Revert "[tests] Adds e2e tests for the /inference/v1/generate endpoint. (#1433)" (#2015)

This reverts commit 95e74d1.

Author: ahg-g

test/e2e/configs_test.go

@@ -121,55 +121,10 @@ schedulingProfiles:
     weight: 2
 `
 
-// generateEncodeConfig is the encode-only EPP config for /inference/v1/generate.
-// Uses single-profile-handler so the EPP routes directly to encode pods without
-// requiring a decode stage.
-const generateEncodeConfig = `apiVersion: llm-d.ai/v1alpha1
-kind: EndpointPickerConfig
-plugins:
-- type: vllmhttp-parser
-- type: encode-filter
-- type: max-score-picker
-- type: single-profile-handler
-requestHandler:
-  parsers:
-   - pluginRef: vllmhttp-parser
-schedulingProfiles:
-- name: default
-  plugins:
-  - pluginRef: encode-filter
-  - pluginRef: max-score-picker
-`
-
-// generatePrefillConfig is the prefill-only EPP config for /inference/v1/generate.
-// Uses single-profile-handler so the EPP routes directly to prefill pods without
-// requiring a decode stage.
-const generatePrefillConfig = `apiVersion: llm-d.ai/v1alpha1
-kind: EndpointPickerConfig
-plugins:
-- type: vllmhttp-parser
-- type: prefill-filter
-- type: max-score-picker
-- type: single-profile-handler
-requestHandler:
-  parsers:
-   - pluginRef: vllmhttp-parser
-schedulingProfiles:
-- name: default
-  plugins:
-  - pluginRef: prefill-filter
-  - pluginRef: max-score-picker
-`
-
 // EPP configuration for running with P/D using the unified disagg-profile-handler
-// pdConfig uses vllmhttp-parser as the request handler so the EPP can parse
-// both OpenAI-style and /inference/v1/generate (token-in) traffic. The parser
-// delegates non-generate paths to the embedded OpenAI parser, so existing
-// chat/completions tests are unaffected.
 const pdConfig = `apiVersion: llm-d.ai/v1alpha1
 kind: EndpointPickerConfig
 plugins:
-- type: vllmhttp-parser
 - type: approx-prefix-cache-producer
   parameters:
     blockSizeTokens: 16

test/e2e/e2e_test.go

@@ -26,10 +26,6 @@ const (
 	// ePDDisaggDir references the Kustomize directory for the deployment
 	// running vLLM with E/P/D (Encode/Prefill/Decode)
 	ePDDisaggDir = "../../deploy/environments/dev/e-p-d"
-	// encodeOnlyDir is the single-component kustomize path for encode-only pods.
-	encodeOnlyDir = "../../deploy/components/vllm-encode"
-	// prefillOnlyDir is the single-component kustomize path for prefill-only pods.
-	prefillOnlyDir = "../../deploy/components/vllm-prefill"
 
 	simplePrompt = "Hello my name is Andrew, I have a doctorate in Rocket Science, and I like interplanetary space exploration"
 	extraPrompt  = "Why is the sky sometimes blue and sometimes red close to sunset?"

test/e2e/generate_endpoint_test.go

@@ -122,22 +122,6 @@ func createModelServersEPDDisagg(encodeReplicas, prefillReplicas, decodeReplicas
 	})
 }
 
-// createModelServersEncodeOnly creates encode-only pods (no prefill, no decode).
-func createModelServersEncodeOnly(replicas int) []string {
-	return createModelServersFromKustomize(encodeOnlyDir, map[string]string{
-		"${EC_CONNECTOR_TYPE}":    "",
-		"${VLLM_REPLICA_COUNT_E}": strconv.Itoa(replicas),
-	})
-}
-
-// createModelServersPrefillOnly creates prefill-only pods (no encode, no decode).
-func createModelServersPrefillOnly(replicas int) []string {
-	return createModelServersFromKustomize(prefillOnlyDir, map[string]string{
-		"${KV_CONNECTOR_TYPE}":    "",
-		"${VLLM_REPLICA_COUNT_P}": strconv.Itoa(replicas),
-	})
-}
-
 // createModelServersEPDUnified creates model server resources for EPD (one deployment for encode/prefill/decode) testing.
 func createModelServersEPDUnified(replicas int) []string {
 	return createModelServersFromKustomize(epdDeploymentDir, map[string]string{