KV Cache miss ratio derivation can produce values >100%

kv_ep_cache_miss_ratio is a metric derived by ns_server and used for the UI Dashboard.

It is calculated as:

get_metric(<<"kv_ep_cache_miss_ratio">>) ->

    percent(_(<<"kv_ep_bg_fetched">>),

            ?cut(promQL:sum_without([<<"op">>, <<"result">>],

                                    _({[{eq, <<"name">>, <<"kv_ops">>},

                                        {eq, <<"op">>, <<"get">>}]}))), 0);

100 * kv_ep_bg_fetched / kv_ops{op="get"}

kv_ep_bg_fetched is the number of performed BGFetches. kv_ops{op="get"} is the number of "get" operations, which includes GET/GAT/Subdoc Lookup. KV can perform BGFetches for non-get operations. For example, sets can require a BGFetch, if XATTRs are used or if replacement is not allowed.

I'm not sure what the "cache miss rate" should represent, but I don't think having a value larger than 100% makes sense (there are more cache misses than requests).

If the intention of the metric is to be specifically looking at get/access operations, it can be replaced with get misses / (miss+hit).

100 * kv_ops{op="get", result="miss"} / sum without (result) (kv_ops{op="get"})