Please use the form below to provide your feedback. Because your feedback is valuable to us, the information you submit in this form is recorded in our issue tracking system (JIRA), which is publicly available. You can track the status of your feedback using the ticket number displayed in the dialog once you submit the form.

Troubleshoot a Workflow

  • Capella AI Services
  • how-to
If your Capella AI Services Workflows complete with errors or have documents that cannot be processed, you can query a document’s extended attributes (XATTRs) data for more information.

Workflows create a META().xattrs.vectorization.status field on a document when that document is skipped during vectorization. You can use a SQL++ query to get information from this field and troubleshoot your Workflow.

If your Workflow returns a LCB_ERR_DURABILITY_AMBIGUOUS error, see Resolve a LCB_ERR_DURABILITY_AMBIGUOUS Error.

For more information about Workflows, see Process Your Data For Capella AI Services.

If you deleted or modified your Workflow’s metadata scope, collections, or Eventing functions, your Workflow might fail to run correctly. You must delete the Workflow and create a new one.

Prerequisites

Run a Troubleshooting Query

To troubleshoot and query the XATTRs for a skipped document:

  1. On the Operational page, click the name of the cluster you used in your Workflow.

  2. Go to Data Tools  Query.

  3. In the query editor, paste the following query, replacing the following values:

    • <$WORKFLOW_ID> with the ID value of your Workflow. You can view and copy your Workflow ID by expanding the entry for your Workflow on the Workflows page.

    • <$BUCKET> with the name of the bucket you selected as the destination for your Workflow.

    • <$SCOPE> with the name of the scope you selected as the destination for your Workflow.

    • <$COLLECTION> with the name of the collection you selected as the destination for your Workflow.

      WITH idx AS (
    SELECT COUNT(1) AS idx_count
    FROM system:indexes
    WHERE name = "<$WORKFLOW_ID>_progress"
      AND state = "online"
)
SELECT VALUE result
FROM (
    SELECT CASE WHEN (SELECT RAW idx_count FROM idx)[0] > 0
        THEN (SELECT META().id AS doc_id,
        {
            "num_malformed_payload_to_model": META().xattrs.vectorization.status.bad_request,
            "num_non_json_objects": META().xattrs.vectorization.status.non_json,
            "num_payload_size_exceeds_model_context_window": META().xattrs.vectorization.status.too_big,
            "num_non_existent_fields": META().xattrs.vectorization.status.no_fields,
            "num_successful_embeddings": META().xattrs.vectorization.status.success
        } AS details
            FROM `<$BUCKET>`.`<$SCOPE>`.`<$COLLECTION>`
            WHERE META().xattrs.vectorization.status IS NOT MISSING
              AND (
                    META().xattrs.vectorization.status.bad_request > 0
                 OR META().xattrs.vectorization.status.non_json   > 0
                 OR META().xattrs.vectorization.status.too_big    > 0
              )
              AND META().xattrs.vectorization.model    IS NOT MISSING
              AND META().xattrs.vectorization.encoding IS NOT MISSING
              AND META().xattrs.vectorization.version  IS NOT MISSING)
        ELSE [{"message": "There looks to be an issue with this workflow’s index status. Contact Couchbase support for further help."}]
    END AS arr
) AS subq UNNEST subq.arr AS result;

  4. Press Enter or click Run.

    The query should return results similar to the following:

    [
  {
    "details": {
      "num_malformed_payload_to_model": 0,
      "num_non_existent_fields": 0,
      "num_non_json_objects": 2,
      "num_payload_size_exceeds_model_context_window": 0,
      "num_successful_embeddings": 0
    },
    "doc_id": "mydoc123"
  }
]

Understand Query Results

The troubleshooting SQL++ query, in the details object for each doc_id, provides the following information:

Error Description

num_malformed_payload_to_model

The number of processing requests to the embedding model that were rejected as malformed or bad requests.

num_non_existent_fields

The number of processing requests that referenced non-existent fields in the document. Check the configuration of your Workflow to make sure your selected source fields exist in the document you’re trying to process.

num_non_json_objects

The number of processing requests that were rejected for JSON serialization issues. Check the content of your document and its fields to make sure the JSON is properly formed, if you’re using a Data from Capella or Structured Data from External sources Workflow.

num_payload_size_exceeds_model_context_window

The number of processing requests that exceeded the embedding model’s context size limit. Check the configuration of the embedding model you chose for your Workflow.

num_successful_embeddings

The number of processing requests that created an embedding.

Resolve a LCB_ERR_DURABILITY_AMBIGUOUS Error

A Workflow can return a LCB_ERR_DURABILITY_AMBIGUOUS error when:

  • The destination cluster is rebalancing.

  • The destination bucket has a Minimum Durability Level set to Replicate to Majority.

If your Workflow fails to run and returns this error, rerun your Workflow after your destination cluster has finished rebalancing:

  1. Go to AI Services  Workflows.

  2. Go to More Options (⋮)  Rerun Workflow.

Next Steps

After you check your Workflow configurations and documents, you can rerun a Workflow by going to More Options (⋮)  Rerun Workflow on the Workflows page.

If you cannot resolve the errors with a Workflow, contact Couchbase Capella Support.