Enable lookup-joins for join benchmarks based on session settings #323
Conversation
specs/joins.toml
Outdated
| @@ -8,6 +8,9 @@ statements = [ | |||
| "optimize table articles with (max_num_segments = 1)", | |||
| ] | |||
|
|
|||
| [session_settings] | |||
| optimizer_equi_join_to_lookup_join = "true" | |||
There was a problem hiding this comment.
do we need it true here? This can make joins which were executed with hashjoin all this time to be executed with lookup join now, no? Isn't it enough to set it only to the other lookup_join.toml file?
There was a problem hiding this comment.
Why would we disable a join optimization in a join benchmark ? For the query
"select * from articles inner join colors on articles.id = colors.id where colors.id = -1"
this optimization is significant.
There was a problem hiding this comment.
my concern is that then we loose benchmark info for a normal inner hash join. Can we track perf regressions for this case then by another query?
If not, imho, we could add it to the specialized file for lookup joins.
|
I believe you need to update this: https://github.com/crate/crate-benchmarks/blob/master/requirements.txt#L14 once cr8 with the new functionality is released. |
mkleen
force-pushed
the
mkleen/enable-lookup-joins
branch
from
4bb4ea4
to
d364cd0
Compare
Summary of the changes / Why this is an improvement
Depending on mfussenegger/cr8#381
Checklist