VTgate --query-timeout is ignored

[angelUrg]

Overview of the Issue

The initial problem is that we have a joins queries like this:
select a.SchemeId as Id, a.Enabled, a.SegmentId, b.CompanyName, shb.IsProfileEnabled from table1 as a join schemes as b on b.Id = a.SchemeId join table2 as shb on a.SchemeId = shb.SchemeId where a.SchemeId = 'xxx' and a.ConfigurationID = 'xxx';

This queries are creating a lot of scatter queries which we want to kill or prevent from running too long so we've tried with --query-timeout.

The query on our test environment was running for 12 seconds so we've set --query-timeout to 3000ms, restarted vtgate but and run the query again - it wasn't killed after 3 seconds, finished successfully after 12 seconds.

For the sake of the test we've changed --query-timeout to 500ms and here we had a result but it was killing 100% of the queries executed , no matter JOIN, SELECT or other.

Reproduction Steps

The part of the vschema for these tables:

{
	"sharded": true,
	"vindexes": {
		"identity_keyspace_binary": {
			"type": "binary"
		}
		"table1": {
			"columnVindexes": [{
				"column": "KeySpaceId",
				"name": "identity_keyspace_binary"
			}]
		},
		"table2": {
			"columnVindexes": [{
				"column": "KeySpaceId",
				"name": "identity_keyspace_binary"
			}]
		},

Binary Version

16.0.2

Operating System and Environment details

CentOS 7
Linux 5.15.49-linuxkit-pr
aarch64

Log Fragments

The scatter queries created by Vitess are not visible, only the one entered by us.

[GrahamCampbell]

I think this is happening because the timeout doesn't apply to the outer query. It is only applied to the individual queries sent to each vttablet.