[PERFORMANCE]: Implement Lazy Service Initialization

[crivetimihai]

Implement lazy service initialization for 50-80% faster cold start times.

Problem

Location: mcpgateway/main.py lines 110-125

All services imported at module load time:

from mcpgateway.services.a2a_service import A2AAgentService
from mcpgateway.services.completion_service import CompletionService
from mcpgateway.services.export_service import ExportService
from mcpgateway.services.gateway_service import GatewayService
from mcpgateway.services.import_service import ImportService
from mcpgateway.services.prompt_service import PromptService
from mcpgateway.services.resource_service import ResourceService
from mcpgateway.services.server_service import ServerService
from mcpgateway.services.tag_service import TagService
from mcpgateway.services.tool_service import ToolService
# ... more services and error classes

Current Startup Timeline

Phase	Time	Notes
Python import	200ms	All dependencies
Database check	50-500ms	Varies by DB
Service init	100-300ms	10+ services
Redis check	50-200ms	If enabled
Plugin load	50-100ms	Plugin manifests
Total	450-1300ms	Cold start

Proposed Solution

Lazy Service Singleton Pattern

from functools import lru_cache

@lru_cache
def get_tool_service():
    from mcpgateway.services.tool_service import ToolService
    return ToolService()

# In routes:
@router.get("/tools")
async def list_tools(
    tool_service: ToolService = Depends(get_tool_service)
):
    return await tool_service.list_tools()

Feature-Based Service Loading

def get_a2a_service():
    if not settings.a2a_enabled:
        raise HTTPException(501, "A2A not enabled")
    from mcpgateway.services.a2a_service import A2AService
    return ServiceRegistry.get(A2AService)

Performance Targets

Metric	Current	Target
Cold start (minimal)	450ms	150ms
Cold start (full)	1300ms	400ms
First request latency	+0ms	+100ms (one-time)

Acceptance Criteria

• Services instantiated on first use, not import
• Database check deferred to first database operation
• Redis check deferred to first cache operation
• Disabled features don't load their services
• Startup time reduced by 50%+
• All existing tests pass
• Passes make verify