mirror of
https://github.com/kubevela/kubevela.git
synced 2026-07-29 10:21:29 +00:00
d7f67354f8
* feat(workflow): add deploy-components step for per-component multi-cluster targeting Adds a pure-CUE workflow-step definition that lets each component in an Application resolve its own target cluster(s) via topology policies, filling the gap between apply-component (single component, static cluster) and deploy (all components, cross-product against policies). Signed-off-by: vishal210893 <vishal210893@gmail.com> * chore(workflow): add rendered deploy-components WorkflowStepDefinition manifest Generated from vela-templates/definitions/internal/workflowstep/deploy-components.cue via make def-install, for installation as part of the vela-core Helm chart. Signed-off-by: vishal210893 <vishal210893@gmail.com> * test(e2e): add multicluster e2e for deploy-components step Covers per-component topology-policy targeting (each component lands only on its resolved cluster, no cross-product) and fast-fail validation when a referenced component name is not present in the application. Signed-off-by: vishal210893 <vishal210893@gmail.com> * docs(workflow): add example doc for deploy-components step Required by the definition-doc CI check for any new WorkflowStepDefinition. Signed-off-by: vishal210893 <vishal210893@gmail.com> * fix(workflow): normalize deploy-components import block to grouped form The rendered manifest was generated with flat "import X" lines; the project's CUE formatter (as used by CI's make reviewable) renders multi-import blocks grouped as import (...). Fixes the check-diff CI job. Signed-off-by: vishal210893 <vishal210893@gmail.com> * fix(workflow): gate deploy-components applies on validation to prevent partial rollout Previously, _missingComponents was derived from deploy's own realized keys, which meant an invalid component name could only be detected after applies for valid components had already started -- a partial rollout on failure. Compute _missingComponents directly from the loaded component set instead (via iteration, matching the pattern deploy itself already uses reliably -- an indexed lookup against the same field was empirically found to evaluate before the load task resolves, causing false positives), and wrap deploy in an if-guard so no component is applied unless every referenced name is valid. Verified against a live multicluster setup: the negative case now applies zero resources before failing, and the positive case is unaffected. Signed-off-by: vishal210893 <vishal210893@gmail.com> * style(workflow): tighten deploy-components comments Comments narrated the debugging history behind each constraint; state them as plain engineering facts instead. Signed-off-by: vishal210893 <vishal210893@gmail.com> * fix(workflow): correct mistyped #do string in multicluster.#GetPlacementsFromTopologyPolicies The exported CUE symbol's #do value ("get-placements-from-tmulticlusterology-policies") never matched the registered provider handler key ("get-placements-from-topology-policies"), making it unusable by any CUE template. Fix the string and rename the symbol to match. Unused elsewhere in the tree, so no other callers are affected. deploy-components keeps its own local copy (for backport safety) but now documents that it mirrors the corrected symbol. Signed-off-by: vishal210893 <vishal210893@gmail.com> * refactor(workflow): use the shared topology-placement symbol in deploy-components Now that multicluster.#GetPlacementsFromTopologyPolicies is fixed, drop the local #GetPlacements duplicate and reference the shared symbol directly. deploy-components is new in this release, so there's no backport scenario where it could ship without the fix landing alongside it. Verified against a live multicluster controller build: positive case resolves placements and applies correctly, negative case still gates deploy on validation with zero partial rollout. Signed-off-by: vishal210893 <vishal210893@gmail.com> --------- Signed-off-by: vishal210893 <vishal210893@gmail.com>