Skip to main content
The RIME Edge Server (RES) is the edge computing engine that powers automated monitoring, detection, and violation handling. RES runs on edge devices at each location, processing branch rules, executing workflows, and managing the complete lifecycle of violations from detection to brand notification.

​
What RES does

RES provides three core capabilities:
  1. Automated rule execution - Processes scheduled branch rules based on cron expressions and operation time windows
  2. Intelligent detection - Runs computer vision models to detect objects, people, and events in camera streams
  3. Violation management - Creates, escalates, and delivers violations to brands through multiple channels

​
Architecture overview

RES uses a master-worker job queue architecture that efficiently processes multiple rules simultaneously while sharing AI model memory across workers. System components:
ComponentRole
Master threadFetches ready jobs from the execution plan and adds them to the queue
Job queueThread-safe Python queue with deduplication to prevent duplicate executions
Worker threadsPull jobs from the queue and execute workflows
Execution plan managerCaches branch rules in memory and syncs with the database
Workflow engineExecutes action sequences defined in workflow configurations
Actions layerDetection actions, business actions, and data actions
Data flow:
  1. Master thread queries the execution plan for ready branch rules
  2. Ready rules are added to the job queue with deduplication
  3. Worker threads pull jobs and execute their workflows
  4. Workflow results trigger business actions (violations, notifications)
  5. Post-processing updates the next run time and releases the job

​
Key concepts

A branch rule connects a rule definition to a specific branch (location). It defines:
  • Operation time window - When the rule should run (e.g., 8:00 AM to 6:00 PM)
  • Next run time - When the rule should execute next
  • Cron expression - The schedule for recurring executions
A workflow is a sequence of actions that execute in order. Workflows can include:
  • Detection actions (computer vision analysis)
  • Business actions (create violations, send notifications)
  • Conditional branching based on detection results
An action detail is a specific configuration of an action for a branch rule. It includes:
  • Stream configuration (which camera to analyze)
  • Model labels (what objects to detect)
  • Zone definitions (areas of interest in the frame)
  • Confidence thresholds
A violation is a record of a detected rule breach. It contains:
  • Timestamp when the violation occurred
  • Attachments (images, video clips)
  • Status tracking for escalation

​
How RES processes rules

1

Fetch ready rules

The master thread queries the execution plan for branch rules that are:
  • Active and not archived
  • Within their operation time window
  • Past their next run time
  • Not currently running
2

Queue for execution

Ready rules are added to the job queue with deduplication to prevent the same rule from running twice simultaneously.
3

Execute workflow

A worker thread claims the job, marks it as running, and executes the workflow:
  1. Process detection actions (analyze camera streams)
  2. Evaluate consolidation logic (combine results from multiple streams)
  3. Execute business actions if conditions are met (create violations, send alerts)
4

Post-process and reschedule

After workflow completion:
  1. Update the branch rule’s next_run_time based on cron expression
  2. Mark the rule as not running
  3. Log execution results

​
Memory and resource management

RES implements several strategies to efficiently manage memory and prevent resource exhaustion:

Memory throttling

Automatically pauses job fetching when RAM exceeds 90% or GPU memory exceeds 85%, resuming when usage drops below safe thresholds.

Single-job mode

Workers exit after each job completion, ensuring complete memory cleanup including thread-local storage.

Stream pooling

Camera streams are shared across workers through reference counting, reducing connection overhead.

Model sharing

AI models are loaded once and shared across all worker threads, minimizing GPU memory usage.

​
Next steps

Running a branch rule

Walk through the complete flow of executing a branch rule from start to finish.

Violation lifecycle

Understand how violations are created, escalated, and delivered to brands.

Detection actions

Learn about the detect action and how to configure computer vision analysis.

Business actions

Explore available business actions for handling detected events.