Spring MCP Server connecting to MCP gateway 0.9.0 facing JVM OutOfMemoryError despite limited number of requests

[ikauwsp]

We are encountering a memory leak issue in our Spring-based MCP server (spring-ai-bom version 1.1.0), which is deployed behind 0.9.0 MCP Gateway. The pod consistently runs for about 3 days before crashing with a java.lang.OutOfMemoryError, even under low request volume.

High-level design:
users → MCP Gateway → Spring MCP Server → APIs

The server is not called directly by users, only through the MCP Gateway

Server Protocol: Streamable HTTP, Sync

Observed behavior

Case 1: Server running without MCP Gateway
• Old Gen heap remains stable
• No OOM occurs
• Memory usage behaves normally under idle

Case 2: Server running with MCP Gateway
• Old Gen heap shows a steady increase over time
• Eventually leads to OutOfMemoryError
• Behavior is reproducible
• Issue only occurs when connected through the Gateway

Suspicion / Hypothesis

From logs and runtime behavior, we observed:
• Frequent client initialization events from the Gateway
• Default health check interval = 60s
• New connections appear to be created repeatedly
• Possible accumulation of:
• idle sessions
• HTTP objects
• SSE connections
• request metadata

We suspect:

The Gateway may be creating new client sessions or HTTP connections repeatedly and not closing them properly.

This could lead to:
• Idle connections piling up
• Objects escaping GC
• Promotion into Old Gen
• Eventual memory exhaustion

We’re wondering:
• Does the MCP Gateway expose configuration for:
• connection reuse?
• session cleanup?
• health check behavior?
• timeout / eviction?
• Is there a best practice for managing long-lived connections in this setup?

If helpful, we can provide more context, jvm old gen heap trends, logs.

Thank you!