fix: Removes Redundant Version References (llm-d#903)

* remove redundant chart version references

---------

Signed-off-by: vezio <tyler.rimaldi@ibm.com>
Signed-off-by: Tyler Vezio Rimaldi <31221081+Vezio@users.noreply.github.com>

Author: Vezio

config/README.md

@@ -330,6 +330,73 @@ scenario:
       uriProtocol: hf     # No PVC, no download job — fetch at runtime
 ```
 
+### Code path
+
+1. `llmdbenchmark/standup/steps/step_04_model_namespace.py` — `_requires_pvc_download()` returns `False` when `uriProtocol != "pvc"`
+2. `config/templates/jinja/13_ms-values.yaml.j2` — conditionally generates `hf://` or `pvc://` URI
+3. `config/templates/jinja/04_download_job.yaml.j2` — only rendered/applied when protocol is `pvc`
+
+## Chart Versions
+
+All Helm chart and component versions are centralized in the `chartVersions` section of `defaults.yaml`. This is the single place to bump versions when upgrading components.
+
+| Field | Default | Description |
+|---|---|---|
+| `chartVersions.istioBase` | `1.29.1` | Istio base chart version |
+| `chartVersions.istiod` | `1.29.1` | Istiod chart version (also used as gateway version) |
+| `chartVersions.llmDInfra` | `auto` | llm-d-infra Helm chart (auto-resolved via helm) |
+| `chartVersions.llmDModelservice` | `auto` | llm-d-modelservice Helm chart (auto-resolved via helm) |
+| `chartVersions.inferencePool` | `v1.3.0` | Inference pool chart version |
+| `chartVersions.gaie` | `v1.3.1` | GAIE chart version |
+| `chartVersions.wva` | `auto` | Workload Variant Autoscaler chart (auto-resolved) |
+| `chartVersions.kgateway` | `v2.2.1` | kgateway chart version |
+| `chartVersions.lws` | `0.8.0` | LeaderWorkerSet chart version |
+
+Versions set to `auto` are resolved at plan time by `VersionResolver` using `helm search repo` or OCI registry queries (skopeo/crane). Fixed versions are used as-is.
+
+### Overriding versions in a scenario
+
+Add a `chartVersions` section to your scenario YAML. Only include the versions you want to change — the rest inherit from defaults:
+
+```yaml
+scenario:
+  - name: "my-upgrade-test"
+    chartVersions:
+      llmDModelservice: "0.5.0"    # pin to specific version
+      kgateway: "v2.3.0"           # upgrade kgateway
+```
+
+### Pinning all versions for reproducibility
+
+To ensure a benchmark run is fully reproducible, pin every `auto` version to a specific value. Run `plan` first to see what `auto` resolves to, then copy those values into your scenario:
+
+```yaml
+scenario:
+  - name: "reproducible-bench"
+    chartVersions:
+      llmDInfra: "v1.4.0"          # was auto
+      llmDModelservice: "v0.4.9"   # was auto
+      wva: "0.5.1"                 # was auto
+```
+
+### Upgrading Istio
+
+Istio uses two charts (`istio-base` and `istiod`) that must be the same version. Override both:
+
+```yaml
+chartVersions:
+  istioBase: "1.30.0"
+  istiod: "1.30.0"
+```
+
+### How `auto` resolution works
+
+1. For charts with a `helmRepositories` entry: queries the repo via `helm search repo` or OCI registry
+2. Falls back to `skopeo list-tags` or `crane ls` for OCI registries
+3. Selects the latest semver-compatible tag
+4. Resolved versions are logged during `plan`: `📦 Resolved chart llmDInfra to v1.4.0 (via repo URL)`
+
+> **Note:** `auto` versions may change between runs as upstream charts release new versions. Pin versions in your scenario for consistent results across runs.
 ## KV Transfer Configuration
 
 The `vllmCommon.kvTransfer` section controls the `--kv-transfer-config` argument passed to the `vllm serve` command. This is how vLLM knows which KV cache transfer connector to use and how to configure it.

config/templates/values/defaults.yaml

@@ -378,6 +378,8 @@ chartVersions:
   inferencePool: v1.3.0
   gaie: v1.3.1
   wva: auto
+  kgateway: v2.2.1
+  lws: "0.8.0"
 
 # ============================================================================
 # SERVICE ACCOUNT CONFIGURATION
@@ -484,7 +486,7 @@ gateway:
   className: istio
   provider: istio
   providerNamespace: istio-system
-  version: *istio_version
+  # Gateway version is derived from chartVersions.istiod at plan time
   logLevel: error
   service:
     type: NodePort
@@ -1209,15 +1211,15 @@ wva:
 # ============================================================================
 gatewayProviders:
   kgateway:
-    chartVersion: v2.2.1
+    # Chart version managed in chartVersions.kgateway
     namespace: kgateway-system
     helmRepository: oci://cr.kgateway.dev/kgateway-dev/charts
 
 # ============================================================================
 # LWS (LeaderWorkerSet)
 # ============================================================================
 lws:
-  chartVersion: "0.8.0"
+  # Chart version managed in chartVersions.lws
   namespace: lws-system
   helmRepository: oci://ghcr.io/kubernetes-sigs/lws
 

llmdbenchmark/parser/version_resolver.py

@@ -318,7 +318,12 @@ def _resolve_chart_versions(self, values: dict, unresolved: list) -> None:
                     unresolved.append(f"chartVersions.{chart_key}")
 
     def _resolve_gateway_version(self, values: dict) -> None:
-        """Resolve gateway version from the istio chart version."""
+        """Resolve gateway version from the istio chart version.
+
+        Kept for backward compatibility with configs that still set
+        ``gateway.version``.  New configs should use ``chartVersions.istiod``
+        directly.
+        """
         gateway = values.get("gateway", {})
         if gateway.get("version") == "auto":
             istio_version = values.get("chartVersions", {}).get("istiod")
@@ -340,7 +345,8 @@ def has_unresolved(self, values: dict) -> list[str]:
         standalone = values.get("standalone", {}).get("image", {})
         if isinstance(standalone, dict) and standalone.get("tag") == "auto":
             unresolved.append("standalone.image.tag")
-        if values.get("gateway", {}).get("version") == "auto":
+        gw_ver = values.get("gateway", {}).get("version")
+        if gw_ver == "auto":
             unresolved.append("gateway.version")
         wva = values.get("wva", {}).get("image", {})
         if isinstance(wva, dict) and wva.get("tag") == "auto":

llmdbenchmark/standup/steps/step_02_admin_prerequisites.py

@@ -339,7 +339,7 @@ def _install_lws_if_needed(
             )
             return
 
-        self._install_lws(cmd, lws_config, errors)
+        self._install_lws(cmd, lws_config, errors, plan_config=plan_config)
 
     def _install_prometheus_crds_if_needed(
         self,
@@ -422,7 +422,7 @@ def _apply_openshift_sccs(
 
     def _install_kgateway(self, cmd: CommandExecutor, plan_config: dict, errors: list):
         kgw = plan_config.get("gatewayProviders", {}).get("kgateway", {})
-        chart_version = kgw.get("chartVersion", "")
+        chart_version = plan_config.get("chartVersions", {}).get("kgateway", "") or kgw.get("chartVersion", "")
         namespace = self._require_config(kgw, "namespace")
         helm_repo = kgw.get("helmRepository", "")
 
@@ -502,8 +502,11 @@ def _install_istio(
         if not result.success:
             errors.append(f"Failed to install Istio via helmfile: {result.stderr}")
 
-    def _install_lws(self, cmd: CommandExecutor, lws_config: dict, errors: list):
-        version = lws_config.get("chartVersion", "")
+    def _install_lws(self, cmd: CommandExecutor, lws_config: dict, errors: list, plan_config: dict | None = None):
+        version = ""
+        if plan_config:
+            version = plan_config.get("chartVersions", {}).get("lws", "")
+        version = version or lws_config.get("chartVersion", "")
         namespace = self._require_config(lws_config, "namespace")
         helm_repo = lws_config.get("helmRepository", "")