Setting the file. One moment.
Subchapter 9.18
references/features-task-orchestration.mdMarkdown7 KBView on GitHub
pnpm -r run <script> schedules a graph of workspace tasks. A task is <project>#<script>; it becomes ready once every task it depends on succeeds, and ready tasks run under --workspace-concurrency. Independent tasks run in unpredictable order.
Configure under tasks in pnpm-workspace.yaml:
tasks:
build:
dependsOn:
- ^build # build in each workspace dependency
test:
dependsOn:
- build # build in the same projectbuild → the build task in the same project.^build → the build task in each selected workspace dependency.tasks entry defaults to depending on the same task in its workspace deps (an unconfigured build behaves as dependsOn: ['^build']), preserving deps-before-dependents order.dependsOn means dependsOn: []. If you set another field (e.g. concurrency) and still want topological order, declare dependsOn: ['^build'] explicitly.--filter/includeWorkspaceRoot selection. A project missing the named script is a pass-through (reported skipped, doesn’t break the chain).tasks:
build:
concurrency: 2 # max 2 instances of build across projects
dependsOn: ['^build']Separate from --workspace-concurrency; a task waiting for its slot doesn’t occupy a workspace slot.
Machine-wide limits shared across pnpm processes that use the same stateDir (including pnpm pipeline):
tasks:
test:rust:
concurrencyGroup: cargo
dependsOn: []
concurrencyGroups:
cargo: 2 # at most 2 cargo tasks at once, across processespnpm run in the same group reuses its parent’s slot. Slots release on process exit/crash.stateDir creates a separate slot pool.priority (integer, default 0) orders waiting tasks for available slots — higher runs first, ties broken by arrival:
tasks:
build:critical: { concurrencyGroup: build, priority: 10 }
build:cleanup: { concurrencyGroup: build, priority: -1 }pnpm tasks status [groups...] # running + waiting tasks per group
pnpm pm tasks status # force built-in if a "tasks" script shadows itA subcommand is required. A waiting task shows its priority only when non-zero.
pnpm -r run --dry-run build # stable topological ordering, no scripts run
pnpm -r run --dry-run --json test # nodes + edges as JSON--resume-from <pkg> — resume at a package’s task, skipping tasks a prior run of the same invocation recorded as passed.--reverse — reverse every edge (dependents run first).--no-bail — keep running independent ready tasks after a failure; tasks depending on the failed one are skipped (default --bail stops dispatching and kills running tasks: SIGTERM, then SIGKILL after 2s on Unix). Accepted only by run/exec/test/start/stop/rebuild/install-test.--workspace-concurrency defaults to min(4, available CPU cores), respecting container CPU quotas; <= 0 means cores minus abs value (min 1).--report-summary statuses: passed|queued|running|skipped|failure (failures carry prefix/message); non-requested tasks are keyed <dir>#<task>.--stream for immediate prefixed output or --aggregate-output.A cycle fails before any script with ERR_PNPM_TASK_CYCLE. Set ignoreWorkspaceCycles: true only for deliberate cycles (pnpm warns and drops ordering among members).
--no-sort and --parallel (implies --no-sort) ignore tasks declarations. Recursive exec has no script name so it doesn’t join dependsOn, but still uses dependency-aware scheduling and --resume-from.
Runs a named set of tasks the way a CI job would: frozen install, then the affected projects’ task graph, with cached results restored.
tasks:
build: { dependsOn: ['^build'], outputs: ['dist/**'], inputs: ['src/**'], env: ['NODE_ENV'] }
test: { dependsOn: ['build'], outputs: [] }
lint: { outputs: [] }
pipelines:
default: [build, test, lint]
release: [build]pnpm pipeline # runs the "default" pipeline
pnpm pipeline release
pnpm pipeline --dry-run --json
pnpm pipeline --full # every project, not just affected-since-base
pnpm pipeline --base <ref> # affected diff base (default origin/main, or pipelineBase)
pnpm pipeline --no-cacheoutputs (outputs: [] = “produces no files”, makes a linter/test cacheable). inputs narrows the cache key (+glob adds to the default); env names hashed vars; cache: false opts out. The key also covers script text, dependency task keys, the lockfile, and the runtime. A hit restores files and replays logs.pnpm run never restores from the pipeline cache; outputs/inputs/env/cache/cargoTargetDir are read by pnpm pipeline only.remoteCache setting shares signed results through a Turborepo-API server (e.g. Vercel Remote Cache) or pnpr. Machines restore only results signed by their trustedKeys; CI publishes with PNPM_REMOTE_CACHE_URL/_TEAM/_TRUSTED_KEYS/_PRIVATE_KEY/_KEY_ID/_BUILDER_ID/_PUBLISH=true. Published results include output files and terminal logs. A missing, untrusted, or unreachable cache is just a miss.pnpm install option (for its frozen install). Errors: ERR_PNPM_NO_PIPELINES, ERR_PNPM_UNKNOWN_PIPELINE, ERR_PNPM_PIPELINE_FAIL. --report/--report-to <url> submit run data; --repo with --watch/--once/--interval watches a repo.cargoTargetDir: target keeps a task’s Cargo target dir between runs/worktrees via immutable snapshots.Workspace install, rebuild, pack, publish, stage, and lifecycle work start a project’s work as soon as its workspace deps finish — following the package graph (not tasks), no longer waiting for unrelated topological groups.