For one of my QS dashboards, I’ve updated the SQL of the dataset by simply adding a single field.
The dataset refresh completed successfully but the dashboard broke. The visuals shows “Fields for this visual no longer in analysis - You can re-add the fields, or remove this visual”.
When looking into the analysis, rows and columns show --Missing field-- but, obviously, those fields are still listed and available in the Data pane. Also, the calculated fields are showing altered names with the last part consisting in numbers instead of the actual name.
This prevents me to update and fix any of my dashboards. What caused this? Any solution?
The symptoms you’re seeing (visuals showing Fields for this visual no longer in analysis, field wells showing --Missing field-- while the fields are clearly still present in the Data pane, and calculated fields appearing with a numeric/hash suffix) are the classic signature of the dataset’s column IDs changing.
Every column in a QuickSight dataset has a hidden internal column ID, separate from the display name you see. Your analyses, dashboards, visuals, filters, and calculated fields all bind to these column IDs, not to the field names. QuickSight only keeps a reference intact when the column ID still matches after a dataset update, so if an ID changes, the reference breaks even though an identically named field is still available in the Data pane.
When a dataset update causes the existing columns’ IDs to be regenerated, the analysis can no longer match its saved references, which produces exactly what you’re seeing:
Visuals show no longer in analysis / --Missing field--, they’re pointing at the old IDs, even though an identically named field (with a new ID) is right there in the Data pane.
Calculated fields appear with a number/hash in the name, when QuickSight can’t resolve the original reference during re-mapping, it displays a generated internal identifier instead of the friendly name.
Column IDs are typically regenerated by structural changes to the dataset. For example, a table being deleted/replaced, or existing columns being reordered, renamed, or having their data type changed. Simply appending a new column to the end of the query normally preserves the existing IDs, so it’s worth checking whether the edit also changed the order, names, or types of the existing columns.
To fix the broken analysis - The underlying data is intact, you just need to re-point the references to the current IDs:
Open the analysis. For each broken visual, drag the correct field from the Fields list back into the field well to replace each --Missing field--.
Open each renamed calculated field, confirm/repair the expression (its inputs may have lost their reference), and rename it back to its intended name.
Re-check filters and actions, they also bind by ID, re-select the field, and recreate any referenced parameters with the same name and data type.
Save and publish the dataset, then reopen and republish the analysis so the corrected mappings flow through to the dashboard.
If you have an asset bundle export (or a prior version) from before the edit, restoring that is the cleanest recovery.
Here are the helpful official AWS documentation and community posts:
Understood about the column IDs. Thank you for the explanation.
I checked my query and confirm it doesn’t have any ORDER BY or data type change. There are three column alias set for hard-coded values that I haven’t changed. The field I added, is right before the last one, which is an aggregated field. Do you think any of this could be the reason because the dashboard breaks? Or is the cause something else?
The dashboard has now been fixed by re-creating the visuals but, I’d like to avoid need to do this every time the dataset needs to be updated. Other than taking a lot of time, it also prevent me to build more complex dashboards as fixing them could require hours.
Thank you for checking your query and sharing those details. I’m glad the dashboard is working again.
I’d also like to correct my earlier explanation about column IDs, and I apologize for any confusion. Visuals and calculated fields in an analysis refer to dataset fields by their field name. A visual can break when a field it uses is renamed, removed, or changes data type.
Since your query has no ORDER BY and the three aliased hard-coded values weren’t changed, these are unlikely to be the cause. To help find it the cause, could you check whether any existing field names in the dataset’s field list changed after the edit.
To avoid this in future updates
Give every column in your query an explicit alias, so field names stay consistent.
Keep existing field names and data types unchanged when you edit the query.
Before making changes, back up the analysis with an asset bundle export and test the change on a copy of the analysis.
Where possible, create reusable calculated fields in the dataset instead of the analysis.
If fields go missing again, you usually don’t need to recreate the visuals. You can drag the correct field from the Fields list back into each field well.