This is a top-level attribute of the pipeline spec.
datum_tries attribute in a Pachyderm pipeline specifies the maximum number of times that Pachyderm will try to process a datum. When a datum fails to process, either because of an error in the processing logic or because it exceeds the
datum_timeout value, Pachyderm will automatically retry the datum until it is successful or the number of
datum_tries has been reached.
Each retry of a datum is treated as a new attempt, and the datum is added back to the job queue for processing. The retry process is transparent to the user and happens automatically within the Pachyderm system.
datum_triesis set to
3by default if unspecified.
- Setting to
1attempts a datum once with no retries.
- If all tries have been exhausted and processing has not succeeded, the datum is marked as
When to Use #
You should consider setting a higher
datum_tries count if your pipeline has a large number of datums that are prone to errors or timeouts, or if the datums you are working with have to be imported or fetched (via data ingress) from an external source.