export¶

The same lifecycle applies whether you drive it with the docsynth CLI locally, deploy.sh on GCP, or the studio.

Run¶

Local: docsynth generate --run-id r1 --total 1000 [--catalogue <uri>] [--format pdf|html].
Cloud (blocking): deploy.sh -c run.yaml run — one execution per slice, waits.
Cloud (detached): deploy.sh -c run.yaml dispatch — returns immediately; a multi-hour run outlives the terminal. This is the studio's default.

Concurrency is just running the worker again — another process, another task — against the same --state; the atomic claim splits the units. job.tasks is how many tasks a Cloud Run execution starts.

Follow¶

docsynth status --run-id r1 --state <uri> — a snapshot; --wait streams until the run is terminal.
docsynth studio status --project <ref> --run r1 [--wait] — reattaches to a dispatched cloud run from any machine that has the project registered, reading the run's state store (no live job to hold). See Studio → Dispatch & status.
deploy.sh status / deploy.sh logs.

Resume¶

A run that leaves failed units exits non-zero, so a --wait, a Cloud Run execution, or a CI step reads a broken run as broken. Recover with:

docsynth generate … --resume (local) or deploy.sh resume (cloud).

Resume re-queues failed units and reclaims units abandoned by crashed workers (expired leases), then continues. It is safe to run repeatedly — generation is idempotent (a re-generated index overwrites identical content).

Pause / cancel¶

docsynth pause / docsynth cancel set the run state so workers stop claiming new units; resume clears a pause.

Export¶

Once a run is complete, load its golden shards into a queryable sink:

docsynth export --run-id r1 --sink <uri> [--storage-prefix …]
deploy.sh export (uses export.sink from the config).
Studio: the export step.

Export is re-runnable and independent of generation — see Invoice → Export.