|
1 | | -#!/bin/bash |
| 1 | +#!/usr/bin/env bash |
2 | 2 |
|
3 | 3 | # Copyright 2025 The Kubernetes Authors. |
4 | 4 | # |
|
14 | 14 | # See the License for the specific language governing permissions and |
15 | 15 | # limitations under the License. |
16 | 16 |
|
17 | | -# This script verifies end-to-end connectivity for an example inference extension test environment based on |
18 | | -# resources from the quickstart guide or e2e test framework. It can optionally launch a "curl" client pod to |
19 | | -# run these tests within the cluster. |
20 | | -# |
21 | | -# USAGE: ./hack/e2e-test.sh |
22 | | -# |
23 | | -# OPTIONAL ENVIRONMENT VARIABLES: |
24 | | -# - TIME: The duration (in seconds) for which the test will run. Defaults to 1 second. |
25 | | -# - CURL_POD: If set to "true", the script will use a Kubernetes pod named "curl" for making requests. |
26 | | -# - IP: Override the detected IP address. If not provided, the script attempts to use a Gateway based on |
27 | | -# the quickstart guide or an Envoy service IP based on the e2e test framework. |
28 | | -# - PORT: Override the detected port. If not provided, the script attempts to use a Gateway based on the |
29 | | -# quickstart guide or an Envoy service IP based on the e2e test framework. |
30 | | -# |
31 | | -# WHAT THE SCRIPT DOES: |
32 | | -# 1. Determines if there is a Gateway named "inference-gateway" in the "default" namespace. If found, it extracts the IP |
33 | | -# address and port from the Gateway's "llm-gw" listener. Otherwise, it falls back to the Envoy service in the "default" namespace. |
34 | | -# 2. Optionally checks for (or creates) a "curl" pod, ensuring it is ready to execute requests. |
35 | | -# 3. Loops for $TIME seconds, sending requests every 5 seconds to the /v1/completions endpoint to confirm successful connectivity. |
36 | | - |
37 | | -set -euo pipefail |
38 | | - |
39 | | -# Determine the directory of this script and build an absolute path to client.yaml. |
40 | | -SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )" |
41 | | -CLIENT_YAML="$SCRIPT_DIR/../test/testdata/client.yaml" |
42 | | - |
43 | | -# TIME is the amount of time, in seconds, to run the test. |
44 | | -TIME=${TIME:-1} |
45 | | -# Optionally use a client curl pod for executing the curl command. |
46 | | -CURL_POD=${CURL_POD:-false} |
47 | | - |
48 | | -check_resource_exists() { |
49 | | - local type=$1 |
50 | | - local name=$2 |
51 | | - local namespace=$3 |
52 | | - |
53 | | - if kubectl get "$type" "$name" -n "$namespace" &>/dev/null; then |
54 | | - return 0 |
55 | | - else |
56 | | - return 1 |
57 | | - fi |
| 17 | +set -euox pipefail |
| 18 | + |
| 19 | +install_kind() { |
| 20 | + if ! command -v kind &>/dev/null; then |
| 21 | + echo "kind not found, installing..." |
| 22 | + [ $(uname -m) = x86_64 ] && curl -Lo ./kind https://kind.sigs.k8s.io/dl/v0.29.0/kind-linux-amd64 |
| 23 | + # For ARM64 |
| 24 | + [ $(uname -m) = aarch64 ] && curl -Lo ./kind https://kind.sigs.k8s.io/dl/v0.29.0/kind-linux-arm64 |
| 25 | + chmod +x ./kind |
| 26 | + mv ./kind /usr/local/bin/kind |
| 27 | + else |
| 28 | + echo "kind is already installed." |
| 29 | + fi |
58 | 30 | } |
59 | 31 |
|
60 | | -check_pod_ready() { |
61 | | - local pod_name=$1 |
62 | | - local namespace=$2 |
63 | | - # Check the Ready condition using jsonpath. Default to False if not found. |
64 | | - local ready_status |
65 | | - ready_status=$(kubectl get pod "$pod_name" -n "$namespace" -o jsonpath='{.status.conditions[?(@.type=="Ready")].status}' 2>/dev/null || echo "False") |
66 | | - if [[ "$ready_status" == "True" ]]; then |
67 | | - return 0 |
68 | | - else |
69 | | - return 1 |
70 | | - fi |
71 | | -} |
72 | | - |
73 | | -# Try to get the Gateway's IP and the port from the listener named "llm-gw" if it exists. |
74 | | -if check_resource_exists "gateway" "inference-gateway" "default"; then |
75 | | - GATEWAY_IP=$(kubectl get gateway inference-gateway -n default -o jsonpath='{.status.addresses[0].value}') |
76 | | - # Use JSONPath to select the port from the listener with name "http" |
77 | | - GATEWAY_PORT=$(kubectl get gateway inference-gateway -n default -o jsonpath='{.spec.listeners[?(@.name=="http")].port}') |
78 | | -else |
79 | | - GATEWAY_IP="" |
80 | | - GATEWAY_PORT="" |
| 32 | +if [ "$USE_KIND" = "true" ]; then |
| 33 | + install_kind # make sure kind cli is installed |
| 34 | + if ! kubectl config current-context >/dev/null 2>&1; then # if no active kind cluster found |
| 35 | + echo "No active kubecontext found. creating a kind cluster for running the tests..." |
| 36 | + kind create cluster --name inference-e2e |
| 37 | + KIND_CLUSTER=inference-e2e IMAGE_TAG=${E2E_IMAGE} make image-kind |
| 38 | + else |
| 39 | + current_context=$(kubectl config current-context) |
| 40 | + current_kind_cluster="${current_context#kind-}" |
| 41 | + echo "Found an active kind cluster ${current_kind_cluster} for running the tests..." |
| 42 | + KIND_CLUSTER=${current_kind_cluster} IMAGE_TAG=${E2E_IMAGE} make image-kind |
| 43 | + fi |
| 44 | +else |
| 45 | + # don't use kind. it's the caller responsibility to load the image into the cluster, we just run the tests. |
| 46 | + # this section is useful when one wants to run an official release or latest main against a cluster other than kind. |
| 47 | + if ! kubectl config current-context >/dev/null 2>&1; then # if no active cluster found |
| 48 | + echo "No active kubecontext found. exiting..." |
| 49 | + exit |
| 50 | + fi |
81 | 51 | fi |
82 | 52 |
|
83 | | -if [[ -n "$GATEWAY_IP" && -n "$GATEWAY_PORT" ]]; then |
84 | | - echo "Using Gateway inference-gateway IP and port from listener 'llm-gw'." |
85 | | - IP=${IP:-$GATEWAY_IP} |
86 | | - PORT=${PORT:-$GATEWAY_PORT} |
87 | | -else |
88 | | - echo "Gateway inference-gateway not found or missing IP/port. Falling back to Envoy service." |
89 | | - # Ensure the Envoy service exists. |
90 | | - if ! check_resource_exists "svc" "envoy" "default"; then |
91 | | - echo "Error: Envoy service not found in namespace 'default'." |
92 | | - exit 1 |
93 | | - fi |
94 | | - IP=${IP:-$(kubectl get svc envoy -n default -o jsonpath='{.spec.clusterIP}')} |
95 | | - PORT=${PORT:-$(kubectl get svc envoy -n default -o jsonpath='{.spec.ports[0].port}')} |
96 | | -fi |
97 | | - |
98 | | -# Optionally verify that the curl pod exists and is ready. |
99 | | -if [[ "$CURL_POD" == "true" ]]; then |
100 | | - if ! check_resource_exists "pod" "curl" "default"; then |
101 | | - echo "Pod 'curl' not found in namespace 'default'. Applying client.yaml from $CLIENT_YAML..." |
102 | | - kubectl apply -f "$CLIENT_YAML" |
103 | | - fi |
104 | | - echo "Waiting for pod 'curl' to be ready..." |
105 | | - # Retry every 5 seconds for up to 30 seconds (6 attempts) |
106 | | - for i in {1..6}; do |
107 | | - if check_pod_ready "curl" "default"; then |
108 | | - echo "Pod 'curl' is now ready." |
109 | | - break |
110 | | - fi |
111 | | - echo "Retry attempt $i: Pod 'curl' not ready; waiting 5 seconds..." |
112 | | - sleep 5 |
113 | | - done |
114 | | - |
115 | | - if ! check_pod_ready "curl" "default"; then |
116 | | - echo "Error: Pod 'curl' is still not ready in namespace 'default' after 30 seconds." |
117 | | - exit 1 |
118 | | - fi |
119 | | -fi |
120 | | - |
121 | | -# Validate that we have a non-empty IP and PORT. |
122 | | -if [[ -z "$IP" ]]; then |
123 | | - echo "Error: Unable to determine a valid IP from either Gateway or Envoy service." |
124 | | - exit 1 |
125 | | -fi |
126 | | - |
127 | | -if [[ -z "$PORT" ]]; then |
128 | | - echo "Error: Unable to determine a valid port from either Gateway or Envoy service." |
129 | | - exit 1 |
130 | | -fi |
131 | | - |
132 | | -echo "Using IP: $IP" |
133 | | -echo "Using PORT: $PORT" |
134 | | - |
135 | | -# Run the test for the specified duration. |
136 | | -end=$((SECONDS + TIME)) |
137 | | -if [[ "$CURL_POD" == "true" ]]; then |
138 | | - while [ $SECONDS -lt $end ]; do |
139 | | - kubectl exec po/curl -- curl -i "$IP:$PORT/v1/completions" \ |
140 | | - -H 'Content-Type: application/json' \ |
141 | | - -d '{"model": "food-review","prompt": "Write as if you were a critic: San Francisco","max_tokens": 100,"temperature": 0}' |
142 | | - sleep 5 |
143 | | - done |
144 | | -else |
145 | | - while [ $SECONDS -lt $end ]; do |
146 | | - curl -i "$IP:$PORT/v1/completions" \ |
147 | | - -H 'Content-Type: application/json' \ |
148 | | - -d '{"model": "food-review","prompt": "Write as if you were a critic: San Francisco","max_tokens": 100,"temperature": 0}' |
149 | | - sleep 5 |
150 | | - done |
151 | | -fi |
| 53 | +echo "Found an active cluster. Running Go e2e tests in ./epp..." |
| 54 | +go test ./test/e2e/epp/ -v -ginkgo.v |
0 commit comments