We noticed there was a discrepancy between the controller gauge for total documents vs. the one from servers for realtime tables. It seems like the former does not include docs from consuming segments and does not handle upsert tables well.
Example stats from an realtime non-upsert table:
- Controller gauge: 1,127,658,711 docs
- Server gauge: 1,095,767,023 docs
Example stats from a realtime upsert table, an order of magnitude different:
- Controller gauge: 30,271,741 docs
- Server gauge: 6,016,144 docs
We noticed there was a discrepancy between the controller gauge for total documents vs. the one from servers for realtime tables. It seems like the former does not include docs from consuming segments and does not handle upsert tables well.
Example stats from an realtime non-upsert table:
Example stats from a realtime upsert table, an order of magnitude different: