Command Center (Operator)

Ingestion Health

Monitor the ingestion pipeline and retry failed jobs across all tenants.

The Ingestion Health panel at /operator/health gives operators a platform-wide view of the knowledge ingestion pipeline and the tools to recover failures.

Health metrics

Four counters summarize the pipeline:

MetricMeaning
PendingJobs queued and waiting to start.
ProcessingJobs in an active stage (parsing, chunking, embedding, indexing).
IndexedDocuments fully processed and searchable.
FailedDocuments whose ingestion gave up.

A refresh control re-polls the current status on demand. The counters come from ingestion.getHealth.

Failed jobs

Below the metrics, a table lists failed jobs across all tenants, with the job id, document title, status, organization, and when it was created. Each failed job has a Retry control (with a confirmation step) that re-enqueues it through the pipeline.

  • Listing: ingestion.listAllFailedJobs
  • Retry: ingestion.retryJob

After a retry, the health metrics and the failed-jobs list refresh so you can confirm the job moved on.

A job reaches failed only after exhausting its retries, and dead_letter for unrecoverable conditions (such as a deleted document or an embedding-dimension mismatch). Understanding which you're looking at tells you whether a retry can succeed. See failure handling.

GP users see a tenant-scoped version of this same health view under organization settings → Knowledge. For the recovery procedure, see ingestion troubleshooting.

On this page