Calculate per query statistics

[jvshahid]

More specifically, for a given select query:

1. Time took to run the query (possibly slow queries can make it to a slow log)
2. Memory consumed by a given query (probably just an estimate)
3. Shards hit by a given query
4. Local aggregation flag
5. Number of points read from each shard
6. Number of servers hit
7. Number of points returned from each server

[elcct]

I have implemented EXPLAIN command that works with any SELECT queries

Example:

explain select value from testing limit 20

Will return:

shard_local says if query was executed on a local server or through remote call
points_read says how many points query processor has received from datastore
points_written says how many points query processor has written
run_time milliseconds spent on executing the query

Other example:

explain select count(value) from testing group by time(1m)

Result:

It can already give some interesting data - like how come empty shard has appeared or that LIMIT clause works per shard...

To an extent this depends on work I did for #294

At the moment it is collecting data from QueryEngine, but I am not sure if data for other engines that participate in SELECT query could be useful

Do you think I should continue with this?

[pauldix]

Yeah, looks great. Not sure we need the other engines to produce stats. As long as we see the same metrics regardless of query, which I think it will because at some point it goes through the engine?

Can you push to a branch and submit a PR so we can have a look?

[Dieterbe]

want!!

[IngaFeick]

+1
Would also be cool if this worked for "list series" commands aswell.

[beckettsean]

closing in favor of #2044