ISSUES-2688 support on cluster for kill & optimize query

[zhang2014]

#2688
cc @sundy-li
I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

[alexey-milovidov]

Ok, but the purpose remains unclear.

[sundy-li]

I think this feature is very useful.
We often face a problem that when a complicate query comes, we want it to be killed right now.
The big query will cause a lot of alarm messages by monitor system (too much IO,CPU usage etc ), we have a shell like

for i in ${hosts[@]};do
	 echo "killing $i"
	  echo "kill query  where elapsed > 320" | $ck --host $i
done

It was less convenient, because we must modify the hosts variable each time we do cluster expansion.

[blinkov]

@sundy-li doesn't your company have some infrastructure to execute arbitrary command in parallel on a tagged subset of servers via ssh? If it does, than such stuff is easily done just by querying localhost with clickhouse-client or curl.

[alexey-milovidov]

Thank you! KILL on cluster is indeed useful.

By the way, if you have a distributed query (that generate many queries on cluster nodes), and if you KILL just the initiating query, all generated queries will also be killed.

About OPTIMIZE. In many cases excessive OPTIMIZE (especially with FINAL) may be harmful, because it will initiate huge merge, that can fill a queue for merges and prevent other merges to run. This feature should be used with caution.