Comment 2 for bug 1402031

A likely workaround for this problem is to turn off the cqd USTAT_ESTIMATE_HBASE_ROW_COUNT before running Update Stats.
This will cause the slower but precisely accurate count(*) method to be used instead of estimating based on HBase meta-information for the HFiles and memstore of the table.

The sporadic nature of the problem suggests some possible conflict with a concurrently executing hbase operation like a flush, compaction, region split, etc. Also, the estimate will not be as accurate if a large percentage of the table's data is still in the memstore, since the row count in the memstore can't be measured as accurately. However, this in itself should not produce a grossly inaccurate estimate as reported in this case.