A collection of small contributions

README.md

@@ -1,21 +1,23 @@
 # llm-load-test-exporter
 
-### Overview
-The purpose of this program is to run [llm-load-test](https://github.com/openshift-psap/llm-load-test) application and then serve the resulting metrics to a /metrics endpoint. This is meant to be run in a kubernetes/openshift environment.
+## Overview
 
-This application is split into 2:
-1. run-llm-load-test: This is the application that actually runs llm-load-test against all models running in a cluster, and saves the output into a volume.
-2. exporter: This application exports the results of running llm-load-test to the /metrics endpoint. 
+The purpose of this program is to run [llm-load-test](https://github.com/openshift-psap/llm-load-test) application and then serve the resulting metrics to a `/metrics` endpoint. This is meant to be run in a kubernetes/OpenShift environment.
 
-### Deploying Application in OpenShift
+This application is split into two parts:
+
+1. `run-llm-load-test`: the application that actually runs `llm-load-test` against all models running in a cluster, and saves the output into a volume.
+2. `exporter`: exports the results of running `llm-load-test` to the `/metrics` endpoint.
+
+## Deploying Application in OpenShift
 
 In order to deploy this application in OpenShift do the following:
 
 1. Modify the namespace this will be deployed to in base/kustomization.yaml
-2. Modify how often the load test is run by modifying WAIT_TIME env variable in base/deployment.yaml
-3. Run `oc create -k base/`
+2. Modify how often the load test is run by modifying `WAIT_TIME` environment variable in `base/deployment.yaml`
+3. Run `oc apply -k base`
 
-### Example output when querying the /metrics endpoint:
+## Example output when querying the /metrics endpoint
 
 ```
 # HELP llm_performance_itl Inter-token Latency (ms)
@@ -40,4 +42,4 @@ llm_performance_throughput{model="granite",namespace="granite-instruct"} 18.8040
 llm_performance_latency{model="granite-internal",namespace="granite-instruct"} 277.4471044540405
 llm_performance_latency{model="granite-test2",namespace="granite-instruct"} 285.0987911224365
 llm_performance_latency{model="granite",namespace="granite-instruct"} 268.87357234954834
-```
+```

base/deployment.yaml

@@ -7,10 +7,10 @@ spec:
   template:
     spec:
       initContainers:                
-        - name: init-container
-          image: alpine/git:latest  
+        - name: clone-repository
+          image: docker.io/alpine/git:latest  
           imagePullPolicy: Always
-          command: ['sh', '-c', 'git clone https://github.com/openshift-psap/llm-load-test.git /shared_data/llm-load-test']
+          command: ['git', 'clone', 'https://github.com/openshift-psap/llm-load-test.git /shared_data/llm-load-test']
           volumeMounts:
             - name: llm-load-test-dir
               mountPath: /shared_data
@@ -19,7 +19,8 @@ spec:
           image: quay.io/rh-ee-istaplet/nerc-tools:llm-load-test-exporter
           imagePullPolicy: Always
           ports:
-            - containerPort: 8080
+            - name: web
+              containerPort: 8080
           volumeMounts:
             - name: llm-load-test-dir
               mountPath: /shared_data

base/files/uwl_metrics_list.yaml

@@ -0,0 +1,3 @@
+matches:
+  - __name__=~"(llm_performance_.*)"
+  - __name__=~"(vllm:.*)"

base/kustomization.yaml

@@ -11,4 +11,10 @@ resources:
   - clusterrole.yaml
   - clusterrolebinding.yaml
   - servicemonitor.yaml
-  - configmap.yaml
+
+configMapGenerator:
+  - name: observability-metrics-custom-allowlist
+    options:
+      disableNameSuffixHash: true
+    files:
+    - files/uwl_metrics_list.yaml

base/service.yaml

@@ -7,4 +7,4 @@ spec:
     - name: web
       protocol: TCP
       port: 8080
-      targetPort: 8080
+      targetPort: web

run-llm-load-test/llm-load-test/Dockerfile → llm-load-test/Containerfile

@@ -2,6 +2,8 @@ FROM python:3.12-slim
 
 WORKDIR /app/
 
-COPY . ./
+COPY requirements.txt ./
 
 RUN pip install -r requirements.txt
+
+COPY . ./