One worker pool means a slow profile starves urgent runs.
Tasks and Pools
Every job runs in a credential-isolated pool — slow profiles never block urgent loads, and the agent decides its own capacity.
Operate
All work in BoltPipeline — pipeline steps, profiles, DDL deploys, lineage pulls — is one row in one queue, claimed by an agent that runs four isolated pools. A profile of a billion-row table cannot starve a time-critical load, and a buggy admin task cannot inherit the credentials of a reader task.
What it solves
Common problems this capability removes from your engineering workflow.
Pool credentials grow to the union of every permission any task needs.
Pipeline sequence is enforced in app code and breaks on race conditions.
Capacity is hard-coded; busy agents drown in oversubscribed work.
How it works
The flow this capability runs end-to-end inside the platform.
- 1Four pools map to four credential tiers — reader, pipeline, admin, system.
- 2Agents tell the platform per-pool capacity on every poll.
- 3The dispatcher never hands out more work than the agent can run.
- 4A watchdog tracks stuck tasks — agents are ephemeral, tasks are accountable.
Related capabilities
Other capabilities in the BoltPipeline platform that work alongside this one.
4Self-Healing Runtime
Operate Orchestrator
Pipeline steps run in true parallel with dedicated connections, classified errors, and policy-driven retry.
4Self-Healing RuntimeCrash-Safe Execution
If the agent crashes mid-pipeline, it resumes from the exact step that was running — committed work is never repeated.
6Agent &Agent Runtime
A stateless executor near your warehouse runs orchestration, profiling, and cataloging with six reliability primitives.
See it on your data.
Try BoltPipeline against your live database — your data never leaves your environment.