~2x degradation of range queries with COUNT

Setup:

• 4 KV nodes, 1 query node, 1 index node
• 1 bucket
• 100M documents (non-DGM)
• 30K updates/sec

Index:

CREATE INDEX by_capped_small ON `bucket-1`(capped_small);

Queries:

SELECT COUNT(*) FROM `bucket-1` WHERE capped_small > $1 AND capped_small < $2;

Those queries match ether 1K or 10K documents. Both cases regressed. Both MOI and Plasma are affected.

See:
http://showfast.sc.couchbase.com/#/timeline/Linux/n1ql/count/MOI
http://showfast.sc.couchbase.com/#/timeline/Linux/n1ql/count/Plasma