buildkite/cluster-migrator

Perfil de buildkite

Última atividade de origem registrada: 8 de setembro de 2026 às 06:12

Catálogo do SkillsMP atualizado: 17 de outubro de 2026 às 00:09

skills coletadas: 1

Estrelas no GitHub: 0
Forks no GitHub: 0

README do repositório

Arquivo de origem do README: README.md

Ler o README original

Expandir README completo

# Cluster Migrator `cluster-migrator` moves Buildkite workloads from unclustered queues to a cluster without an all-at-once cutover. It gradually routes new jobs to cluster queues, compares source and destination activity, cuts over concurrency groups, checks known pipeline blockers, and permanently assigns pipelines to the destination cluster. > [!IMPORTANT] > The queue migration APIs require a Buildkite feature flag. The concurrency-group cutover and pipeline migration contracts remain provisional. Do not use this tool for a production migration until Buildkite has enabled the APIs and confirmed the operational safety gates for your organization. ## How it works A migration has three stages. Finish each stage for the whole organization before starting the next: 1. **Route queues.** Map every unclustered source queue to an existing queue with the same key in the destination cluster, then increase the percentage of new jobs sent there from 0% to 100%. All queues migrate to one destination cluster. Once a queue reaches 100%, its routing can't change. 2. **Cut over concurrency groups.** Once every queue is fully routed, cut over each concurrency group that has been used on unclustered queues: hold the group, let its running jobs finish, and move its waiting jobs to the cluster. 3. **Move pipelines.** Assess each pipeline's queue and concurrency-group blockers, then permanently assign it to the cluster. Repeat for every pipeline. Routing changes affect new jobs only. Jobs already created remain on the queue selected when they were created. A concurrency group first used after every queue it uses reaches 100% starts in the cluster, so it doesn't need a cutover. Do not move a pipeline until every concurrency group it uses is in the cluster. Pipeline readiness reports any that aren't as blockers. Concurrency groups are identified as `<scope>/<key>`: `custom/<key>` for a `concurrency_group` set in pipeline YAML, or `step/<id>` for a group Buildkite created for a step. Equal keys in different scopes are different groups. Keys may contain slashes, such as `custom/deploy/production`. ## Prerequisites Before starting: - Ask Buildkite to enable the queue migration APIs for the organization. - Enable Advanced Queue Metrics for the organization so `queue metrics` can monitor source and destination activity. - Create the destination cluster and its queues. Each destination queue must have the same exact, case-sensitive key as its source queue. - Prepare enough agent capacity in the destination cluster for the traffic you will route. - Create an API token that can access exactly one Buildkite organization with these scopes: - `read_organizations` - `read_clusters` - `write_clusters` - `read_pipelines` - `write_pipelines` Export the token: ```shell export BUILDKITE_API_TOKEN=<token> ``` The CLI discovers the organization from the token and caches the slug locally. It refreshes a stale cache entry automatically. ## Install Install the latest release with [mise](https://mise.jdx.dev/): ```shell mise use --global github:buildkite/cluster-migrator ``` Alternatively, download a binary from [GitHub Releases](https://github.com/buildkite/cluster-migrator/releases). Releases support macOS and Linux on x86-64 and ARM64. ## Migrate a workload The examples below migrate the `test` queue and the `monorepo` pipeline to the `production` cluster. Use a cluster name or ID and a pipeline slug, exact name, or ID. ### 1. Inspect current migrations ```shell cluster-migrator queue status ``` Pass a queue key to inspect one migration: ```shell cluster-migrator queue status test ``` ### 2. Configure the queue at 0% Preview the change, then create the mapping: ```shell cluster-migrator --dry-run queue configure test \ --destination-cluster production cluster-migrator queue configure test \ --destination-cluster production ``` Configuration does not route traffic. Scale the destination agents before raising the percentage. Every queue in an organization must migrate to the same cluster. Buildkite rejects configuring a queue for a different cluster from the organization's other queue migrations. ### 3. Increase routing gradually Percentages are absolute, not relative increments: ```shell cluster-migrator queue set-percent test --to 10 cluster-migrator queue metrics test cluster-migrator queue set-percent test --to 30 cluster-migrator queue metrics test cluster-migrator queue set-percent test --to 100 ``` Setting 100% is final. Once a queue reaches 100%, Buildkite rejects any change to its routing, because concurrency groups can start in the cluster from then on. Before each increase, confirm the destination has enough connected agents, waiting jobs remain controlled, and p95 wait time is acceptable. `queue metrics` compares the latest source activity with the latest and peak destination activity over the reported observation window. It provides context, not proof that a queue is safe to increase; also check your normal dispatch, queue-latency, and stranded-job observability. Routing changes take effect when jobs are created, so observed traffic may take time to reflect the new percentage. For an infrequently used queue, trigger a representative build before continuing. Once the destination is healthy, safely reduce the corresponding source-agent capacity before provisioning and routing the next increment. To stop sending new jobs to the cluster queue before it reaches 100%: ```shell cluster-migrator queue set-percent test --to 0 ``` This does not move existing jobs or remove the queue mapping. Once a queue reaches 100%, Buildkite rejects any change to its routing. ### 4. Cut over concurrency groups Start once every queue is routed 100%. Cut over every concurrency group that has been used on unclustered queues before moving any pipeline. Ask Buildkite for the list of these groups. Pipeline readiness also reports any group a pipeline still needs. Request a cutover for one group: ```shell cluster-migrator concurrency-group cutover custom/deploy \ --destination-cluster production ``` To request cutovers for a list of groups in `groups.txt`, one per line: ```shell while IFS= read -r group; do cluster-migrator concurrency-group cutover "$group" \ --destination-cluster production </dev/null done < groups.txt ``` Each cutover runs on the server, so the loop doesn't wait for them to finish. Rerun the command for any group that was rejected once its blockers are resolved. Buildkite checks the cutover before starting it, and rejects it without changes when: - The group already has a migration that is in progress, finished, or targets another cluster. - Any queue the group used in the last 10 minutes, or still has unfinished jobs on, is not routed 100% to the destination. - Ungrouped jobs were created on those unclustered queues in the last 10 minutes. Older ungrouped jobs still running there don't block the cutover, but pipeline readiness blocks their pipelines until they finish. Jobs in other concurrency groups don't block, because they only leave the queue when their own group cuts over. A rejected cutover prints its blockers and next steps and exits non-zero. Once they're resolved, run the same command again. An accepted cutover holds the group, waits for its running unclustered jobs to finish, and then moves its waiting jobs to the cluster. The command returns once the cutover starts; the rest runs on the server. Check progress at any time: ```shell cluster-migrator concurrency-group status custom/deploy ``` Concurrency-group migrations only move forward. Once a group routes to the cluster, it cannot return to unclustered queues. A forward migration cancelled before routing switches remains unclustered and can cut over again to the same destination. ### 5. Assess each pipeline After all queues used by a pipeline reach 100%, check its known blockers: ```shell cluster-migrator pipeline readiness monorepo \ --destination-cluster production ``` The assessment reports queue blockers, active source jobs, concurrency-group blockers, observation windows, and when the observations can refresh. A `no_known_blockers` result is not proof of readiness: incomplete observation windows may omit dependencies. A blocked assessment prints corrective next steps and exits non-zero. After a pipeline moves, step uploads that target a queue missing from the destination cluster will fail. Check static and dynamically generated pipeline steps for queue keys that may not appear in the readiness observation window. ### 6. Move the pipeline Previewing a move runs the readiness assessment without changing the pipeline: ```shell cluster-migrator --dry-run pipeline move monorepo \ --destination-cluster production ``` Move the pipeline only after reviewing the assessment and your operational signals: ```shell cluster-migrator pipeline move monorepo \ --destination-cluster production ``` The CLI has no pipeline rollback command. The server revalidates authoritative migration invariants when applying a move; an earlier `no_known_blockers` result is not authorization by itself. Route every source queue to 100% first, then cut over every concurrency group, then assess and move each pipeline. ## Command reference | Command | Purpose | | --- | --- | | `queue configure <queue> --destination-cluster <cluster>` | Map a source queue to an existing destination queue at 0%. | | `queue status [queue]` | Show one queue migration or list all migrations. | | `queue set-percent <queue> --to <0-100>` | Set the absolute percentage of new jobs routed to the cluster queue. | | `queue metrics <queue>` | Compare recent source and destination activity. | | `concurrency-group cutover <scope>/<key> --destination-cluster <cluster>` | Start holding, draining, and moving a concurrency group into the cluster, or report why it can't start. `--dry-run` checks only the group's migration state. | | `concurrency-group status [<scope>/<key>]` | Show one concurrency-group migration's state and unclustered jobs, or list every migration. A group whose cutover hasn't started has no migration. | | `pipeline readiness <pipeline> --destination-cluster <cluster>` | Assess known queue and concurrency-group blockers. | | `pipeline move <pipeline> --destination-cluster <cluster>` | Permanently assign a pipeline to the cluster. | Run `cluster-migrator <command> --help` for full usage. ### Global flags | Flag | Purpose | | --- | --- | | `--dry-run` | Validate and display a mutation without applying it. | | `--json` | Write machine-readable JSON instead of terminal output. | | `--timeout <duration>` | Set the maximum command duration; defaults to `10m`. | | `--endpoint <url>` | Override the Buildkite REST API endpoint. Also available as `BUILDKITE_API_ENDPOINT`. | | `--api-token <token>` | Supply the API token. Prefer `BUILDKITE_API_TOKEN` to keep it out of shell history and process listings. | All mutations are non-interactive. The CLI reads fresh server state before applying routing changes. ## Output and automation Human-readable output includes current state, results, observation freshness, and suggested next steps. Missing metric values appear as `—`, not zero. Use `--json` in scripts: ```shell cluster-migrator --json queue status cluster-migrator --json queue metrics test cluster-migrator --json pipeline readiness monorepo \ --destination-cluster production ``` JSON preserves the API's `null` metric and observation values. Terminal-only guidance is omitted. Queue metrics and pipeline readiness may initially be unavailable while Buildkite prepares an observation. The CLI follows the server's retry interval for up to one minute, writes prolonged-wait progress to stderr, and keeps stdout clean for the final human-readable or JSON result. ## Development Install the pinned tools, build the binary, and run the development checks with [mise](https://mise.jdx.dev/): ```shell mise install mise run build mise run format mise run lint mise run test mise run vet mise run vulnerability ``` The build task creates `./cluster-migrator`. Run `mise run release-check` when changing the release configuration. The pre-commit hook runs formatting, linting, tests, and `go vet` in parallel. ## License MIT. See [LICENSE](LICENSE).

Carregando o README do repositório…

Skills neste repositório

Analistas de garantia de qualidade de software e testadores

1 categorias ocupacionais · 100% classificado

skill

ocupação

descrição

atualizado

writing-cli-commands

Analistas de garantia de qualidade de software e testadores

Guides CLI command design for cluster-migrator. Use when adding commands or changing CLI output and its tests.

Idioma do texto original: inglês

2026-09-08

Mostrando 1 de 1 skills coletadas.

Instalar com um assistente de IA

Copie este prompt para o assistente de IA que você usa.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/buildkite/cluster-migrator.