Can Phoenix CBO prune columns when statistics are missing?

Goal

Determine whether Phoenix 5.x can reduce I/O for low‑traffic scans by pruning column families when statistics are missing.

Constraints

• Sparse, intermittent queries on a large HBase table.
• Column family cardinality statistics may be stale or absent.
• Bucket pruning depends on high‑cardinality indexed columns.

Uncertainty

The optimizer may default to full region scans if it cannot estimate selectivity, potentially negating cost savings.

Questions

Can the cost‑based optimizer still prune unnecessary column families when column statistics are missing, and if so, what triggers the fallback to a full scan? Does bucket pruning consistently avoid scanning irrelevant regions when the index distribution changes over time?