[statsig-node] Fix EPIPE errors with proper HTTP agent connection pooling and timeouts

[holbrookab-augment]

Fix EPIPE Errors with Proper HTTP Agent Connection Pooling and Timeouts

Problem

The Node.js SDK was experiencing EPIPE (broken pipe / socket connection issues) errors for us when making HTTP requests. These errors occurred due to:

1. Improper connection pooling: The previous implementation using https-proxy-agent didn't properly handle connection reuse
2. Missing timeout configuration: No freeSocketTimeout was configured, causing stale connections to be reused after the server had closed them
3. Inadequate keep-alive settings: Connection pooling wasn't configured with proper timeout values

Solution

This PR replaces https-proxy-agent with agentkeepalive and hpagent to implement proper HTTP agent connection pooling with comprehensive timeout configuration:

Key Changes

1. Replaced dependencies:

   • Removed: https-proxy-agent@^7.0.6
   • Added: agentkeepalive@^4.5.0 and hpagent@^1.2.0

2. Implemented shared HTTP/HTTPS agents with proper timeout configuration:

   • timeout: 30000 - Request timeout matching SDK defaults
   • freeSocketTimeout: 15000 - Prevents EPIPE errors from stale connections
   • keepAlive: true and keepAliveMsecs: 1000 - Enables connection reuse

3. Enhanced proxy support:

   • Updated createProxyAgent → createProxyAgents to return both HTTP and HTTPS agents
   • Both proxy agents use the same timeout configuration as direct connections
   • Added getAgent helper function to select the appropriate agent based on URL protocol

Technical Details

• agentkeepalive: Provides freeSocketTimeout support across all Node.js versions, which is critical for preventing EPIPE errors when connections are reused after being idle
• hpagent: Supports all standard agent options (including keepAlive and freeSocketTimeout) for proxy scenarios, ensuring consistent behavior whether using a proxy or direct connections

Benefits

• ✅ Prevents EPIPE errors by properly handling stale connections
• ✅ Improves connection reliability through proper connection pooling
• ✅ Maintains backward compatibility - no breaking changes to the API
• ✅ Works consistently with or without proxy configuration

Testing

• All existing tests pass (120 tests)
• No breaking changes to the public API
• Backward compatible with existing proxy configurations
• Tested it separately in our own infrastructure: this package contains the changed code). We don't use the proxy but the other changes have not resulted in any prior socket issues in 2 days of current test.

Files Changed

• statsig-node/package.json - Updated dependencies
• statsig-node/src/lib/index.ts - Core HTTP agent implementation
• statsig-node/pnpm-lock.yaml - Lock file updates

[holbrookab-augment]

@weihao-statsig not urgent, but I made this PR based on some issues we've had recently. Hope it helps!

[holbrookab-augment]

@weihao-statsig kind bump

[weihao-statsig]

Ah sorry I missed this earlier, we are actually planning to kill the network layer for node.

[holbrookab-augment]

@weihao-statsig sorry, what does this mean? The keepAlive / freeSocketTimeout issues are important to us...by killing the network layer will that help with this?