I've previously been able to follow this workaround to render large datasets in dashboards.
https://support.esri.com/en/technical-article/000014750
This worked in 10.8 but on a staging server testing things in 10.9.1 I've found the MaxRecordCountFactor does not appear to be honored by Web Maps and apps. Monitoring the browser traffic it appears all queries sent to the end point have that parameter fixed at 3. Is there any way to set this somewhere else, perhaps in a map's definition? I've only seen ways to modify service definitions.
The issue is also compounded by the fact that dashboard filters seem to apply them after the tile requests (as opposed to a web map filter/query which seems to apply before the request to the endpoint) so even if the data set is being filtered down to a manageable level by dashboard filters, because the per tile requests pre-filter exceed the maximum we get missing data. Is there any kind of documentation on what filters are applied when? Like maybe URL parameters behave different so I could try those and replace my selector with web links to the dashboard with a different parameter?
It's pretty common that I'd rather a web map or dashboard perform slowly or even outright crash than display a visualization that is arbitrarily omitting data. I've tried some suggested techniques to reduce datasets but they've often at odds with other things I'm trying to do. For example I can reduce a data set dramatically by consolidating stacked points into a count field, but when I then combine that with clustering there is no way to weight the clustering by that field. A cluster combining 2 points with a stacked count of 1 vs. 2 points with a stacked count of 1000 is sized identical because the size is based on feature count, not my stacked count attribute.
So then I tried the full dataset and so the cluster size can be set by the actual feature count but now I'm running into this MaxRecordCountFactor which means as higher zooms the cluster totals are completely wrong with no indication that they are incorrect.
