Problem:
Depending on moderate load (see attach munin report), I'm seeing inconsistent time for count()

Mon Sep 6 15:45:33
> db.blogs.count()
169096641
Mon Sep 6 15:54:22

Mon Sep 6 19:06:20
> db.blogs.count()
170166577
Mon Sep 6 19:06:29

The second clocks in a 9 seconds (when there is no load on the system), the first at 9 minutes with a system under moderate load.

I'm guessing that the count is a computed value rather than stored. If we use a stored value in future, then insert and delete operations will be gated by the latch time for the counter, so its a trade off.

Environment:

• 3 shard cluster, 2 members per shard, plus a config and router node (all separate machines)

Reproduce:
See steps above

Business Case:
Scalability