Skip to main content
Everyone

Ingest statuses explained

Every upload, connector pull, new version and Q&A pair becomes an ingest job with two labels: a Status, how far the job has got, and a Stage, what the pipeline is doing. The Ingest status section lists both, newest first, and refreshes every two seconds while any row on the page is still queued or processing.

The Ingest status section of Northwind Supplier Handbook: three Q&A pair rows and the document rows, all Indexed, and supplier-form-scan.pdf stopped at Chunking with an Error tag, Failed.
The Ingest status section of Northwind Supplier Handbook: three Q&A pair rows and the document rows, all Indexed, and supplier-form-scan.pdf stopped at Chunking with an Error tag, Failed.
Enlarge

Each row gives the Document, with its source and id underneath, then Stage, Progress, Started and Status. All Status filters by status, and you can tick more than one. Select the Started heading to reverse the order. Search by name only covers the rows on the page you are looking at.

Status: how far the job has got​

Each status is shown in the colours of its badge in the product.

StatusWhat it meansWhat you can do
QueuedAccepted, and waiting for a worker to pick it up.Wait.
ProcessingA worker is running the Stage shown beside it.Wait. The Progress bar shows the status, not how much of the work is done.
IndexedFinished. The content is in the index and answers can cite it.Nothing.
FailedThe job stopped and will not continue on its own.Read the Error tag beside the stage, then retry the document from Documents.

Stage: where in the pipeline it is​

StageWhat is happening
ParsingReading the file and pulling out its text.
ChunkingSplitting that text into the passages an answer can cite.
DeduplicatingChecking those passages against what is already held.
EmbeddingTurning each passage into something a question can be matched against.
IndexingWriting the result to the index.
Rendering previewBuilding the preview shown when a document is opened.

A Q&A pair only passes through Embedding and Indexing. Its row is titled with the question and labelled Q&A pair underneath. Three more stages are housekeeping: Purging (clearing a removed document out of the index), Reassigning access (applying changed access tags) and Superseded. A row for something since removed is titled Deleted document or Deleted Q&A pair.

Superseded is not a failure​

A row reading Indexed with the stage Superseded was overtaken: a newer version, or a removal, arrived while it ran. It carries no Error tag and nothing is missing.

Reading a failure​

The Stage column keeps the stage that broke, so you can see where a job stopped. Hover or focus the Error tag beside it: the first line is a plain explanation, the second the raw message from the pipeline. Elie retries a temporary problem, such as a service briefly unavailable, by itself. When those retries run out the tag starts Automatic retries were exhausted and offers a manual retry.

The Error tag on supplier-form-scan.pdf hovered: The file was read but produced no content to index. Re-upload a document with extractable text. Below it, the raw message: no chunks produced.
The Error tag on supplier-form-scan.pdf hovered: The file was read but produced no content to index. Re-upload a document with extractable text. Below it, the raw message: no chunks produced.

There is no retry on this section. Use Retry on the document's row in Documents, or Retry indexing on the document itself; each document allows five manual retries. A failed Q&A pair cannot be retried: edit it or add it again. See Managing knowledge base documents.

When documents look like copies of each other, a Flagged duplicates card under the table names each one, the document it resembles and how closely its content matches.

The same labels on Documents​

Each document carries an overall Status using the same colours, plus two of its own while it is being removed:

StatusWhat it means
DeletingA removal is in progress.
DeletedRemoved from the index. Hidden from Documents unless you tick Deleted in the All Status filter.

A document with a new version on the way shows its current version beside the status, as in Indexed · v1, with a second chip for the new one: (v2) processing…, or (v2) failed if it did not index. The current version keeps answering until the new one is indexed.

Next​

Last updated on Something wrong with this page?