Gateway Controller Restart may not update Policy Engine/Router in some cases

[renuka-fernando]

Description

Problem

When the gateway controller restarts, the policy engine reconnects successfully but doesn't receive any configuration updates. APIs remain broken until someone manually deploys a new API or makes a change.

Steps to Reproduce:

1. Deploy an API successfully
2. Restart the gateway controller
3. Policy engine reconnects but has no config
4. API requests return 404 (broken)
5. Only after deploying a NEW API does everything work again

Root Cause (Simple Explanation)

The xDS protocol uses version numbers to track changes. When the controller restarts:

1. Controller: Loads configs from database, creates snapshot with version "1"
2. Policy Engine: Still has version "1" in memory from before the restart
3. xDS Protocol: Compares versions → "1" equals "1" → "no update needed"
4. Result: Policy engine never gets the configs until version changes

The version counter resets to 0 on every controller restart, causing version collisions.

How to Fix

Simple Solution (Recommended)

Make snapshot versions unique across restarts by including a timestamp.

Change version format from:

• "1", "2", "3" (just a counter)

To:

• "1739750400-1", "1739750400-2" (timestamp + counter)

Files to Change:

1. gateway/gateway-controller/pkg/storage/memory.go

   • Add startupTimestamp field to store when the controller started
   • Change IncrementSnapshotVersion() to return "{timestamp}-{counter}" instead of just counter
   • Change return type from int64 to string

2. gateway/gateway-controller/pkg/xds/snapshot.go

   • Use the string version directly (already compatible with string versions)

3. gateway/gateway-controller/pkg/api/handlers/handlers.go

   • Update status callback to accept string version instead of int64

Why This Works:

• Every restart gets a new timestamp
• Versions are always unique across restarts
• Policy engine sees different version → gets update immediately
• xDS protocol supports string versions natively

Alternative Quick Fix (Client-Side)

Reset policy engine's version memory on every reconnection.

File: gateway/gateway-runtime/policy-engine/internal/xdsclient/client.go

Add after line 258 (after c.setState(StateConnected)):

// Reset versions to force full sync
c.mu.Lock()
c.policyChainVersion = ""
c.apiKeyVersion = ""
c.lazyResourceVersion = ""
c.mu.Unlock()

Trade-off: Simple fix, but policy engine reprocesses everything on every reconnection.

Testing the Fix

# 1. Deploy an API
curl -X POST http://localhost:9090/apis -d @test-api.yaml

# 2. Verify it works
curl http://localhost:8080/test-path
# Should return 200 OK

# 3. Restart controller
docker compose restart gateway-controller

# 4. Wait 5 seconds for reconnection
sleep 5

# 5. Test API again (WITHOUT redeploying)
curl http://localhost:8080/test-path
# Should STILL return 200 OK (this is the fix!)

Priority

High - This breaks all deployed APIs on controller restart, requiring manual intervention to restore service.

Additional Context

• The gateway controller uses go-control-plane's State-of-the-World xDS protocol
• Version comparison is done by the go-control-plane library
• This only affects persistent mode (when configs are stored in database)
• Fresh deployments work fine - only restarts are affected