Platform Analytics Dashboard Visualization Disappearing

[Christopher Sik]

I was editing a multi-tab Platform Analytics Dashboard yesterday. I resized a couple of single score widgets on one tab (we'll call it tab 13 for this discussion), saved my changes, exited out of Edit mode, and then navigated to a different tab (tab 2) in the same dashboard.

But tab 2 loaded with the visualization widgets from the tab 13 and not the usual set of widgets which had existed there for over a year.

I checked other tabs in the dashboard and the widgets there all loaded normally. Then I returned to tab 13  and noted that my resize changes had reverted back. So I went back into edit mode, set them again, and saved my work. I refreshed the dashboard and the changes appeared to stick. So I went back to tab 2, only to find that no widgets were displayed. They all just vanished as if they never existed.

I couldn't find them in the library either.  Fortunately, I had a recent version of the dashboard in a sub-prod instance and was able to export the visualizations from that instance and import them into PROD to reconstruct tab 2. There were a couple that only existed in PROD and appear to be gone forever now.

This is not the first time this has happened either. A few months ago a different tab in this dashboard was impacted the same way.

Has anyone encountered a bug like this before?

[Naveen20 · Accepted solution]

The absence of any audit trail or even records in pa_m2m_dashboard_tabs suggests the Platform Analytics dashboard framework may be storing widget/tab configuration differently than classic PA dashboards — likely in-memory or in a client-side cache layer that gets flushed and rewritten on save rather than making incremental server-side updates to individual records.

So what probably happened is that when you saved your resize changes on tab 13, the platform serialized the entire dashboard state back to the server, but the in-memory representation had already corrupted the tab associations. The save then overwrote the correct configuration wholesale with the corrupted one. Since it was a full overwrite rather than individual record updates, there's no granular audit trail — the platform treated it as a single "dashboard saved" event, not as discrete widget deletes or moves.

This makes the bug significantly harder to recover from and reinforces the update set snapshot approach before any edits. When you open the support case, emphasize specifically that there's no audit trail for the widget loss — that's important for the engineering team because it narrows the defect to the dashboard serialization/deserialization path rather than a simple tab reference mixup. 

[bagills]

This is helpful for tracking down the deleted records to enable recovering lost changes. However, this does not address the issue or explains how to prevent it in the future. 

Is there any guidance on this? We are still observing this issue in Australia Patch 2