Leaderboard Deploy agents

Implement intelligent failure handling that makes contextual decisions about retry, cleanup, and error propagation based on the type and likelihood of recovery from specific failures.

Consider these key principles:

• Analyze failure context: Use HTTP response codes and error types to determine if retry is appropriate (e.g., retry on rate limits/timeouts, but fail fast on 404s for missing resources)
• Implement bounded retries: Avoid infinite retry loops that could cause thundering herd problems during outages
• Clean up on failure: Remove partial state or broken installations when operations fail to prevent corrupted system state
• Use strict error handling: Enable proper error propagation with mechanisms like set -euo pipefail in bash scripts

Example implementation:

#!/usr/bin/env bash
set -euo pipefail

retry_with_context() {
    local url="$1"
    local temp_file="$2"
    local max_attempts=3
    
    for attempt in $(seq 1 $max_attempts); do
        if curl --fail --location "$url" > "$temp_file"; then
            return 0
        fi
        
        # Get HTTP code to decide if retry makes sense
        http_code=$(curl -s -o /dev/null -w "%{http_code}" "$url")
        case $http_code in
            404|403) 
                echo "Resource not found (HTTP $http_code), not retrying"
                return 1
                ;;
            429|5*) 
                echo "Temporary failure (HTTP $http_code), retrying in 2s..."
                sleep 2
                ;;
        esac
    done
    
    # Cleanup on final failure
    rm -f "$temp_file"
    return 1
}

This approach balances user experience with system reliability by making informed decisions about when failures are recoverable versus when they indicate fundamental problems that shouldn’t be retried.