KILL QUERY should work during all phases of a query

[beckettsean]

Feature Request

Opening a feature request kicks off a discussion.
Requests may be closed if we're not actively planning to work on them.

Proposal:

Make it so queries can be interrupted during the planning phase, not just the execution phase.

Current behavior:

KILL QUERY can't interrupt queries until the query is at the execution phase, actually sampling data from disk. With large datasets, a query like SELECT * FROM /.*/ will spend a long time and a very large amount of RAM just instantiating the query engine iterators. During that phase, KILL QUERY won't interrupt the process and the RAM use will continue to rise.

Desired behavior:

KILL QUERY should terminate the query as quickly as feasible no matter what phase the query is in.

Use case:

Currently there are queries that can't be killed before they OOM the system even if the query is detected and a KILL is issued.

[jsternberg]

This is blocked until #7646 is merged. That PR changes the structure of the query engine enough that any work on this will conflict with that PR.

[hardiksondagar]

I'm still facing this issue in 1.6x

> show queries
qid  query                                                                                                                                                                              database   duration status
---  -----                                                                                                                                                                              --------   -------- ------
5649 SELECT SensorId, first("2001") AS Value FROM Reading WHERE time >= '2018-09-15 12:15:00' AND time < '2018-09-15 12:30:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId smartsense 5m16s    killed
5651 SELECT SensorId, last("2001") AS Value FROM Reading WHERE time >= '2018-09-15 21:30:00' AND time < '2018-09-15 21:45:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId  smartsense 5m14s    killed
5652 SELECT SensorId, first("2001") AS Value FROM Reading WHERE time >= '2018-09-15 18:30:00' AND time < '2018-09-15 18:45:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId smartsense 5m14s    killed
5712 SHOW QUERIES                                                                                                                                                                                  380µs    running
5647 SELECT SensorId, last("2001") AS Value FROM Reading WHERE time >= '2018-09-15 09:00:00' AND time < '2018-09-15 09:15:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId  smartsense 5m19s    killed
5648 SELECT SensorId, first("2001") AS Value FROM Reading WHERE time >= '2018-09-15 06:00:00' AND time < '2018-09-15 06:15:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId smartsense 5m19s    killed
5650 SELECT SensorId, last("2001") AS Value FROM Reading WHERE time >= '2018-09-15 15:15:00' AND time < '2018-09-15 15:30:00' AND SensorId =~ /^17[0-9]{1}[0-9]{2}$/ GROUP BY SensorId  smartsense 5m16s    killed

1  [|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||100.0%] Tasks: 58, 210 thr, 106 kthr; 5 running
2  [|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||100.0%] Load average: 4.09 3.34 2.03
3  [|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||100.0%] Uptime: 283 days(!), 22:30:51
4  [|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||100.0%]
Mem[||||||||||||||||||||||||||||||||||||||||||||||||||||||13.4G/15.7G]
Swp[||                                                     774M/32.0G]

[VadimSelyanik]

Still have this issue in 1.7.0