Hi all!
The AlloyDB documentation mentions the following:
AlloyDB refreshes the data within a columnar unit—a set of data blocks within a columnar table—when all of these conditions are met:
The columnar unit's table has exceeded the scan-threshold defined by the separate refresh_threshold_scan_count flag.
The percentage of the columnar unit's underlying data blocks that have become invalid exceeds the value of the refresh_threshold_percentage flag. Columnar data blocks become invalid due to changes in the row-based data that the columnar table draws upon.
This made me believe the columnar engine itself is eventually consistent and could yield stale results? Or does it skip invalid blocks and draw upon the row storage instead?
I was unable to find a good answer on this. I'd imagine there would be a clear warning in the documentation that results could be stale and haven't seen it (yet), hence I'd assume it is strongly consistent.
Note for my use case eventual consistency would be fine, but preferably within a few seconds max.
Hope someone has an explanation.
Kind regards,
Niels
LinkedIn
Twitter
