fix(monitoring): set MemoryTelemetry logger to INFO for production

[waleedlatif1]

Summary

• Production logger defaults to ERROR-only, which silently suppressed our logger.info('Memory snapshot', ...) calls
• Adds { logLevel: 'INFO' } override specifically for the MemoryTelemetry logger so snapshots appear in CloudWatch

Test plan

• Deploy to staging and verify Memory snapshot entries appear in /ecs/sim-staging/us-east-1/app CloudWatch logs every 60s

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Feb 28, 2026 9:54pm

[greptile-apps]

Greptile Summary

This PR implements memory telemetry for production monitoring and includes several defensive improvements to prevent memory leaks and unbounded growth.

Key changes:

• Added MemoryTelemetry logger with logLevel: 'INFO' override to ensure memory snapshots appear in CloudWatch (fixes the production logger's ERROR-only default)
• Implemented SSE connection tracking across all streaming endpoints (a2a-message, a2a-resubscribe, mcp-events, wand, workflow-execute, execution-stream-reconnect)
• Added protection against unbounded memory growth during Redis failures by limiting pending event queues
• Limited stderr accumulation in isolated VM workers to 64KB
• Refactored cleanup logic in A2A resubscribe handler to eliminate duplication

Monitoring improvements:
Memory snapshots now log every 60s including heap stats, RSS, active resources, and SSE connection counts by route, enabling correlation between connection leaks and memory spikes.

Defensive measures:

• Event buffers drop oldest events when exceeding limits during sustained Redis failures
• Worker stderr is capped to prevent unbounded accumulation
• SSE tracking uses guard flags to prevent double-decrement in all exit paths

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• All changes are defensive improvements for production monitoring and memory leak prevention. The SSE connection tracking logic properly uses guard flags to prevent double-decrement, cleanup functions are correctly called in all exit paths (finally blocks and cancel methods), and the memory telemetry logger override is the intended fix for production visibility.
• No files require special attention - verify in staging that Memory snapshot logs appear in CloudWatch as expected

Important Files Changed

Filename	Overview
apps/sim/lib/monitoring/memory-telemetry.ts	New file that logs memory snapshots with INFO level override to ensure production visibility
apps/sim/lib/monitoring/sse-connections.ts	New SSE connection tracker using in-memory Map for memory leak diagnostics
apps/sim/instrumentation-node.ts	Starts memory telemetry monitoring on app initialization
apps/sim/app/api/a2a/serve/[agentId]/route.ts	Added SSE connection tracking with proper increment/decrement guards, refactored cleanup logic
apps/sim/app/api/mcp/events/route.ts	Added SSE connection tracking with cleanup guard to prevent double-decrement
apps/sim/lib/execution/event-buffer.ts	Added pending event limit with oldest-first eviction during Redis outages

_{Last reviewed commit: 4788705}

[greptile-apps]

_{11 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}