A bug was discovered where users had committed changes that were picked up by the streaming update, but lost after a bulk ingest. These edits were valid, so there was understandable consternation.
The problem is that we had a special pointCounts column function used only during bulk ingest that omitted a number of categories of edits from consideration if the underlying geometry was a point, including nodes tagged as buildings (which is permissible).
This is a quick fix PR that seems to address the problem. We ought to revisit this code, in case further streamlining is possible now that points/ways/relations are all being counted similarly.
A bug was discovered where users had committed changes that were picked up by the streaming update, but lost after a bulk ingest. These edits were valid, so there was understandable consternation.
The problem is that we had a special
pointCountscolumn function used only during bulk ingest that omitted a number of categories of edits from consideration if the underlying geometry was a point, including nodes tagged as buildings (which is permissible).This is a quick fix PR that seems to address the problem. We ought to revisit this code, in case further streamlining is possible now that points/ways/relations are all being counted similarly.