The execution JSON contains the runtime conditions like duration, workflow state, the initial trigger, the data that passes through the workflow, and so on.

Related: Workflow, Workflow Definition, Workflow ID

Workflow ID​

The execution ID of a Conductor workflow instance. The workflow ID can be used to track the execution progress and workflow state.

Workflow Scheduler​

Allows workflows to be scheduled and executed on a regular basis. The workflow scheduler is configured using cron expressions.