Hosted Studio Debug

Hosted Studio Debug Skill

Debug GenLayer Studio deployments via ArgoCD CLI.

Workload Manifests

Kubernetes manifests are in sibling repo ../devexp-apps-workload (assume by default, ask user if not found):

devexp-apps-workload/workload/
├── dev/           # studio-dev, rally-studio-dev
├── stg/           # studio-stg
├── prd/           # studio-prd, rally-studio-prd

Each contains Deployments, Services, Ingresses, ExternalSecrets managed by ArgoCD.

Prerequisites

• Logged into ArgoCD CLI
• Access to target cluster

Full Status Check

To get complete visibility into Studio status, check all components across all replicas:

# 1. Overall app health
argocd app get <app>-workload

# 2. List all pods and their status
argocd app resources <app>-workload --kind Pod

# 3. Check consensus worker logs (all 4 replicas in prd)
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep -iE "(error|exception|timeout|failed)"

# 4. Check JSON-RPC logs (all 2 replicas in prd)
argocd app logs <app>-workload --name studio-jsonrpc --tail 500 2>&1 | grep -iE "(error|exception|timeout|failed)"

# 5. Check webdriver logs
argocd app logs <app>-workload --name studio-webdriver --tail 200 2>&1 | grep -iE "(error|exception|crash)"

Quick Commands

# Check app health
argocd app get <app>-workload

# List all resources (pods, deployments, services, etc.)
argocd app resources <app>-workload

# List all pods with their status
argocd app resources <app>-workload --kind Pod

# Tail consensus worker logs (aggregates from ALL replicas)
argocd app logs <app>-workload --name studio-consensus-worker --tail 200

# Tail JSON-RPC logs (aggregates from ALL replicas)
argocd app logs <app>-workload --name studio-jsonrpc --tail 200

# Check for errors across all replicas
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep -i error
argocd app logs <app>-workload --name studio-jsonrpc --tail 500 2>&1 | grep -i error

Multi-Replica Log Access

Important: The argocd app logs --name <container> command automatically aggregates logs from ALL pod replicas. This ensures full visibility across the entire deployment.

# Get logs from all consensus worker replicas (production has 4)
argocd app logs <app>-workload --name studio-consensus-worker --tail 500

# Get logs from all JSON-RPC replicas (production has 2)
argocd app logs <app>-workload --name studio-jsonrpc --tail 500

# Get logs from a specific pod (if needed for isolation)
argocd app logs <app>-workload --pod <pod-name> --tail 200

# List pods first to get pod names
argocd app resources <app>-workload --kind Pod

Common Issues

Transaction Timeouts

Consensus worker timeouts usually mean GenVM Manager is unresponsive. Check all 4 workers in production:

# Check for GenVM timeouts across all consensus worker replicas
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep -E "(timeout|SocketTimeoutError|127.0.0.1:3999)"

# Empty stdout = GenVM never started
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep "stdout=''"

Root cause: GenVM Manager (genvm-modules manager --port 3999) becomes unresponsive. Fix: Worker restart (auto or manual) restarts GenVM Manager.

Key files:

• backend/node/genvm/origin/base_host.py:463 - where timeouts occur
• backend/node/base.py - Manager.create() spawns GenVM
• backend/consensus/worker_service.py - worker startup/health

Pod Restarts

# Check restart timestamps in logs
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep -E "(Started|Uvicorn running)"

# Via kubectl
kubectl get pods -n <namespace> -o wide
kubectl get events -n <namespace> --sort-by='.lastTimestamp'

External API Failures

Contracts calling external APIs (Twitter, etc.) through proxies:

# Check for HTTP errors in contract execution
argocd app logs <app>-workload --name studio-consensus-worker --tail 500 2>&1 | grep -E "(HTTP|fetch|proxy|api)"

Environment Apps

Components

Note: When debugging, always check logs from ALL replicas to get complete visibility. The argocd app logs --name <container> command handles this automatically.