| 1 |
# Licensed under the Apache License, Version 2.0 (the "License"); you may |
| 2 |
# not use this file except in compliance with the License. You may obtain |
| 3 |
# a copy of the License at |
| 4 |
# |
| 5 |
# http://www.apache.org/licenses/LICENSE-2.0 |
| 6 |
# |
| 7 |
# Unless required by applicable law or agreed to in writing, software |
| 8 |
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT |
| 9 |
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the |
| 10 |
# License for the specific language governing permissions and limitations |
| 11 |
# under the License. |
| 12 |
import enum |
| 13 |
import re |
| 14 |
|
| 15 |
from magnum.api import utils as api_utils |
| 16 |
from magnum.common import clients |
| 17 |
from magnum.common import exception |
| 18 |
from magnum.common import neutron |
| 19 |
from magnum.common import short_id |
| 20 |
from magnum.drivers.common import driver |
| 21 |
from magnum.objects import fields |
| 22 |
from oslo_log import log as logging |
| 23 |
from oslo_utils import strutils |
| 24 |
from oslo_utils import uuidutils |
| 25 |
|
| 26 |
from magnum_capi_helm.common import app_creds |
| 27 |
from magnum_capi_helm.common import ca_certificates |
| 28 |
from magnum_capi_helm.common import capi_monitor |
| 29 |
from magnum_capi_helm import conf |
| 30 |
from magnum_capi_helm import driver_utils |
| 31 |
from magnum_capi_helm import helm |
| 32 |
from magnum_capi_helm import kubernetes |
| 33 |
|
| 34 |
LOG = logging.getLogger(__name__) |
| 35 |
CONF = conf.CONF |
| 36 |
NODE_GROUP_ROLE_CONTROLLER = "master" |
| 37 |
|
| 38 |
|
| 39 |
class NodeGroupState(enum.Enum): |
| 40 |
NOT_PRESENT = 1 |
| 41 |
PENDING = 2 |
| 42 |
READY = 3 |
| 43 |
FAILED = 4 |
| 44 |
|
| 45 |
|
| 46 |
class Driver(driver.Driver): |
| 47 |
def __init__(self): |
| 48 |
self._helm_client = helm.Client() |
| 49 |
self.__k8s_client = None |
| 50 |
|
| 51 |
@property |
| 52 |
def _k8s_client(self): |
| 53 |
if not self.__k8s_client: |
| 54 |
self.__k8s_client = kubernetes.Client.load() |
| 55 |
return self.__k8s_client |
| 56 |
|
| 57 |
@property |
| 58 |
def provides(self): |
| 59 |
return [ |
| 60 |
{ |
| 61 |
"server_type": "vm", |
| 62 |
# NOTE(johngarbutt) we don't depend on a specific OS, |
| 63 |
# we depend on kubeadm images with cloud-init |
| 64 |
"os": "ubuntu", |
| 65 |
"coe": "kubernetes", |
| 66 |
}, |
| 67 |
{ |
| 68 |
"server_type": "vm", |
| 69 |
"os": "flatcar", |
| 70 |
"coe": "kubernetes", |
| 71 |
}, |
| 72 |
] |
| 73 |
|
| 74 |
def _update_control_plane_nodegroup_status(self, cluster, nodegroup): |
| 75 |
# The status of the master nodegroup is determined by the Cluster API |
| 76 |
# control plane object |
| 77 |
kcp = self._k8s_client.get_kubeadm_control_plane( |
| 78 |
driver_utils.get_k8s_resource_name(cluster, "control-plane"), |
| 79 |
driver_utils.cluster_namespace(cluster), |
| 80 |
) |
| 81 |
|
| 82 |
ng_state = NodeGroupState.NOT_PRESENT |
| 83 |
if kcp: |
| 84 |
ng_state = NodeGroupState.PENDING |
| 85 |
|
| 86 |
kcp_spec = kcp.get("spec", {}) if kcp else {} |
| 87 |
kcp_status = kcp.get("status", {}) if kcp else {} |
| 88 |
|
| 89 |
# The control plane object is what controls the Kubernetes version |
| 90 |
# If it is known, report it |
| 91 |
kube_version = kcp_status.get("version", kcp_spec.get("version")) |
| 92 |
if cluster.coe_version != kube_version: |
| 93 |
cluster.coe_version = kube_version |
| 94 |
cluster.save() |
| 95 |
|
| 96 |
kcp_true_conditions = { |
| 97 |
cond["type"] |
| 98 |
for cond in kcp_status.get("conditions", []) |
| 99 |
if cond["status"] == "True" |
| 100 |
} |
| 101 |
kcp_ready = all( |
| 102 |
cond in kcp_true_conditions |
| 103 |
for cond in ( |
| 104 |
"MachinesReady", |
| 105 |
"Ready", |
| 106 |
"EtcdClusterHealthy", |
| 107 |
"ControlPlaneComponentsHealthy", |
| 108 |
) |
| 109 |
) |
| 110 |
target_replicas = kcp_spec.get("replicas") |
| 111 |
current_replicas = kcp_status.get("replicas") |
| 112 |
updated_replicas = kcp_status.get("updatedReplicas") |
| 113 |
ready_replicas = kcp_status.get("readyReplicas") |
| 114 |
if ( |
| 115 |
kcp_ready |
| 116 |
and target_replicas == current_replicas |
| 117 |
and current_replicas == updated_replicas |
| 118 |
and updated_replicas == ready_replicas |
| 119 |
): |
| 120 |
ng_state = NodeGroupState.READY |
| 121 |
|
| 122 |
# TODO(mkjpryor) Work out a way to determine FAILED state |
| 123 |
return self._update_nodegroup_status(cluster, nodegroup, ng_state) |
| 124 |
|
| 125 |
def _update_worker_nodegroup_status(self, cluster, nodegroup): |
| 126 |
# The status of a worker nodegroup is determined by the corresponding |
| 127 |
# Cluster API machine deployment |
| 128 |
md = self._k8s_client.get_machine_deployment( |
| 129 |
driver_utils.get_k8s_resource_name(cluster, nodegroup.name), |
| 130 |
driver_utils.cluster_namespace(cluster), |
| 131 |
) |
| 132 |
|
| 133 |
ng_state = NodeGroupState.NOT_PRESENT |
| 134 |
if md: |
| 135 |
ng_state = NodeGroupState.PENDING |
| 136 |
|
| 137 |
# When a machine deployment is deleted, it disappears straight |
| 138 |
# away even when there are still machines belonging to it that |
| 139 |
# are deleting |
| 140 |
# In that case, we want to keep the nodegroup as DELETE_IN_PROGRESS |
| 141 |
# until all the machines for the node group are gone |
| 142 |
if ( |
| 143 |
not md |
| 144 |
and nodegroup.status.startswith("DELETE_") |
| 145 |
and self._nodegroup_machines_exist(cluster, nodegroup) |
| 146 |
): |
| 147 |
LOG.debug( |
| 148 |
f"Node group {nodegroup.name} " |
| 149 |
f"for cluster {cluster.uuid} " |
| 150 |
"machine deployment gone, but machines still found." |
| 151 |
) |
| 152 |
ng_state = NodeGroupState.PENDING |
| 153 |
|
| 154 |
md_status = md.get("status", {}) if md else {} |
| 155 |
md_phase = md_status.get("phase") |
| 156 |
if md_phase: |
| 157 |
if md_phase == "Running": |
| 158 |
ng_state = NodeGroupState.READY |
| 159 |
elif md_phase in {"Failed", "Unknown"}: |
| 160 |
ng_state = NodeGroupState.FAILED |
| 161 |
|
| 162 |
return self._update_nodegroup_status(cluster, nodegroup, ng_state) |
| 163 |
|
| 164 |
def _update_nodegroup_status(self, cluster, nodegroup, ng_state): |
| 165 |
# For delete we are waiting for not present |
| 166 |
if nodegroup.status.startswith("DELETE_"): |
| 167 |
if ng_state == NodeGroupState.NOT_PRESENT: |
| 168 |
if not nodegroup.is_default: |
| 169 |
# Conductor will delete default nodegroups |
| 170 |
# when cluster is deleted, but non default |
| 171 |
# node groups should be deleted here. |
| 172 |
nodegroup.destroy() |
| 173 |
LOG.debug( |
| 174 |
f"Node group deleted: {nodegroup.name} " |
| 175 |
f"for cluster {cluster.uuid} " |
| 176 |
f"which is_default: {nodegroup.is_default}" |
| 177 |
) |
| 178 |
# signal the node group has been deleted |
| 179 |
return None |
| 180 |
|
| 181 |
LOG.debug( |
| 182 |
f"Node group not yet delete: {nodegroup.name} " |
| 183 |
f"for cluster {cluster.uuid}" |
| 184 |
) |
| 185 |
return nodegroup |
| 186 |
|
| 187 |
is_update_operation = nodegroup.status.startswith("UPDATE_") |
| 188 |
is_create_operation = nodegroup.status.startswith("CREATE_") |
| 189 |
if not is_update_operation and not is_create_operation: |
| 190 |
LOG.warning( |
| 191 |
f"Node group: {nodegroup.name} in unexpected " |
| 192 |
f"state: {nodegroup.status} in cluster {cluster.uuid}" |
| 193 |
) |
| 194 |
elif ng_state == NodeGroupState.READY: |
| 195 |
nodegroup.status = ( |
| 196 |
fields.ClusterStatus.UPDATE_COMPLETE |
| 197 |
if is_update_operation |
| 198 |
else fields.ClusterStatus.CREATE_COMPLETE |
| 199 |
) |
| 200 |
LOG.debug( |
| 201 |
f"Node group ready: {nodegroup.name} " |
| 202 |
f"in cluster {cluster.uuid}" |
| 203 |
) |
| 204 |
nodegroup.save() |
| 205 |
|
| 206 |
elif ng_state == NodeGroupState.FAILED: |
| 207 |
nodegroup.status = ( |
| 208 |
fields.ClusterStatus.UPDATE_FAILED |
| 209 |
if is_update_operation |
| 210 |
else fields.ClusterStatus.CREATE_FAILED |
| 211 |
) |
| 212 |
LOG.debug( |
| 213 |
f"Node group failed: {nodegroup.name} " |
| 214 |
f"in cluster {cluster.uuid}" |
| 215 |
) |
| 216 |
nodegroup.save() |
| 217 |
elif ng_state == NodeGroupState.NOT_PRESENT: |
| 218 |
LOG.debug( |
| 219 |
f"Node group not yet found: {nodegroup.name} " |
| 220 |
f"state:{nodegroup.status} in cluster {cluster.uuid}" |
| 221 |
) |
| 222 |
else: |
| 223 |
LOG.debug( |
| 224 |
f"Node group still pending: {nodegroup.name} " |
| 225 |
f"state:{nodegroup.status} in cluster {cluster.uuid}" |
| 226 |
) |
| 227 |
|
| 228 |
return nodegroup |
| 229 |
|
| 230 |
def _nodegroup_machines_exist(self, cluster, nodegroup): |
| 231 |
cluster_name = driver_utils.chart_release_name(cluster) |
| 232 |
nodegroup_name = driver_utils.sanitized_name(nodegroup.name) |
| 233 |
machines = self._k8s_client.get_all_machines_by_label( |
| 234 |
{ |
| 235 |
"capi.stackhpc.com/cluster": cluster_name, |
| 236 |
"capi.stackhpc.com/component": "worker", |
| 237 |
"capi.stackhpc.com/node-group": nodegroup_name, |
| 238 |
}, |
| 239 |
driver_utils.cluster_namespace(cluster), |
| 240 |
) |
| 241 |
return bool(machines) |
| 242 |
|
| 243 |
def _update_cluster_api_address(self, cluster, capi_cluster): |
| 244 |
# As soon as we know the API address, we should set it |
| 245 |
# This means users can access the API even if the create is |
| 246 |
# not complete, which could be useful for debugging failures, |
| 247 |
# e.g. with addons |
| 248 |
if not capi_cluster: |
| 249 |
# skip update if cluster not yet created |
| 250 |
return |
| 251 |
|
| 252 |
if cluster.status not in [ |
| 253 |
fields.ClusterStatus.CREATE_IN_PROGRESS, |
| 254 |
fields.ClusterStatus.UPDATE_IN_PROGRESS, |
| 255 |
]: |
| 256 |
# only update api-address when updating or creating |
| 257 |
return |
| 258 |
|
| 259 |
api_endpoint = capi_cluster["spec"].get("controlPlaneEndpoint") |
| 260 |
if api_endpoint: |
| 261 |
api_address = ( |
| 262 |
f"https://{api_endpoint['host']}:{api_endpoint['port']}" |
| 263 |
) |
| 264 |
if cluster.api_address != api_address: |
| 265 |
cluster.api_address = api_address |
| 266 |
cluster.save() |
| 267 |
LOG.debug(f"Found api_address for {cluster.uuid}") |
| 268 |
|
| 269 |
def _update_status_updating(self, cluster, capi_cluster): |
| 270 |
# If the cluster is not yet ready then the create/update |
| 271 |
# is still in progress |
| 272 |
true_conditions = { |
| 273 |
cond["type"] |
| 274 |
for cond in capi_cluster.get("status", {}).get("conditions", []) |
| 275 |
if cond["status"] == "True" |
| 276 |
} |
| 277 |
for cond in ("InfrastructureReady", "ControlPlaneReady", "Ready"): |
| 278 |
if cond not in true_conditions: |
| 279 |
return |
| 280 |
|
| 281 |
is_update_operation = cluster.status.startswith("UPDATE_") |
| 282 |
|
| 283 |
# Check the status of the addons |
| 284 |
addons = self._k8s_client.get_addons_by_label( |
| 285 |
{ |
| 286 |
"addons.stackhpc.com/cluster": driver_utils.chart_release_name( |
| 287 |
cluster |
| 288 |
) |
| 289 |
}, |
| 290 |
driver_utils.cluster_namespace(cluster), |
| 291 |
) |
| 292 |
for addon in addons: |
| 293 |
addon_phase = addon.get("status", {}).get("phase") |
| 294 |
if addon_phase and addon_phase in {"Failed", "Unknown"}: |
| 295 |
# If the addon is failed, mark the cluster as failed |
| 296 |
cluster.status = ( |
| 297 |
fields.ClusterStatus.UPDATE_FAILED |
| 298 |
if is_update_operation |
| 299 |
else fields.ClusterStatus.CREATE_FAILED |
| 300 |
) |
| 301 |
cluster.save() |
| 302 |
return |
| 303 |
elif addon_phase and addon_phase == "Deployed": |
| 304 |
# If the addon is deployed, move on to the next one |
| 305 |
continue |
| 306 |
else: |
| 307 |
# If there are any addons that are not deployed or failed, |
| 308 |
# wait for the next invocation to check again |
| 309 |
LOG.debug( |
| 310 |
f"addon {addon['metadata']['name']} not yet deployed " |
| 311 |
f"for {cluster.uuid}" |
| 312 |
) |
| 313 |
return |
| 314 |
|
| 315 |
# If we get this far, the cluster has completed successfully |
| 316 |
cluster.status = ( |
| 317 |
fields.ClusterStatus.UPDATE_COMPLETE |
| 318 |
if is_update_operation |
| 319 |
else fields.ClusterStatus.CREATE_COMPLETE |
| 320 |
) |
| 321 |
cluster.save() |
| 322 |
|
| 323 |
def _update_status_deleting(self, context, cluster): |
| 324 |
# Once the Cluster API cluster is gone, we need to clean up |
| 325 |
# the secrets we created |
| 326 |
self._k8s_client.delete_all_secrets_by_label( |
| 327 |
"magnum.openstack.org/cluster-uuid", |
| 328 |
cluster.uuid, |
| 329 |
driver_utils.cluster_namespace(cluster), |
| 330 |
) |
| 331 |
|
| 332 |
# We also need to clean up the appcred that we made |
| 333 |
app_creds.delete_app_cred(context, cluster) |
| 334 |
|
| 335 |
cluster.status = fields.ClusterStatus.DELETE_COMPLETE |
| 336 |
cluster.save() |
| 337 |
|
| 338 |
def _get_capi_cluster(self, cluster): |
| 339 |
release_name = driver_utils.chart_release_name(cluster) |
| 340 |
if release_name: |
| 341 |
return self._k8s_client.get_capi_cluster( |
| 342 |
release_name, |
| 343 |
driver_utils.cluster_namespace(cluster), |
| 344 |
) |
| 345 |
|
| 346 |
def _update_all_nodegroups_status(self, cluster): |
| 347 |
"""Returns True if any node group still in progress.""" |
| 348 |
nodegroups = [] |
| 349 |
for nodegroup in cluster.nodegroups: |
| 350 |
if nodegroup.role == NODE_GROUP_ROLE_CONTROLLER: |
| 351 |
updated_nodegroup = ( |
| 352 |
self._update_control_plane_nodegroup_status( |
| 353 |
cluster, nodegroup |
| 354 |
) |
| 355 |
) |
| 356 |
else: |
| 357 |
updated_nodegroup = self._update_worker_nodegroup_status( |
| 358 |
cluster, nodegroup |
| 359 |
) |
| 360 |
if updated_nodegroup: |
| 361 |
nodegroups.append(updated_nodegroup) |
| 362 |
|
| 363 |
# Return True if any are still in progress |
| 364 |
for nodegroup in nodegroups: |
| 365 |
if nodegroup.status.endswith("_IN_PROGRESS"): |
| 366 |
return True |
| 367 |
return False |
| 368 |
|
| 369 |
def update_cluster_status(self, context, cluster): |
| 370 |
# NOTE(mkjpryor) |
| 371 |
# Because Kubernetes operators are built around reconciliation loops, |
| 372 |
# Cluster API clusters don't really go into an error state |
| 373 |
# Hence we only currently handle transitioning from IN_PROGRESS |
| 374 |
# states to COMPLETE |
| 375 |
|
| 376 |
# TODO(mkjpryor) Add a timeout for create/update/delete |
| 377 |
|
| 378 |
capi_cluster = self._get_capi_cluster(cluster) |
| 379 |
|
| 380 |
if capi_cluster: |
| 381 |
# Update the cluster API address if it is known |
| 382 |
# so users can get their coe credentials |
| 383 |
self._update_cluster_api_address(cluster, capi_cluster) |
| 384 |
|
| 385 |
# Update the nodegroups first |
| 386 |
# to ensure API never returns an inconsistent state |
| 387 |
nodegroups_in_progress = self._update_all_nodegroups_status( |
| 388 |
cluster |
| 389 |
) |
| 390 |
|
| 391 |
if cluster.status in { |
| 392 |
fields.ClusterStatus.CREATE_IN_PROGRESS, |
| 393 |
fields.ClusterStatus.UPDATE_IN_PROGRESS, |
| 394 |
}: |
| 395 |
LOG.debug("Checking on an update for %s", cluster.uuid) |
| 396 |
# If the cluster does not exist yet, |
| 397 |
# create is still in progress |
| 398 |
if not capi_cluster: |
| 399 |
LOG.debug(f"capi_cluster not yet created for {cluster.uuid}") |
| 400 |
return |
| 401 |
if nodegroups_in_progress: |
| 402 |
LOG.debug(f"Node groups are not all ready for {cluster.uuid}") |
| 403 |
return |
| 404 |
self._update_status_updating(cluster, capi_cluster) |
| 405 |
|
| 406 |
elif cluster.status == fields.ClusterStatus.DELETE_IN_PROGRESS: |
| 407 |
LOG.debug("Checking on a delete for %s", cluster.uuid) |
| 408 |
# If the Cluster API cluster still exists, |
| 409 |
# the delete is still in progress |
| 410 |
if capi_cluster: |
| 411 |
LOG.debug(f"capi_cluster still found for {cluster.uuid}") |
| 412 |
return |
| 413 |
self._update_status_deleting(context, cluster) |
| 414 |
|
| 415 |
def get_monitor(self, context, cluster): |
| 416 |
return capi_monitor.CAPIMonitor(context, cluster) |
| 417 |
|
| 418 |
def _k8s_resource_labels(self, cluster): |
| 419 |
# TODO(johngarbutt) need to check these are safe labels |
| 420 |
name = driver_utils.chart_release_name(cluster) |
| 421 |
return { |
| 422 |
"magnum.openstack.org/project-id": cluster.project_id[:63], |
| 423 |
"magnum.openstack.org/user-id": cluster.user_id[:63], |
| 424 |
"magnum.openstack.org/cluster-uuid": cluster.uuid[:63], |
| 425 |
"cluster.x-k8s.io/cluster-name": name, |
| 426 |
} |
| 427 |
|
| 428 |
def _create_appcred_secret(self, context, cluster): |
| 429 |
string_data = app_creds.get_app_cred_string_data(context, cluster) |
| 430 |
name = self._get_app_cred_name(cluster) |
| 431 |
self._k8s_client.apply_secret( |
| 432 |
name, |
| 433 |
{ |
| 434 |
"metadata": {"labels": self._k8s_resource_labels(cluster)}, |
| 435 |
"stringData": string_data, |
| 436 |
}, |
| 437 |
driver_utils.cluster_namespace(cluster), |
| 438 |
) |
| 439 |
|
| 440 |
def _ensure_certificate_secrets(self, context, cluster): |
| 441 |
# Magnum creates CA certs for each of the Kubernetes components that |
| 442 |
# must be trusted by the cluster |
| 443 |
# In particular, this is required for "openstack coe cluster config" |
| 444 |
# to work, as that doesn't communicate with the driver and instead |
| 445 |
# relies on the correct CA being trusted by the cluster |
| 446 |
|
| 447 |
# Cluster API looks for specific named secrets for each of the CAs, |
| 448 |
# and generates them if they don't exist, so we create them here |
| 449 |
# with the correct certificates in |
| 450 |
for ( |
| 451 |
name, |
| 452 |
data, |
| 453 |
) in ca_certificates.get_certificate_string_data( |
| 454 |
context, cluster |
| 455 |
).items(): |
| 456 |
self._k8s_client.apply_secret( |
| 457 |
driver_utils.get_k8s_resource_name(cluster, name), |
| 458 |
{ |
| 459 |
"metadata": {"labels": self._k8s_resource_labels(cluster)}, |
| 460 |
"type": "cluster.x-k8s.io/secret", |
| 461 |
"stringData": data, |
| 462 |
}, |
| 463 |
driver_utils.cluster_namespace(cluster), |
| 464 |
) |
| 465 |
|
| 466 |
def _label(self, cluster, key, default): |
| 467 |
all_labels = helm.mergeconcat( |
| 468 |
cluster.cluster_template.labels, cluster.labels |
| 469 |
) |
| 470 |
if not all_labels: |
| 471 |
return default |
| 472 |
raw = all_labels.get(key, default) |
| 473 |
# NOTE(johngarbutt): filtering untrusted user input |
| 474 |
return re.sub(r"[^a-zA-Z0-9\.\-\/ _]+", "", raw) |
| 475 |
|
| 476 |
def _get_label_bool(self, cluster, label, default): |
| 477 |
cluster_label = self._label(cluster, label, "") |
| 478 |
return strutils.bool_from_string(cluster_label, default=default) |
| 479 |
|
| 480 |
def _get_label_int(self, cluster, label, default): |
| 481 |
cluster_label = self._label(cluster, label, "") |
| 482 |
if not cluster_label: |
| 483 |
return default |
| 484 |
try: |
| 485 |
return int(cluster_label) |
| 486 |
except ValueError: |
| 487 |
return default |
| 488 |
|
| 489 |
def _get_chart_version(self, cluster): |
| 490 |
version = cluster.cluster_template.labels.get( |
| 491 |
"capi_helm_chart_version", |
| 492 |
CONF.capi_helm.default_helm_chart_version, |
| 493 |
) |
| 494 |
# NOTE(johngarbutt): filtering untrusted user input |
| 495 |
return re.sub(r"[^a-z0-9\.\-\+]+", "", version) |
| 496 |
|
| 497 |
def _get_kube_version(self, image): |
| 498 |
# The image should have a property containing the Kubernetes version |
| 499 |
kube_version = image.get("kube_version") |
| 500 |
if not kube_version: |
| 501 |
raise exception.MagnumException( |
| 502 |
message=f"Image {image.id} does not " |
| 503 |
"have a kube_version property." |
| 504 |
) |
| 505 |
raw = kube_version.lstrip("v") |
| 506 |
# TODO(johngarbutt) more validation required? |
| 507 |
return re.sub(r"[^0-9\.]+", "", raw) |
| 508 |
|
| 509 |
def _get_os_distro(self, image): |
| 510 |
os_distro = image.get("os_distro") |
| 511 |
if not os_distro: |
| 512 |
raise exception.MagnumException( |
| 513 |
message=f"Image {image.id} does not " |
| 514 |
"have an os_distro property." |
| 515 |
) |
| 516 |
return re.sub(r"[^a-zA-Z0-9\.\-\/ ]+", "", os_distro) |
| 517 |
|
| 518 |
def _get_image_details(self, context, image_identifier): |
| 519 |
osc = clients.OpenStackClients(context) |
| 520 |
image = api_utils.get_openstack_resource( |
| 521 |
osc.glance().images, image_identifier, "images" |
| 522 |
) |
| 523 |
return ( |
| 524 |
image.id, |
| 525 |
self._get_kube_version(image), |
| 526 |
self._get_os_distro(image), |
| 527 |
) |
| 528 |
|
| 529 |
def _get_app_cred_name(self, cluster): |
| 530 |
return driver_utils.get_k8s_resource_name(cluster, "cloud-credentials") |
| 531 |
|
| 532 |
def _get_etcd_config(self, cluster): |
| 533 |
# Support new-style and legacy labels for volume size and type, with |
| 534 |
# new-style labels taking precedence |
| 535 |
etcd_size = self._get_label_int( |
| 536 |
cluster, |
| 537 |
"etcd_blockdevice_size", |
| 538 |
self._get_label_int(cluster, "etcd_volume_size", 0), |
| 539 |
) |
| 540 |
if etcd_size > 0: |
| 541 |
etcd_block_device = {"size": etcd_size} |
| 542 |
# The block device type can be either local or volume |
| 543 |
etcd_bd_type = self._label( |
| 544 |
cluster, "etcd_blockdevice_type", "volume" |
| 545 |
) |
| 546 |
if etcd_bd_type == "local": |
| 547 |
etcd_block_device["type"] = "Local" |
| 548 |
else: |
| 549 |
etcd_block_device["type"] = "Volume" |
| 550 |
|
| 551 |
etcd_volume_type = self._label( |
| 552 |
cluster, |
| 553 |
"etcd_blockdevice_volume_type", |
| 554 |
self._label(cluster, "etcd_volume_type", ""), |
| 555 |
) |
| 556 |
if etcd_volume_type: |
| 557 |
etcd_block_device["volumeType"] = etcd_volume_type |
| 558 |
|
| 559 |
etcd_volume_az = self._label( |
| 560 |
cluster, "etcd_blockdevice_volume_az", "" |
| 561 |
) |
| 562 |
if etcd_volume_az: |
| 563 |
etcd_block_device["availabilityZone"] = etcd_volume_az |
| 564 |
return {"blockDevice": etcd_block_device} |
| 565 |
else: |
| 566 |
return {} |
| 567 |
|
| 568 |
def _get_dns_nameservers(self, cluster): |
| 569 |
dns_nameserver = cluster.cluster_template.dns_nameserver |
| 570 |
if dns_nameserver: |
| 571 |
return strutils.split_by_commas(dns_nameserver) |
| 572 |
else: |
| 573 |
return None |
| 574 |
|
| 575 |
def _get_monitoring_enabled(self, cluster): |
| 576 |
# NOTE(mkjpryor) default off, like heat driver, |
| 577 |
# as requires cinder and takes a while |
| 578 |
return self._get_label_bool(cluster, "monitoring_enabled", False) |
| 579 |
|
| 580 |
def _get_kube_dash_enabled(self, cluster): |
| 581 |
# NOTE(mkjpryor) default on, like the heat driver |
| 582 |
return self._get_label_bool(cluster, "kube_dashboard_enabled", True) |
| 583 |
|
| 584 |
def _get_autoheal_enabled(self, cluster): |
| 585 |
return self._get_label_bool(cluster, "auto_healing_enabled", True) |
| 586 |
|
| 587 |
def _get_autoscale_enabled(self, cluster): |
| 588 |
return self._get_label_bool(cluster, "auto_scaling_enabled", False) |
| 589 |
|
| 590 |
def _get_autoscale_values(self, cluster, nodegroup): |
| 591 |
auto_scale = self._get_autoscale_enabled(cluster) |
| 592 |
min_nodes, max_nodes = self._validate_allowed_node_counts( |
| 593 |
cluster, nodegroup |
| 594 |
) |
| 595 |
auto_scale_args = {} |
| 596 |
if auto_scale and min_nodes != max_nodes: |
| 597 |
auto_scale_args["autoscale"] = "true" |
| 598 |
auto_scale_args["machineCountMin"] = min_nodes |
| 599 |
auto_scale_args["machineCountMax"] = max_nodes |
| 600 |
return auto_scale_args |
| 601 |
|
| 602 |
def _get_k8s_keystone_auth_enabled(self, cluster): |
| 603 |
return self._get_label_bool(cluster, "keystone_auth_enabled", False) |
| 604 |
|
| 605 |
def _get_fixed_network_id(self, context, cluster): |
| 606 |
network = cluster.fixed_network |
| 607 |
if not network: |
| 608 |
return |
| 609 |
if network and uuidutils.is_uuid_like(network): |
| 610 |
return network |
| 611 |
else: |
| 612 |
return neutron.get_network( |
| 613 |
context, network, source="name", target="id", external=False |
| 614 |
) |
| 615 |
|
| 616 |
def _validate_allowed_flavor(self, context, requested_flavor): |
| 617 |
# Compare requested flavor with allowed for Kubernetes node |
| 618 |
flavors = ( |
| 619 |
clients.OpenStackClients(context) |
| 620 |
.nova() |
| 621 |
.flavors.list(min_ram=CONF.capi_helm.minimum_flavor_ram) |
| 622 |
) |
| 623 |
for flavor in flavors: |
| 624 |
vcpus = flavor.vcpus |
| 625 |
LOG.debug( |
| 626 |
f"Checking if {requested_flavor} matches " |
| 627 |
f"{flavor.id} or {flavor.name}" |
| 628 |
) |
| 629 |
if requested_flavor in [flavor.id, flavor.name]: |
| 630 |
if vcpus < CONF.capi_helm.minimum_flavor_vcpus: |
| 631 |
raise exception.MagnumException( |
| 632 |
message=f"Flavor {requested_flavor} does not " |
| 633 |
f"have enough CPU to run Kubernetes. " |
| 634 |
f"Minimum {CONF.capi_helm.minimum_flavor_vcpus} " |
| 635 |
"vcpus required." |
| 636 |
) |
| 637 |
return |
| 638 |
raise exception.MagnumException( |
| 639 |
message=f"Flavor {requested_flavor} does not " |
| 640 |
f"have enough RAM to run Kubernetes. " |
| 641 |
f"Minimum {CONF.capi_helm.minimum_flavor_ram} MB required." |
| 642 |
) |
| 643 |
|
| 644 |
def _is_default_worker_nodegroup(self, cluster, nodegroup): |
| 645 |
return cluster.default_ng_worker.id == nodegroup.id |
| 646 |
|
| 647 |
def _get_node_counts(self, cluster, nodegroup): |
| 648 |
|
| 649 |
# NOTE(scott): In CAPI MachineDeployment resources created by the |
| 650 |
# capi-helm-charts, the `replicas` field is omitted when autoscaling |
| 651 |
# is enabled (since we're relinquishing control over node count to |
| 652 |
# the autoscaler) so if a user creates a nodegroup where only |
| 653 |
# node_count is provided and min/max are not, we need to be careful |
| 654 |
# about setting the default min/max values to equal node_count. |
| 655 |
min_nodes = nodegroup.node_count |
| 656 |
max_nodes = nodegroup.node_count |
| 657 |
|
| 658 |
if nodegroup.min_node_count is not None: |
| 659 |
min_nodes = nodegroup.min_node_count |
| 660 |
if nodegroup.max_node_count is not None: |
| 661 |
max_nodes = nodegroup.max_node_count |
| 662 |
|
| 663 |
# If min/max node counts are not defined on the default |
| 664 |
# worker group then fall back to equivalent cluster labels |
| 665 |
if self._is_default_worker_nodegroup(cluster, nodegroup): |
| 666 |
# NOTE(scott): Magnum seems to set min_node_count = 1 |
| 667 |
# on the default group so treat this as if it were None |
| 668 |
if nodegroup.min_node_count == 1: |
| 669 |
min_nodes = nodegroup.node_count |
| 670 |
|
| 671 |
# We still want to be able to override the default node |
| 672 |
# group values with labels for consistent behaviour with |
| 673 |
# Magnum Heat driver. |
| 674 |
min_nodes = self._get_label_int( |
| 675 |
cluster, "min_node_count", min_nodes |
| 676 |
) |
| 677 |
max_nodes = self._get_label_int( |
| 678 |
cluster, "max_node_count", max_nodes |
| 679 |
) |
| 680 |
|
| 681 |
return min_nodes, max_nodes |
| 682 |
|
| 683 |
def _validate_allowed_node_counts(self, cluster, nodegroup): |
| 684 |
min_nodes, max_nodes = self._get_node_counts(cluster, nodegroup) |
| 685 |
|
| 686 |
LOG.debug( |
| 687 |
f"Checking if node group {nodegroup.name} has valid " |
| 688 |
f"node count parameters (count, min, max) = " |
| 689 |
f"{(nodegroup.node_count, min_nodes, max_nodes)}" |
| 690 |
) |
| 691 |
|
| 692 |
if min_nodes is not None: |
| 693 |
# ClusterAPI Provider OpenStack (CAPO) |
| 694 |
# doesn't support scale to zero yet. |
| 695 |
if min_nodes < 1: |
| 696 |
raise exception.NodeGroupInvalidInput( |
| 697 |
message="Min node count must be greater than " |
| 698 |
"or equal to 1 for all node groups." |
| 699 |
) |
| 700 |
if min_nodes > nodegroup.node_count: |
| 701 |
raise exception.NodeGroupInvalidInput( |
| 702 |
message="Min node count must be less than " |
| 703 |
"or equal to current node count" |
| 704 |
) |
| 705 |
if max_nodes is not None and max_nodes < min_nodes: |
| 706 |
raise exception.NodeGroupInvalidInput( |
| 707 |
message="Max node count must be greater than " |
| 708 |
"or equal to min node count" |
| 709 |
) |
| 710 |
|
| 711 |
return min_nodes, max_nodes |
| 712 |
|
| 713 |
def _get_csi_cinder_availability_zone(self, cluster): |
| 714 |
return self._label( |
| 715 |
cluster, |
| 716 |
"csi_cinder_availability_zone", |
| 717 |
CONF.capi_helm.csi_cinder_availability_zone, |
| 718 |
) |
| 719 |
|
| 720 |
def _get_csi_cinder_reclaim_policy(self, cluster): |
| 721 |
return self._label( |
| 722 |
cluster, |
| 723 |
"csi_cinder_reclaim_policy", |
| 724 |
CONF.capi_helm.csi_cinder_reclaim_policy, |
| 725 |
) |
| 726 |
|
| 727 |
def _get_csi_cinder_fstype(self, cluster): |
| 728 |
return self._label( |
| 729 |
cluster, |
| 730 |
"csi_cinder_fstype", |
| 731 |
CONF.capi_helm.csi_cinder_fstype, |
| 732 |
) |
| 733 |
|
| 734 |
def _get_csi_cinder_allow_volume_expansion(self, cluster): |
| 735 |
return self._get_label_bool( |
| 736 |
cluster, |
| 737 |
"csi_cinder_allow_volume_expansion", |
| 738 |
CONF.capi_helm.csi_cinder_allow_volume_expansion, |
| 739 |
) |
| 740 |
|
| 741 |
def _get_octavia_provider(self, cluster): |
| 742 |
return self._label(cluster, "octavia_provider", "amphora") |
| 743 |
|
| 744 |
def _get_octavia_lb_algorithm(self, cluster): |
| 745 |
provider = self._get_octavia_provider(cluster) |
| 746 |
return self._label( |
| 747 |
cluster, |
| 748 |
"octavia_lb_algorithm", |
| 749 |
"SOURCE_IP_PORT" if provider.lower() == "ovn" else "ROUND_ROBIN", |
| 750 |
) |
| 751 |
|
| 752 |
def _get_allowed_cidrs(self, cluster): |
| 753 |
cidr_list = cluster.labels.get("api_master_lb_allowed_cidrs", "") |
| 754 |
LOG.debug(f"CIDR list {cidr_list}") |
| 755 |
if isinstance(cidr_list, str) and cidr_list != "": |
| 756 |
return cidr_list.split(",") |
| 757 |
return False |
| 758 |
|
| 759 |
def _storageclass_definitions(self, context, cluster): |
| 760 |
"""Query cinder API to retrieve list of available volume types. |
| 761 |
|
| 762 |
@return dict(dict,list(dict)) containing storage classes |
| 763 |
""" |
| 764 |
LOG.debug("Retrieve volume types from cinder for StorageClasses.") |
| 765 |
client = clients.OpenStackClients(context) |
| 766 |
availability_zone = self._get_csi_cinder_availability_zone(cluster) |
| 767 |
c_client = client.cinder() |
| 768 |
volume_types = [i.name for i in c_client.volume_types.list()] |
| 769 |
# Use the default volume type if defined. Otherwise use the first |
| 770 |
# type returned by cinder. |
| 771 |
default_volume_type = CONF.capi_helm.csi_cinder_default_volume_type |
| 772 |
LOG.debug( |
| 773 |
f"Default volume type: {default_volume_type}" |
| 774 |
f" Volume types: {volume_types}" |
| 775 |
) |
| 776 |
if not default_volume_type: |
| 777 |
default_volume_type = volume_types[0] |
| 778 |
LOG.warning( |
| 779 |
f"Default volume type not defined." |
| 780 |
f" Using {default_volume_type}." |
| 781 |
) |
| 782 |
elif default_volume_type not in volume_types: |
| 783 |
# If default does not exist throw an error. |
| 784 |
raise exception.MagnumException( |
| 785 |
message=f"{default_volume_type} is not a" |
| 786 |
" valid Cinder volume type." |
| 787 |
) |
| 788 |
default_storage_class = {} |
| 789 |
additional_storage_classes = [] |
| 790 |
allow_expansion = self._get_csi_cinder_allow_volume_expansion(cluster) |
| 791 |
reclaim_policy = self._get_csi_cinder_reclaim_policy(cluster) |
| 792 |
allowed_topologies = CONF.capi_helm.csi_cinder_allowed_topologies |
| 793 |
fstype = self._get_csi_cinder_fstype(cluster) |
| 794 |
|
| 795 |
for volume_type in volume_types: |
| 796 |
storage_class = { |
| 797 |
"name": driver_utils.sanitized_name(volume_type), |
| 798 |
"reclaimPolicy": reclaim_policy, |
| 799 |
"allowVolumeExpansion": allow_expansion, |
| 800 |
"availabilityZone": availability_zone, |
| 801 |
"volumeType": volume_type, |
| 802 |
"allowedTopologies": allowed_topologies, |
| 803 |
"fstype": fstype, |
| 804 |
"enabled": True, |
| 805 |
} |
| 806 |
if volume_type == default_volume_type: |
| 807 |
default_storage_class = storage_class |
| 808 |
else: |
| 809 |
additional_storage_classes.append(storage_class) |
| 810 |
return dict( |
| 811 |
defaultStorageClass=default_storage_class, |
| 812 |
additionalStorageClasses=additional_storage_classes, |
| 813 |
) |
| 814 |
|
| 815 |
def _process_node_groups(self, cluster): |
| 816 |
nodegroups = cluster.nodegroups |
| 817 |
nodegroup_set = [] |
| 818 |
for ng in nodegroups: |
| 819 |
if ng.role != NODE_GROUP_ROLE_CONTROLLER: |
| 820 |
nodegroup_item = dict( |
| 821 |
name=driver_utils.sanitized_name(ng.name), |
| 822 |
machineFlavor=ng.flavor_id, |
| 823 |
machineCount=ng.node_count, |
| 824 |
) |
| 825 |
if self._get_autoscale_enabled(cluster): |
| 826 |
values = self._get_autoscale_values(cluster, ng) |
| 827 |
nodegroup_item = helm.mergeconcat(nodegroup_item, values) |
| 828 |
nodegroup_set.append(nodegroup_item) |
| 829 |
return nodegroup_set |
| 830 |
|
| 831 |
def _update_helm_release(self, context, cluster, nodegroups=None): |
| 832 |
if nodegroups is None: |
| 833 |
nodegroups = cluster.nodegroups |
| 834 |
|
| 835 |
image_id, kube_version, os_distro = self._get_image_details( |
| 836 |
context, cluster.cluster_template.image_id |
| 837 |
) |
| 838 |
|
| 839 |
network_id = self._get_fixed_network_id(context, cluster) |
| 840 |
subnet_id = neutron.get_fixed_subnet_id(context, cluster.fixed_subnet) |
| 841 |
|
| 842 |
values = { |
| 843 |
"kubernetesVersion": kube_version, |
| 844 |
"machineImageId": image_id, |
| 845 |
"machineSSHKeyName": cluster.keypair or None, |
| 846 |
"cloudCredentialsSecretName": self._get_app_cred_name(cluster), |
| 847 |
"etcd": self._get_etcd_config(cluster), |
| 848 |
"apiServer": { |
| 849 |
"enableLoadBalancer": True, |
| 850 |
"loadBalancerProvider": self._get_octavia_provider(cluster), |
| 851 |
}, |
| 852 |
"clusterNetworking": { |
| 853 |
"dnsNameservers": self._get_dns_nameservers(cluster), |
| 854 |
"externalNetworkId": neutron.get_external_network_id( |
| 855 |
context, cluster.cluster_template.external_network_id |
| 856 |
), |
| 857 |
"internalNetwork": { |
| 858 |
"networkFilter": ( |
| 859 |
{"id": network_id} if network_id else None |
| 860 |
), |
| 861 |
"subnetFilter": ({"id": subnet_id} if subnet_id else None), |
| 862 |
# This is only used if a fixed network is not specified |
| 863 |
"nodeCidr": self._label( |
| 864 |
cluster, "fixed_subnet_cidr", "10.0.0.0/24" |
| 865 |
), |
| 866 |
}, |
| 867 |
}, |
| 868 |
"osDistro": os_distro, |
| 869 |
"controlPlane": { |
| 870 |
"machineFlavor": cluster.master_flavor_id, |
| 871 |
"machineCount": cluster.master_count, |
| 872 |
"healthCheck": { |
| 873 |
"enabled": self._get_autoheal_enabled(cluster), |
| 874 |
}, |
| 875 |
}, |
| 876 |
"nodeGroupDefaults": { |
| 877 |
"healthCheck": { |
| 878 |
"enabled": self._get_autoheal_enabled(cluster), |
| 879 |
}, |
| 880 |
}, |
| 881 |
"nodeGroups": self._process_node_groups(cluster), |
| 882 |
"addons": { |
| 883 |
"openstack": { |
| 884 |
"csiCinder": self._storageclass_definitions( |
| 885 |
context, cluster |
| 886 |
), |
| 887 |
"cloudConfig": { |
| 888 |
"LoadBalancer": { |
| 889 |
"lb-provider": self._get_octavia_provider(cluster), |
| 890 |
"lb-method": self._get_octavia_lb_algorithm( |
| 891 |
cluster |
| 892 |
), |
| 893 |
"create-monitor": self._get_label_bool( |
| 894 |
cluster, "octavia_lb_healthcheck", True |
| 895 |
), |
| 896 |
} |
| 897 |
}, |
| 898 |
}, |
| 899 |
"monitoring": { |
| 900 |
"enabled": self._get_monitoring_enabled(cluster) |
| 901 |
}, |
| 902 |
"kubernetesDashboard": { |
| 903 |
"enabled": self._get_kube_dash_enabled(cluster) |
| 904 |
}, |
| 905 |
# TODO(mkjpryor): can't enable ingress until code exists to |
| 906 |
# remove the load balancer |
| 907 |
"ingress": {"enabled": False}, |
| 908 |
}, |
| 909 |
} |
| 910 |
|
| 911 |
# Add boot disk details, if defined in config file. |
| 912 |
# Helm chart defaults to ephemeral disks, if unset. |
| 913 |
boot_volume_type = self._label( |
| 914 |
cluster, "boot_volume_type", CONF.cinder.default_boot_volume_type |
| 915 |
) |
| 916 |
if boot_volume_type: |
| 917 |
disk_type_details = { |
| 918 |
"controlPlane": { |
| 919 |
"machineRootVolume": { |
| 920 |
"volumeType": boot_volume_type, |
| 921 |
} |
| 922 |
}, |
| 923 |
"nodeGroupDefaults": { |
| 924 |
"machineRootVolume": { |
| 925 |
"volumeType": boot_volume_type, |
| 926 |
} |
| 927 |
}, |
| 928 |
} |
| 929 |
values = helm.mergeconcat(values, disk_type_details) |
| 930 |
|
| 931 |
boot_volume_size_gb = self._get_label_int( |
| 932 |
cluster, "boot_volume_size", CONF.cinder.default_boot_volume_size |
| 933 |
) |
| 934 |
if boot_volume_size_gb: |
| 935 |
disk_size_details = { |
| 936 |
"controlPlane": { |
| 937 |
"machineRootVolume": { |
| 938 |
"diskSize": boot_volume_size_gb, |
| 939 |
} |
| 940 |
}, |
| 941 |
"nodeGroupDefaults": { |
| 942 |
"machineRootVolume": { |
| 943 |
"diskSize": boot_volume_size_gb, |
| 944 |
} |
| 945 |
}, |
| 946 |
} |
| 947 |
values = helm.mergeconcat(values, disk_size_details) |
| 948 |
|
| 949 |
# Sometimes you need to add an extra network |
| 950 |
# for things like Cinder CSI CephFS Native |
| 951 |
extra_network_name = self._label(cluster, "extra_network_name", "") |
| 952 |
if extra_network_name: |
| 953 |
network_details = { |
| 954 |
"nodeGroupDefaults": { |
| 955 |
"machineNetworking": { |
| 956 |
"ports": [ |
| 957 |
{}, |
| 958 |
{ |
| 959 |
"network": { |
| 960 |
"name": extra_network_name, |
| 961 |
}, |
| 962 |
"securityGroups": [], |
| 963 |
}, |
| 964 |
], |
| 965 |
}, |
| 966 |
}, |
| 967 |
} |
| 968 |
values = helm.mergeconcat(values, network_details) |
| 969 |
|
| 970 |
if self._get_k8s_keystone_auth_enabled(cluster): |
| 971 |
k8s_keystone_auth_config = { |
| 972 |
"authWebhook": "k8s-keystone-auth", |
| 973 |
"addons": { |
| 974 |
"openstack": { |
| 975 |
"k8sKeystoneAuth": { # addon subchart configuration |
| 976 |
"enabled": True, |
| 977 |
"values": { |
| 978 |
"openstackAuthUrl": context.auth_url, |
| 979 |
"projectId": context.project_id, |
| 980 |
}, |
| 981 |
} |
| 982 |
} |
| 983 |
}, |
| 984 |
} |
| 985 |
values = helm.mergeconcat(values, k8s_keystone_auth_config) |
| 986 |
LOG.debug( |
| 987 |
"Enable K8s keystone auth webhook for" |
| 988 |
f" project: {context.project_id} auth url: {context.auth_url}" |
| 989 |
) |
| 990 |
|
| 991 |
api_lb_allowed_cidrs = self._get_allowed_cidrs(cluster) |
| 992 |
if isinstance(api_lb_allowed_cidrs, list): |
| 993 |
allowed_cidrs_config = { |
| 994 |
"apiServer": {"allowedCidrs": api_lb_allowed_cidrs} |
| 995 |
} |
| 996 |
values = helm.mergeconcat(values, allowed_cidrs_config) |
| 997 |
|
| 998 |
self._helm_client.install_or_upgrade( |
| 999 |
driver_utils.chart_release_name(cluster), |
| 1000 |
CONF.capi_helm.helm_chart_name, |
| 1001 |
values, |
| 1002 |
repo=CONF.capi_helm.helm_chart_repo, |
| 1003 |
version=self._get_chart_version(cluster), |
| 1004 |
namespace=driver_utils.cluster_namespace(cluster), |
| 1005 |
) |
| 1006 |
|
| 1007 |
def _generate_release_name(self, cluster): |
| 1008 |
if cluster.stack_id: |
| 1009 |
return |
| 1010 |
|
| 1011 |
# Make sure no duplicate names |
| 1012 |
# by generating 12 character random id |
| 1013 |
random_bit = short_id.generate_id() |
| 1014 |
base_name = driver_utils.sanitized_name(cluster.name) |
| 1015 |
# valid release names are 53 chars long |
| 1016 |
# and stack_id is 12 characters |
| 1017 |
# but we also use this to derive hostnames |
| 1018 |
trimmed_name = base_name[:30] |
| 1019 |
# Save the full name, so users can rename in the API |
| 1020 |
cluster.stack_id = f"{trimmed_name}-{random_bit}".lower() |
| 1021 |
# be sure to save this before we use it |
| 1022 |
cluster.save() |
| 1023 |
|
| 1024 |
def create_cluster(self, context, cluster, cluster_create_timeout): |
| 1025 |
LOG.info("Starting to create cluster %s", cluster.uuid) |
| 1026 |
|
| 1027 |
self._validate_allowed_flavor(context, cluster.master_flavor_id) |
| 1028 |
nodegroups = cluster.nodegroups |
| 1029 |
for ng in nodegroups: |
| 1030 |
self._validate_allowed_flavor(context, ng.flavor_id) |
| 1031 |
# we generate this name (on the initial create call only) |
| 1032 |
# so we hit no issues with duplicate cluster names |
| 1033 |
# and it makes renaming clusters in the API possible |
| 1034 |
self._generate_release_name(cluster) |
| 1035 |
|
| 1036 |
# NOTE(johngarbutt) all node groups should already |
| 1037 |
# be in the CREATE_IN_PROGRESS state |
| 1038 |
self._k8s_client.ensure_namespace( |
| 1039 |
driver_utils.cluster_namespace(cluster) |
| 1040 |
) |
| 1041 |
self._create_appcred_secret(context, cluster) |
| 1042 |
self._ensure_certificate_secrets(context, cluster) |
| 1043 |
|
| 1044 |
self._update_helm_release(context, cluster) |
| 1045 |
|
| 1046 |
def update_cluster( |
| 1047 |
self, context, cluster, scale_manager=None, rollback=False |
| 1048 |
): |
| 1049 |
# Cluster API refuses to update things like cluster networking, |
| 1050 |
# so it is safest not to implement this for now |
| 1051 |
# TODO(mkjpryor) Check what bits of update we can support |
| 1052 |
raise NotImplementedError( |
| 1053 |
"Updating a cluster in this way is not currently supported" |
| 1054 |
) |
| 1055 |
|
| 1056 |
def delete_cluster(self, context, cluster): |
| 1057 |
LOG.info("Starting to delete cluster %s", cluster.uuid) |
| 1058 |
|
| 1059 |
# Copy the helm driver by marking all node groups |
| 1060 |
# as delete in progress here, as note done by conductor |
| 1061 |
# We do this before calling uninstall_release because |
| 1062 |
# update_cluster_status can get called before we return |
| 1063 |
for ng in cluster.nodegroups: |
| 1064 |
ng.status = fields.ClusterStatus.DELETE_IN_PROGRESS |
| 1065 |
ng.save() |
| 1066 |
|
| 1067 |
release_name = driver_utils.chart_release_name(cluster) |
| 1068 |
# Only attempt deletion of CAPI resources if they were created in |
| 1069 |
# the first place e.g. if trust creation fails during cluster create |
| 1070 |
# then no CAPI resources will have been created. |
| 1071 |
if release_name: |
| 1072 |
# Begin the deletion of the cluster resources by uninstalling the |
| 1073 |
# Helm release. |
| 1074 |
# Note that this just marks the resources for deletion, |
| 1075 |
# it does not wait for the resources to be deleted. |
| 1076 |
self._helm_client.uninstall_release( |
| 1077 |
release_name, |
| 1078 |
namespace=driver_utils.cluster_namespace(cluster), |
| 1079 |
) |
| 1080 |
|
| 1081 |
def resize_cluster( |
| 1082 |
self, |
| 1083 |
context, |
| 1084 |
cluster, |
| 1085 |
resize_manager, |
| 1086 |
node_count, |
| 1087 |
nodes_to_remove, |
| 1088 |
nodegroup=None, |
| 1089 |
): |
| 1090 |
if nodes_to_remove: |
| 1091 |
LOG.warning("Removing specific nodes is not currently supported") |
| 1092 |
self._update_helm_release(context, cluster) |
| 1093 |
|
| 1094 |
def upgrade_cluster( |
| 1095 |
self, |
| 1096 |
context, |
| 1097 |
cluster, |
| 1098 |
cluster_template, |
| 1099 |
max_batch_size, |
| 1100 |
nodegroup, |
| 1101 |
scale_manager=None, |
| 1102 |
rollback=False, |
| 1103 |
): |
| 1104 |
# TODO(mkjpryor) check that the upgrade is viable |
| 1105 |
# e.g. not a downgrade, not an upgrade by more than one minor version |
| 1106 |
|
| 1107 |
# Updating the template will likely apply for all nodegroups |
| 1108 |
# So mark them all as having an update in progress |
| 1109 |
for nodegroup in cluster.nodegroups: |
| 1110 |
nodegroup.status = fields.ClusterStatus.UPDATE_IN_PROGRESS |
| 1111 |
self._validate_allowed_flavor(context, nodegroup.flavor_id) |
| 1112 |
nodegroup.save() |
| 1113 |
|
| 1114 |
# Move the cluster to the new template |
| 1115 |
cluster.cluster_template_id = cluster_template.uuid |
| 1116 |
cluster.status = fields.ClusterStatus.UPDATE_IN_PROGRESS |
| 1117 |
cluster.save() |
| 1118 |
cluster.refresh() |
| 1119 |
|
| 1120 |
self._update_helm_release(context, cluster) |
| 1121 |
|
| 1122 |
def create_nodegroup(self, context, cluster, nodegroup): |
| 1123 |
nodegroup.status = fields.ClusterStatus.CREATE_IN_PROGRESS |
| 1124 |
self._validate_allowed_flavor(context, nodegroup.flavor_id) |
| 1125 |
if self._get_autoscale_enabled(cluster): |
| 1126 |
self._validate_allowed_node_counts(cluster, nodegroup) |
| 1127 |
nodegroup.save() |
| 1128 |
|
| 1129 |
self._update_helm_release(context, cluster) |
| 1130 |
|
| 1131 |
def update_nodegroup(self, context, cluster, nodegroup): |
| 1132 |
nodegroup.status = fields.ClusterStatus.UPDATE_IN_PROGRESS |
| 1133 |
self._validate_allowed_flavor(context, nodegroup.flavor_id) |
| 1134 |
if self._get_autoscale_enabled(cluster): |
| 1135 |
self._validate_allowed_node_counts(cluster, nodegroup) |
| 1136 |
nodegroup.save() |
| 1137 |
|
| 1138 |
self._update_helm_release(context, cluster) |
| 1139 |
|
| 1140 |
def delete_nodegroup(self, context, cluster, nodegroup): |
| 1141 |
nodegroup.status = fields.ClusterStatus.DELETE_IN_PROGRESS |
| 1142 |
nodegroup.save() |
| 1143 |
|
| 1144 |
# Remove the nodegroup being deleted from the nodegroups |
| 1145 |
# for the Helm release |
| 1146 |
self._update_helm_release( |
| 1147 |
context, |
| 1148 |
cluster, |
| 1149 |
[ng for ng in cluster.nodegroups if ng.name != nodegroup.name], |
| 1150 |
) |
| 1151 |
|
| 1152 |
def create_federation(self, context, federation): |
| 1153 |
raise NotImplementedError("Will not implement 'create_federation'") |
| 1154 |
|
| 1155 |
def update_federation(self, context, federation): |
| 1156 |
raise NotImplementedError("Will not implement 'update_federation'") |
| 1157 |
|
| 1158 |
def delete_federation(self, context, federation): |
| 1159 |
raise NotImplementedError("Will not implement 'delete_federation'") |