Skip to content

Execution state model

Beampipe separates its control phase from submission, scheduler, DALiuGE, and output-verification facts. Those facts can disagree, so none is allowed to stand in for the others.

Explore the control phase

NO EXTERNAL WORK

Discover and admit

Register source identity, prepare archive metadata, compare signatures, and evaluate readiness and automation limits.

The public status is a compact operator projection: pending, running, awaiting_scheduler, not_submitted, completed, failed, retrying, or cancelled. Exact position remains in control_phase and execution_phase.

Independent external axes

submissionnot_startedin_flightsubmitteduncertain / failed
schedulernot_submittedpendingrunningsucceededfailed / cancelled / unknown
DALiuGEnot_createdbuilding / deployingrunningfinishedfailed / unreachable
outputsnot_startedpending / verifyingverifiedfailed / unknown

The reducer derives the next action from all axes. A succeeded scheduler allocation with an active DALiuGE session is inconsistent and requires review; it is never promoted silently to success.

Submission uncertainty

Before external I/O, Beampipe persists submission intent and deterministic external identity. If SSH disconnects after sbatch, the submission axis becomes uncertain. Reconciliation searches by stable job name and checks squeue and sacct; retry stays blocked until the system can prove no scheduler job or DALiuGE session exists.

Retry gate

Observed stateSafe action
Failed before manifest creationRegenerate from pinned inputs
Known translation or pre-submit failureResume with immutable artifacts
Submission uncertainReconcile; retry blocked
External job or session existsMonitor or cancel; do not duplicate
beampipe execution retry "$EXECUTION_ID" \
  --reason "Translator endpoint restored after planned maintenance"

Every retry requires a reason and increments retry_count. Meaningful transitions, claims, recovery, retry, cancellation, and administrative actions append provenance or claim-history records.

Continue with the operator procedure for recovery and cancellation.