fix: move keyword routing tests to e2e framework and validate matched_keywords (#694)

* Change the test localtion to e2e/testcase and remove not needed code that was for other tests cases that not done yet.

Signed-off-by: Senan Zedan <[email protected]>

* Update keyword routing tests to match reference implementation

Simplified from 31 to 13 tests covering core OR/AND/NOR operators.
Aligned configuration with documented keyword.yaml behavior.

Signed-off-by: Senan Zedan <[email protected]>

* Remove accidentally committed test report files

Signed-off-by: Senan Zedan <[email protected]>

* Fix keyword routing: change yaml field from 'category' to 'name'

CRITICAL BUG FIX: The KeywordRule struct expects yaml field 'name',
but the config was using 'category'. This caused all keyword rules
to be silently ignored, resulting in "no decision matched" errors.

Changes:
- keyword_rules: changed all instances of 'category:' to 'name:'
- This allows keyword classifier to properly load and match rules

Root cause: YAML unmarshaling failed silently because field names
didn't match the struct tags (yaml:"name" vs category in config)

Signed-off-by: Senan Zedan <[email protected]>

* Fix keyword routing in values.yaml: field names and decision types

CRITICAL BUG FIX: The values.yaml file is what actually gets deployed
to Kubernetes via Helm, not config.yaml. Fixed both issues:

1. keyword_rules: changed 'category:' to 'name:' (3 rules)
   - urgent_request, sensitive_data, exclude_spam

2. decision conditions: changed type from 'domain' to 'keyword' (3 decisions)
   - urgent_request_decision, sensitive_data_decision, exclude_spam_decision

Root cause: The config.yaml was just a reference file. The actual
deployment uses values.yaml which had the same bugs - keyword rules
were not being loaded at all.

Signed-off-by: Senan Zedan <[email protected]>

* Fix model references: change qwen3 to base-model

The vllm simulator backend only provides 'base-model', not 'qwen3'.
All decision modelRefs and default_model now use 'base-model' to
match the available backend model.

This fixes HTTP 404 errors: 'The model qwen3 does not exist'

Signed-off-by: Senan Zedan <[email protected]>

---------

Signed-off-by: Senan Zedan <[email protected]>

Author: szedan-rh

.github/workflows/integration-test-k8s.yml

@@ -22,7 +22,7 @@ jobs:
     strategy:
       fail-fast: false  # Continue testing other profiles even if one fails
       matrix:
-        profile: [ai-gateway, aibrix]
+        profile: [ai-gateway, aibrix, routing-strategies]
 
     steps:
       - name: Check out the repo