Run Lifecycle
Pending, running, and terminal act runs. Each row freezes compileHash at create, then moves through phases you can read in Runs and tie to audit.
This page is the behavior chapter after Workflows and cells. You already know what a run is. Here you interpret status, timeline phases, and timestamps when you debug production act. For the first run, start with Run first workflow.
What you see in Runs
Each row is one governed invocation. Act runs that launch a cell move from pending through running to a terminal status. Infer and decide rows may complete without a cell Job and are not the primary act run-minute path.
| Field | Meaning |
|---|---|
id | Stable run id (run_…) for API, audit filters, and deep links. |
workflowId | Optional wf_ id binding policy defaults and placement. |
executionTier | infer, decide, or act. Only act paths that launch a cell follow the full cell timeline. |
status | pending, running, succeeded, failed, or blocked. |
compileHash | Policy bundle fingerprint captured at create. Does not change when you publish new policy. |
clusterId | Hosted Blekline runtime or your connected cluster. Null on some decide paths. |
cellProfileId | Profile that set CPU, memory, and max duration for the cell Job. |
k8sJobName | Job name on cluster act; hosted runs use Blekline compute routing. |
cellDurationMs | Billable wall time in the cell after completion (see Billing & limits). |
latencyBootMs | Create to runningAt when reported by the executor. |
createdAt / completedAt | UTC bounds for the run row and timeline. |
Status model
The control plane stores five statuses on every run. Live act runs refresh in the console while status is pending or running.
| Status | Terminal | Meaning | In console |
|---|---|---|---|
pending | No | Run row exists; cell Job is scheduling or awaiting assign on hosted or cluster paths. | Status pill; timeline shows Job scheduled as active. |
running | No | Cell is executing act; duration and runningAt update while live. | Detail page polls every few seconds until terminal. |
succeeded | Yes | Act finished; cellDurationMs set when the agent reports completion. | Timeline terminal phase Succeeded. |
failed | Yes | Executor or in-cluster agent reported failure. | Workers tab when execution plane is connected; check cluster logs. |
blocked | Yes | Governance blocked the path (policy or approval hold). | Event History may link to rule approvals. |
Act-tier timeline
For executionTier: act, the Timeline tab follows the real run record: policy snapshot, scheduling, cell execution, then a terminal outcome. It does not invent orchestration steps your worker never reported.
| Phase | What happens |
|---|---|
| Run created | POST /api/workspace/cell-runs returns the row with initial status. |
| Policy snapshot | compileHash and executionPolicyId bind published runtime policy at birth. |
| Job scheduled | Assign queue or cell Job enqueue for hosted warm pool or cluster executor. |
| Cell running | Pool operator or agent sets running; infrastructure snapshot shown on the timeline. |
| Terminal | succeeded, failed, or blocked with completedAt and optional cellDurationMs. |
Hosted act omits clusterId on some rows; cluster act sets clusterId and uses your connected executor. See Spin up cells.
Infer and decide runs
Non-act tiers can record a run for identity and audit context without scheduling a cell Job. The timeline shows a shorter decide path. These paths are not the hero meter for act run-minutes. See Billing & limits.
compileHash snapshot
compileHash is captured when the run is created from published runtime policy and the optional workflow binding. New publishes do not alter older run rows. The run Policy tab and audit filters use the same hash. Author policy on Runtime policy.
Duration and metering
When a cell finishes, the agent can report cellDurationMs. That value drives act run-minutes on plans that meter cell-backed act. latencyBootMs is the gap from create to runningAt when the executor reports it. Profile max duration still caps how long a cell may run.
Blocked and failed
blocked means governance stopped the path. The run detail page may surface a link to Rule approvals or related improve-loop surfaces. failed means the executor or agent reported an execution error. Use the Workers tab when your workspace has a verified execution plane and cluster act is connected.
For turning blocks into policy changes, see Threat signals & approvals.
Who advances status
| Actor | Role |
|---|---|
| Workspace session | Creates runs via console or POST /api/workspace/cell-runs with a scoped token. |
| Cell pool operator | POST /api/compute/cell-pool/run-lifecycle after assign (cluster dispatcher token). |
| Cell agent | PATCH run status and cellDurationMs when BLEKLINE_WORKSPACE_TOKEN is mounted on the Job. |
Request and response shapes for create and patch live in Cell runs and the OpenAPI catalog.
Console map
Where to look
Inspect runs, timeline, and audit from the same detail page.
Inspect
- RunsFilter by status, workflow, and placement.
- Run timelinePhases derived from the run record (no fictional orchestration steps).
- Event HistoryAudit log preset filtered by run id.
When blocked
- Rule approvalsHuman-reviewed publish after block telemetry.
- Threat signalsProposals that feed the improve loop.