Actor Errors

Overview

Actor errors come from two sources:

1. Direct actor errors - From the actor workflow itself (stored in workflow state as ActorError)
2. Runner pool errors - From serverless endpoints, tracked by the runner pool error tracker workflow

Errors use a two-layer representation:

• ActorError - Internal enum stored in workflow state (NoCapacity, RunnerNoResponse)
• ActorError - API-facing enum with enriched context (PoolError, NoCapacity, RunnerNoResponse)

The runner pool error tracker monitors serverless endpoint health. When errors occur, it stores them with timestamps. Errors are cleared after 3 consecutive successes (hysteresis).

Error Sources

Direct Actor Errors

Stored in actor workflow state as failure_reason:

• NoCapacity - No runner available to allocate the actor
• RunnerNoResponse - Runner was allocated but became unresponsive (includes runner_id)

Set when:

• Runner allocation fails or times out
• Runner disconnects while actor was allocated

Runner Pool Errors

Tracked separately by runner_pool_error_tracker workflow for serverless configs:

• RunnerPoolError - Contains error details from failed endpoint requests. Variants include:
  • ServerlessHttpError - HTTP error (status code, body)
  • ServerlessStreamEndedEarly - SSE stream ended before runner initialized
  • ServerlessConnectionError - Network/connection error
  • ServerlessInvalidBase64 - Invalid base64 in SSE message
  • ServerlessInvalidPayload - Invalid protocol payload
  • InternalError - Internal errors (namespace not found, config not found, etc.)

The error tracker:

• Receives errors from the serverless pool manager
• Stores active_error with timestamp
• Clears error after 3 consecutive successful runner startups
• Makes state queryable for API enrichment and guard fail-fast

Error Consumption

API Response Enrichment

When fetching actors via the API (pegboard::ops::actor::get):

1. Check if actor has failure_reason: NoCapacity
2. Query the error tracker workflow for the actor's runner config
3. If pool has active error, return PoolError with details
4. Otherwise return plain NoCapacity

This provides users with actionable error info (e.g., "serverless endpoint returned 500") instead of generic "no capacity".

Guard Fail-Fast

When guard is waiting for an actor to become ready:

1. Wait 1 second before starting checks (allow normal startup)
2. Every 2 seconds, poll the error tracker workflow
3. If active error exists, return actor_runner_failed immediately

This prevents clients from waiting the full ready timeout when the runner pool is known to be failing.

Runner Configs API

The runner configs list endpoint enriches serverless configs with current pool errors:

1. For each serverless config, find its error tracker workflow
2. If active error exists, populate pool_error field
3. Allows dashboards to show pool health status

Error Flow Diagram

---
config:
  theme: mc
  look: classic
---
flowchart TD
    subgraph Sources
        AWF[Actor Workflow]
        SE[Serverless Endpoint]
    end

    subgraph Storage
        WFS[Workflow State
failure_reason: ActorError]
        ET[Error Tracker Workflow
active_error: RunnerPoolError]
    end

    subgraph Consumers
        API[Actor Get API]
        Guard[Guard Fail-Fast]
        RC[Runner Configs API]
    end

    AWF -->|NoCapacity
RunnerNoResponse| WFS
    SE -->|Connection errors| ET

    WFS --> API
    ET --> API
    API -->|ActorError| Client1[Client]

    ET --> Guard
    Guard -->|actor_runner_failed| Client2[Client]

    ET --> RC
    RC -->|pool_error| Dashboard[Dashboard]