Failed Job Notification

Keboola logs the outcome of every job in an orchestration, including whether it finished, errored, or was skipped. A Beam agent reads that job log after each scheduled run and, when a job errors, notifies the data team with the job name, error message, and which downstream steps depend on it. Jobs that finish with a warning but no hard error are logged without an alert, since those often resolve on their own. The agent does not rerun a failed job or edit its configuration, it only surfaces the failure so a person can decide whether to fix and retry it.

Row Count Anomaly Flags

Keboola reports how many rows each extractor or writer produces on a given run. A Beam agent reads that row count against the recent average for the same job and, when a run comes in far below or above that average, flags the output and notifies the data team before anything downstream consumes it. Jobs still building up a reliable history, or ones already flagged for a known seasonal pattern like a monthly close, are excluded from the check. The agent does not stop a job outright, it only flags the output, leaving the decision to hold or proceed with a person.

Orchestration Dependency Holds

Keboola orchestrations run a sequence of jobs where later steps depend on earlier ones finishing cleanly. A Beam agent watches that sequence and, when an early job fails, holds the downstream jobs that depend on it rather than letting them run against incomplete data, then notifies the data team of exactly what is paused and why. Jobs with no dependency on the failed step continue on schedule as normal. The agent does not reorder the orchestration or remove a job from the sequence, it only pauses runs that would otherwise process on bad input, until a person clears it.