compute

Cyoda Compute Node Implementation

Compute nodes are optional — many Cyoda workflows work entirely with built-in processors and criteria. Only add a compute node when you need custom logic that Cyoda cannot express natively.

Do you need a compute node?

A compute node is needed when a workflow transition requires:

• Calling an external API or service
• Running custom validation or calculation logic
• Modifying entity data in a way built-in processors don't support

If your workflow only needs state changes, simple field criteria, or scheduled transitions — you likely don't need a compute node.

Implementation Guide

Refer to the detailed patterns reference: resources/grpc-patterns.md

Connection steps:

1. Connect to CloudEventsService.startStreaming via gRPC
2. Send CalculationMemberJoinEvent with your tags — see grpc-patterns.md
3. Receive CalculationMemberGreetEvent with your memberId
4. Handle incoming requests (processors or criteria)
5. Maintain keep-alive, implement reconnection

For processor implementation: see examples/processor.md For criteria implementation: see examples/criteria.md

Authentication

Use the JWT token from the active profile in ~/.config/cyoda/cyoda-plugin-config.json:

PROFILE=$(jq -r '.active // "default"' "$HOME/.config/cyoda/cyoda-plugin-config.json" 2>/dev/null || echo "default"); jq -r --arg p "$PROFILE" '.profiles[$p].token // "none — local mock auth"' "$HOME/.config/cyoda/cyoda-plugin-config.json" 2>/dev/null || echo "no token — local mock auth"

Include as bearer token in gRPC metadata: Authorization: Bearer {token}

For local cyoda-go with mock auth, authentication may not be required — check with /cyoda:docs.

Schema Details — Fetch Before Implementing

Before writing any implementation code, fetch the authoritative schemas:

1. cyoda help grpc proto — CloudEvent proto definition; generate language-specific stubs from this
2. cyoda help cloudevents json — JSON schemas for every event type with exact field names

Do NOT rely on field names from grpc-patterns.md or examples/ — those show structure and sequence only. The schemas are the authoritative source for field names, required fields, and types. Using stale example field names instead of checking the schemas is a common source of bugs.

Production Requirements

• Idempotency: deduplicate using requestId
• Thread safety: synchronize all stream writes
• Timeout: respond within 60 seconds
• Reconnection: exponential backoff (see grpc-patterns.md)
• Monitoring: track request latency, keep-alive intervals, reconnection count