|
| 1 | +//! Ensures that `Pod`s are configured and running for each [`v1alpha2::AirflowCluster`] |
| 2 | +//! |
| 3 | +//! Pipeline architecture: dereference -> validate -> build -> apply -> update_status |
| 4 | +
|
| 5 | +pub mod apply; |
| 6 | +pub mod build; |
| 7 | +pub mod dereference; |
| 8 | +pub mod update_status; |
| 9 | +pub mod validate; |
| 10 | + |
| 11 | +// --------------------------------------------------------------------------- |
| 12 | +// Imports |
| 13 | +// --------------------------------------------------------------------------- |
| 14 | + |
| 15 | +use std::{ |
| 16 | + collections::BTreeMap, |
| 17 | + marker::PhantomData, |
| 18 | + sync::Arc, |
| 19 | +}; |
| 20 | + |
| 21 | +use const_format::concatcp; |
| 22 | +use product_config::ProductConfigManager; |
| 23 | +use snafu::{ResultExt, Snafu}; |
| 24 | +use stackable_operator::{ |
| 25 | + cli::OperatorEnvironmentOptions, |
| 26 | + cluster_resources::{ClusterResourceApplyStrategy, ClusterResources}, |
| 27 | + commons::{ |
| 28 | + affinity::StackableAffinity, |
| 29 | + product_image_selection::ResolvedProductImage, |
| 30 | + resources::{NoRuntimeLimits, Resources}, |
| 31 | + }, |
| 32 | + crd::listener, |
| 33 | + k8s_openapi::api::{ |
| 34 | + apps::v1::StatefulSet, |
| 35 | + core::v1::{ |
| 36 | + ConfigMap, Container as K8sContainer, EnvVar, PersistentVolumeClaim, PodTemplateSpec, |
| 37 | + Service, ServiceAccount, Volume, VolumeMount, |
| 38 | + }, |
| 39 | + policy::v1::PodDisruptionBudget, |
| 40 | + rbac::v1::RoleBinding, |
| 41 | + }, |
| 42 | + kube::{ |
| 43 | + Resource, |
| 44 | + api::ObjectMeta, |
| 45 | + core::{DeserializeGuard, error_boundary}, |
| 46 | + runtime::{controller::Action, reflector::ObjectRef}, |
| 47 | + }, |
| 48 | + logging::controller::ReconcilerError, |
| 49 | + product_logging::spec::ContainerLogConfig, |
| 50 | + role_utils::RoleGroupRef, |
| 51 | + shared::time::Duration, |
| 52 | +}; |
| 53 | +use strum::{EnumDiscriminants, IntoStaticStr}; |
| 54 | + |
| 55 | +use crate::{ |
| 56 | + crd::{ |
| 57 | + APP_NAME, AirflowExecutor, AirflowRole, AirflowStorageConfig, OPERATOR_NAME, |
| 58 | + v1alpha2, |
| 59 | + }, |
| 60 | + framework::{ |
| 61 | + HasName, HasUid, NameIsValidLabelValue, |
| 62 | + product_logging::framework::{ValidatedContainerLogConfigChoice, VectorContainerLogConfig}, |
| 63 | + types::{ |
| 64 | + kubernetes::{NamespaceName, Uid}, |
| 65 | + operator::ClusterName, |
| 66 | + }, |
| 67 | + }, |
| 68 | +}; |
| 69 | + |
| 70 | +use self::{ |
| 71 | + apply::Applier, |
| 72 | + build::build, |
| 73 | + dereference::dereference, |
| 74 | + update_status::update_status, |
| 75 | + validate::validate_cluster, |
| 76 | +}; |
| 77 | + |
| 78 | +// --------------------------------------------------------------------------- |
| 79 | +// Constants and context |
| 80 | +// --------------------------------------------------------------------------- |
| 81 | + |
| 82 | +pub const AIRFLOW_CONTROLLER_NAME: &str = "airflowcluster"; |
| 83 | +pub const CONTAINER_IMAGE_BASE_NAME: &str = "airflow"; |
| 84 | +pub const AIRFLOW_FULL_CONTROLLER_NAME: &str = |
| 85 | + concatcp!(AIRFLOW_CONTROLLER_NAME, '.', OPERATOR_NAME); |
| 86 | + |
| 87 | +pub struct Ctx { |
| 88 | + pub client: stackable_operator::client::Client, |
| 89 | + pub product_config: ProductConfigManager, |
| 90 | + pub operator_environment: OperatorEnvironmentOptions, |
| 91 | +} |
| 92 | + |
| 93 | +// --------------------------------------------------------------------------- |
| 94 | +// Validated types |
| 95 | +// --------------------------------------------------------------------------- |
| 96 | + |
| 97 | +pub(crate) struct Prepared; |
| 98 | +pub(crate) struct Applied; |
| 99 | + |
| 100 | +pub(crate) struct KubernetesResources<T> { |
| 101 | + pub stateful_sets: Vec<StatefulSet>, |
| 102 | + pub config_maps: Vec<ConfigMap>, |
| 103 | + pub services: Vec<Service>, |
| 104 | + pub service_accounts: Vec<ServiceAccount>, |
| 105 | + pub role_bindings: Vec<RoleBinding>, |
| 106 | + pub pod_disruption_budgets: Vec<PodDisruptionBudget>, |
| 107 | + pub listeners: Vec<listener::v1alpha1::Listener>, |
| 108 | + pub _status: PhantomData<T>, |
| 109 | +} |
| 110 | + |
| 111 | +#[derive(Clone, Debug)] |
| 112 | +pub struct ValidatedRoleConfig { |
| 113 | + pub pdb_enabled: bool, |
| 114 | + pub pdb_max_unavailable: Option<u16>, |
| 115 | + pub listener_class: Option<String>, |
| 116 | + pub group_listener_name: Option<String>, |
| 117 | +} |
| 118 | + |
| 119 | +#[derive(Clone, Debug)] |
| 120 | +pub struct ValidatedRoleGroupConfig { |
| 121 | + pub resources: Resources<AirflowStorageConfig, NoRuntimeLimits>, |
| 122 | + pub logging: ValidatedLogging, |
| 123 | + pub affinity: StackableAffinity, |
| 124 | + pub graceful_shutdown_timeout: Duration, |
| 125 | + pub config_file_content: String, |
| 126 | +} |
| 127 | + |
| 128 | +#[derive(Clone)] |
| 129 | +pub struct PrecomputedPodData { |
| 130 | + pub env_vars: Vec<EnvVar>, |
| 131 | + pub airflow_commands: Vec<String>, |
| 132 | + pub auth_volumes: Vec<Volume>, |
| 133 | + pub auth_volume_mounts: Vec<VolumeMount>, |
| 134 | + pub extra_volumes: Vec<Volume>, |
| 135 | + pub extra_volume_mounts: Vec<VolumeMount>, |
| 136 | + pub git_sync_containers: Vec<K8sContainer>, |
| 137 | + pub git_sync_init_containers: Vec<K8sContainer>, |
| 138 | + pub git_sync_volumes: Vec<Volume>, |
| 139 | + pub git_sync_volume_mounts: Vec<VolumeMount>, |
| 140 | + pub vector_container: Option<K8sContainer>, |
| 141 | + pub service_account_name: String, |
| 142 | + pub replicas: Option<u16>, |
| 143 | + pub pod_overrides: PodTemplateSpec, |
| 144 | + pub executor: AirflowExecutor, |
| 145 | + pub executor_template_configmap_name: Option<String>, |
| 146 | + pub listener_volume_claim_template: Option<PersistentVolumeClaim>, |
| 147 | +} |
| 148 | + |
| 149 | +#[derive(Clone, Debug)] |
| 150 | +pub struct ValidatedLogging { |
| 151 | + pub airflow_container: ValidatedContainerLogConfigChoice, |
| 152 | + pub vector_container: Option<VectorContainerLogConfig>, |
| 153 | + pub git_sync_container_log_config: ContainerLogConfig, |
| 154 | +} |
| 155 | + |
| 156 | +impl ValidatedLogging { |
| 157 | + pub fn is_vector_agent_enabled(&self) -> bool { |
| 158 | + self.vector_container.is_some() |
| 159 | + } |
| 160 | +} |
| 161 | + |
| 162 | +// REVIEW: ValidatedAirflowCluster is the central validated type. It holds all data needed |
| 163 | +// by the build stage so that build() can be infallible. All optional-after-merge fields |
| 164 | +// are unwrapped during validation, and logging is pre-validated into ValidatedLogging. |
| 165 | +#[derive(Clone)] |
| 166 | +pub struct ValidatedAirflowCluster { |
| 167 | + metadata: ObjectMeta, |
| 168 | + pub image: ResolvedProductImage, |
| 169 | + pub name: ClusterName, |
| 170 | + pub namespace: NamespaceName, |
| 171 | + pub uid: Uid, |
| 172 | + pub role_groups: BTreeMap<AirflowRole, BTreeMap<String, ValidatedRoleGroupConfig>>, |
| 173 | + pub precomputed_pod_data: BTreeMap<AirflowRole, BTreeMap<String, PrecomputedPodData>>, |
| 174 | + pub executor_template_config_maps: Vec<ConfigMap>, |
| 175 | + pub role_configs: BTreeMap<AirflowRole, ValidatedRoleConfig>, |
| 176 | + pub executor: AirflowExecutor, |
| 177 | +} |
| 178 | + |
| 179 | +impl ValidatedAirflowCluster { |
| 180 | + #[allow(clippy::too_many_arguments)] |
| 181 | + pub fn new( |
| 182 | + image: ResolvedProductImage, |
| 183 | + name: ClusterName, |
| 184 | + namespace: NamespaceName, |
| 185 | + uid: Uid, |
| 186 | + role_groups: BTreeMap<AirflowRole, BTreeMap<String, ValidatedRoleGroupConfig>>, |
| 187 | + precomputed_pod_data: BTreeMap<AirflowRole, BTreeMap<String, PrecomputedPodData>>, |
| 188 | + executor_template_config_maps: Vec<ConfigMap>, |
| 189 | + role_configs: BTreeMap<AirflowRole, ValidatedRoleConfig>, |
| 190 | + executor: AirflowExecutor, |
| 191 | + ) -> Self { |
| 192 | + Self { |
| 193 | + metadata: ObjectMeta { |
| 194 | + name: Some(name.to_string()), |
| 195 | + namespace: Some(namespace.to_string()), |
| 196 | + uid: Some(uid.to_string()), |
| 197 | + ..ObjectMeta::default() |
| 198 | + }, |
| 199 | + image, |
| 200 | + name, |
| 201 | + namespace, |
| 202 | + uid, |
| 203 | + role_groups, |
| 204 | + precomputed_pod_data, |
| 205 | + executor_template_config_maps, |
| 206 | + role_configs, |
| 207 | + executor, |
| 208 | + } |
| 209 | + } |
| 210 | + |
| 211 | + pub fn rolegroup_ref(&self, role: &AirflowRole, role_group: &str) -> RoleGroupRef<Self> { |
| 212 | + RoleGroupRef { |
| 213 | + cluster: ObjectRef::from_obj(self), |
| 214 | + role: role.to_string(), |
| 215 | + role_group: role_group.to_string(), |
| 216 | + } |
| 217 | + } |
| 218 | +} |
| 219 | + |
| 220 | +impl HasName for ValidatedAirflowCluster { |
| 221 | + fn to_name(&self) -> String { |
| 222 | + self.name.to_string() |
| 223 | + } |
| 224 | +} |
| 225 | + |
| 226 | +impl HasUid for ValidatedAirflowCluster { |
| 227 | + fn to_uid(&self) -> Uid { |
| 228 | + self.uid.clone() |
| 229 | + } |
| 230 | +} |
| 231 | + |
| 232 | +impl Resource for ValidatedAirflowCluster { |
| 233 | + type DynamicType = |
| 234 | + <v1alpha2::AirflowCluster as stackable_operator::kube::Resource>::DynamicType; |
| 235 | + type Scope = <v1alpha2::AirflowCluster as stackable_operator::kube::Resource>::Scope; |
| 236 | + |
| 237 | + fn kind(dt: &Self::DynamicType) -> std::borrow::Cow<'_, str> { |
| 238 | + v1alpha2::AirflowCluster::kind(dt) |
| 239 | + } |
| 240 | + |
| 241 | + fn group(dt: &Self::DynamicType) -> std::borrow::Cow<'_, str> { |
| 242 | + v1alpha2::AirflowCluster::group(dt) |
| 243 | + } |
| 244 | + |
| 245 | + fn version(dt: &Self::DynamicType) -> std::borrow::Cow<'_, str> { |
| 246 | + v1alpha2::AirflowCluster::version(dt) |
| 247 | + } |
| 248 | + |
| 249 | + fn plural(dt: &Self::DynamicType) -> std::borrow::Cow<'_, str> { |
| 250 | + v1alpha2::AirflowCluster::plural(dt) |
| 251 | + } |
| 252 | + |
| 253 | + fn meta(&self) -> &ObjectMeta { |
| 254 | + &self.metadata |
| 255 | + } |
| 256 | + |
| 257 | + fn meta_mut(&mut self) -> &mut ObjectMeta { |
| 258 | + &mut self.metadata |
| 259 | + } |
| 260 | +} |
| 261 | + |
| 262 | +impl NameIsValidLabelValue for ValidatedAirflowCluster { |
| 263 | + fn to_label_value(&self) -> String { |
| 264 | + self.name.to_label_value() |
| 265 | + } |
| 266 | +} |
| 267 | + |
| 268 | +// --------------------------------------------------------------------------- |
| 269 | +// Error types and reconcile |
| 270 | +// --------------------------------------------------------------------------- |
| 271 | + |
| 272 | +#[derive(Snafu, Debug, EnumDiscriminants)] |
| 273 | +#[strum_discriminants(derive(IntoStaticStr))] |
| 274 | +pub enum Error { |
| 275 | + #[snafu(display("AirflowCluster object is invalid"))] |
| 276 | + InvalidAirflowCluster { |
| 277 | + source: error_boundary::InvalidObject, |
| 278 | + }, |
| 279 | + |
| 280 | + #[snafu(display("failed to dereference resources"))] |
| 281 | + Dereference { source: dereference::Error }, |
| 282 | + |
| 283 | + #[snafu(display("failed to validate cluster"))] |
| 284 | + Validate { source: validate::Error }, |
| 285 | + |
| 286 | + #[snafu(display("failed to create cluster resources"))] |
| 287 | + CreateClusterResources { |
| 288 | + source: stackable_operator::cluster_resources::Error, |
| 289 | + }, |
| 290 | + |
| 291 | + #[snafu(display("failed to apply resources"))] |
| 292 | + Apply { source: apply::Error }, |
| 293 | + |
| 294 | + #[snafu(display("failed to update status"))] |
| 295 | + UpdateStatus { source: update_status::Error }, |
| 296 | +} |
| 297 | + |
| 298 | +type Result<T, E = Error> = std::result::Result<T, E>; |
| 299 | + |
| 300 | +impl ReconcilerError for Error { |
| 301 | + fn category(&self) -> &'static str { |
| 302 | + ErrorDiscriminants::from(self).into() |
| 303 | + } |
| 304 | +} |
| 305 | + |
| 306 | +// REVIEW: The reconcile pipeline is structured as five sequential stages: |
| 307 | +// 1. dereference — async, fallible: resolve external references |
| 308 | +// 2. validate — sync, fallible: validate and merge configs |
| 309 | +// 3. build — sync, infallible: construct Kubernetes resources |
| 310 | +// 4. apply — async, fallible: apply resources to the cluster |
| 311 | +// 5. update_status — async, fallible: patch status on the CRD |
| 312 | +pub async fn reconcile( |
| 313 | + airflow: Arc<DeserializeGuard<v1alpha2::AirflowCluster>>, |
| 314 | + ctx: Arc<Ctx>, |
| 315 | +) -> Result<Action> { |
| 316 | + tracing::info!("Starting reconcile"); |
| 317 | + |
| 318 | + let airflow = airflow |
| 319 | + .0 |
| 320 | + .as_ref() |
| 321 | + .map_err(error_boundary::InvalidObject::clone) |
| 322 | + .context(InvalidAirflowClusterSnafu)?; |
| 323 | + |
| 324 | + // --- dereference (async, fallible) --- |
| 325 | + let dereferenced = dereference( |
| 326 | + &ctx.client, |
| 327 | + airflow, |
| 328 | + CONTAINER_IMAGE_BASE_NAME, |
| 329 | + &ctx.operator_environment.image_repository, |
| 330 | + crate::built_info::PKG_VERSION, |
| 331 | + ) |
| 332 | + .await |
| 333 | + .context(DereferenceSnafu)?; |
| 334 | + |
| 335 | + // --- validate (sync, fallible) --- |
| 336 | + let validated = |
| 337 | + validate_cluster(airflow, &dereferenced, &ctx.product_config).context(ValidateSnafu)?; |
| 338 | + |
| 339 | + // REVIEW: build() is infallible — all validation and fallible operations (config |
| 340 | + // generation, PodBuilder/ContainerBuilder usage, logging validation) happen in the |
| 341 | + // validate stage. The build stage purely assembles Kubernetes resource structs. |
| 342 | + // --- build (sync, infallible) --- |
| 343 | + let prepared = build(&validated); |
| 344 | + |
| 345 | + // --- apply (async, fallible) --- |
| 346 | + let cluster_resources = ClusterResources::new( |
| 347 | + APP_NAME, |
| 348 | + OPERATOR_NAME, |
| 349 | + AIRFLOW_CONTROLLER_NAME, |
| 350 | + &airflow.object_ref(&()), |
| 351 | + ClusterResourceApplyStrategy::from(&airflow.spec.cluster_operation), |
| 352 | + &airflow.spec.object_overrides, |
| 353 | + ) |
| 354 | + .context(CreateClusterResourcesSnafu)?; |
| 355 | + |
| 356 | + let applied = Applier::new(&ctx.client, cluster_resources) |
| 357 | + .apply(prepared) |
| 358 | + .await |
| 359 | + .context(ApplySnafu)?; |
| 360 | + |
| 361 | + // --- update status (async, fallible) --- |
| 362 | + update_status(&ctx.client, airflow, applied) |
| 363 | + .await |
| 364 | + .context(UpdateStatusSnafu)?; |
| 365 | + |
| 366 | + Ok(Action::await_change()) |
| 367 | +} |
| 368 | + |
| 369 | +pub fn error_policy( |
| 370 | + _obj: Arc<DeserializeGuard<v1alpha2::AirflowCluster>>, |
| 371 | + error: &Error, |
| 372 | + _ctx: Arc<Ctx>, |
| 373 | +) -> Action { |
| 374 | + match error { |
| 375 | + Error::InvalidAirflowCluster { .. } => Action::await_change(), |
| 376 | + _ => Action::requeue(*Duration::from_secs(10)), |
| 377 | + } |
| 378 | +} |
0 commit comments