Isaac Sim을 위한 Private EKS Auto Mode

set -euo pipefail
cd "$ISAAC_REPO"
isaac status > "$ISAAC_EVIDENCE/stacks.json"
jq . "$ISAAC_EVIDENCE/stacks.json"
export ISAAC_KUBECONFIG="$(isaac kubeconfig)"
test "$(ik config current-context)" = gpu-cluster-01

"$ISAAC_PYTHON" - <<'PY'
import json
from infra.isaac import deployment as d
cfg = d.config()
aws = d.session(cfg)
d.validate_network(aws, cfg)
out = d.outputs(aws, cfg)
cluster = aws.client("eks").describe_cluster(name=cfg["clusterName"])["cluster"]
assert cluster["name"] == "gpu-cluster-01"
assert cluster["arn"] == out["ClusterArn"]
assert cluster.get("tags", {}).get("DeploymentOwner") == d.OWNER
network = cluster["resourcesVpcConfig"]
assert network["vpcId"] == cfg["vpcId"]
assert set(network["subnetIds"]) == set(cfg["privateSubnetIds"])
assert network["endpointPrivateAccess"] is True
assert set(network["publicAccessCidrs"]) == set(cfg["managementCidrs"])
print(json.dumps({
    "cluster": cluster["name"], "status": cluster["status"],
    "version": cluster["version"], "privateEndpoint": network["endpointPrivateAccess"],
    "managementCidrs": network["publicAccessCidrs"], "privateSubnets": network["subnetIds"],
}, indent=2))
assert cluster["status"] == "ACTIVE"
PY

isaac plan --component cluster
jq '{stackName,changeSetType,createdAt,changes}' .local/isaac/cluster-plan.json

isaac apply --plan .local/isaac/cluster-plan.json
isaac status

export ISAAC_KUBECONFIG="$(isaac kubeconfig)"
ik get nodeclasses,nodepools -o json \
  > "$ISAAC_EVIDENCE/node-config-before.json"
jq '.items[] | {kind,name:.metadata.name,labels:.metadata.labels,spec}' \
  "$ISAAC_EVIDENCE/node-config-before.json"

isaac nodes
ik get nodeclass isaac-sim-private -o json \
  > "$ISAAC_EVIDENCE/gpu-nodeclass.json"
ik get nodepool isaac-sim-spot -o json \
  > "$ISAAC_EVIDENCE/gpu-nodepool.json"
jq '{name:.metadata.name,subnets:.spec.subnetSelectorTerms,
     publicIP:.spec.advancedNetworking.associatePublicIPAddress,
     storage:.spec.ephemeralStorage}' "$ISAAC_EVIDENCE/gpu-nodeclass.json"
jq '{requirements:.spec.template.spec.requirements,
     taints:.spec.template.spec.taints,disruption:.spec.disruption,limits:.spec.limits}' \
  "$ISAAC_EVIDENCE/gpu-nodepool.json"

isaac plan --component addons
jq '{stackName,changes}' .local/isaac/addons-plan.json

isaac apply --plan .local/isaac/addons-plan.json
isaac status
ik -n kube-system rollout status deployment/coredns --timeout=300s

isaac plan --component build
jq '{stackName,changes}' .local/isaac/build-plan.json

isaac apply --plan .local/isaac/build-plan.json
isaac status

ik -n robot-atlas-isaac get jobs,pods -o wide
ik apply --dry-run=server -f infra/isaac/gpu-probe.yaml
ik apply -f infra/isaac/gpu-probe.yaml
ik -n robot-atlas-isaac wait --for=condition=complete \
  job/isaac-gpu-probe --timeout=900s
ik -n robot-atlas-isaac logs job/isaac-gpu-probe \
  > "$ISAAC_EVIDENCE/gpu-probe.txt"
ik get nodes -l karpenter.sh/nodepool=isaac-sim-spot -o json \
  > "$ISAAC_EVIDENCE/gpu-nodes.json"
jq '.items[] | {name:.metadata.name,type:.metadata.labels["node.kubernetes.io/instance-type"],
     capacity:.metadata.labels["karpenter.sh/capacity-type"],
     providerID:.spec.providerID,addresses:.status.addresses}' \
  "$ISAAC_EVIDENCE/gpu-nodes.json"
cat "$ISAAC_EVIDENCE/gpu-probe.txt"

"$ISAAC_PYTHON" - <<'PY'
import json, os, re
from pathlib import Path
from infra.isaac import deployment as d
cfg = d.config()
aws = d.session(cfg)
nodes = json.loads((Path(os.environ["ISAAC_EVIDENCE"]) / "gpu-nodes.json").read_text())["items"]
ids = []
for node in nodes:
    assert node["metadata"]["labels"].get("karpenter.sh/nodepool") == "isaac-sim-spot"
    instance_id = node["spec"]["providerID"].rsplit("/", 1)[-1]
    assert re.fullmatch(r"i-(?:[0-9a-f]{8}|[0-9a-f]{17})", instance_id)
    ids.append(instance_id)
assert ids, "No captured GPU node: inspect a current probe, not an empty EC2 list"
rows = [i for r in aws.client("ec2").describe_instances(InstanceIds=ids)["Reservations"]
        for i in r["Instances"]]
assert {i["InstanceId"] for i in rows} == set(ids)
for instance in rows:
    assert instance["VpcId"] == cfg["vpcId"] and instance["SubnetId"] in cfg["privateSubnetIds"]
    assert not instance.get("PublicIpAddress")
    assert all(not eni.get("Association", {}).get("PublicIp")
               for eni in instance.get("NetworkInterfaces", []))
print(json.dumps([{"id": i["InstanceId"], "type": i["InstanceType"],
                   "state": i["State"]["Name"], "publicIp": i.get("PublicIpAddress")}
                  for i in rows], indent=2))
PY