Running
Run individual output steps or the whole flow. Prep processes the full data at this point, not the sample, so this is where problems the sample hid appear.
The run summary reports rows written and any errors. Check the row count against expectation every time — it is the cheapest verification available.
Scheduling options
| Option | Notes |
|---|---|
| Tableau Prep Conductor | Add-on for Server/Cloud; proper scheduling, monitoring, alerting |
| Command line | tableau-prep-cli driven by any scheduler |
| Manual | Someone remembers to press the button |
Conductor is the supported answer and costs extra. The command line is the common workaround where it is not licensed.
Incremental refresh
Where supported, an input can be configured for incremental refresh — processing only rows newer than the highest value of a chosen field.
Much faster on large sources, but blind to rows that were updated or deleted after the fact. The usual pattern is incremental most of the time with a periodic full refresh.
Making flows maintainable
- 1Name every step for what it does, not Clean 1 and Clean 2
- 2Add descriptions to steps whose purpose is not obvious
- 3Filter early so later steps process fewer rows
- 4Remove unneeded fields at the input
- 5Keep one flow to one purpose rather than one enormous flow
Prep or the warehouse
If the same shaping is needed by several teams and tools, it probably belongs in the warehouse rather than in a Prep flow — a view or a scheduled transformation serves everyone, not only Tableau.
Prep is at its best for Tableau-specific preparation, and for work owned by analysts rather than engineers.