Naming is hard but to me providing what Caleb mentioned through something like WITH OPTIONS sounds reasonable. Thanks for bringing it up.
On Sat, 21 Dec 2024 at 2:46, Joel Shepherd <sheph...@amazon.com> wrote: > WITH INDEX (or something equivalent) seems really useful. > > Less opinionated on the specific syntax, but I think there is a lot of > value in the form of predictable, controllable performance, in giving > developers more direct control over query execution, whether that's > index selection or even lower-level decisions. If you've experienced the > thrill of operating a database with a cost-based planner that abruptly > selects a new, sub-optimal plan due to a change in statistics or > configuration, you'll appreciate language features that yield some > planning control back to you. It does increase the burden on the > developer to understand how best to execute the query, but it makes > their intent much more obvious, and easier to adjust as the system changes. > > -- Joel. > > On 12/20/2024 12:28 PM, Caleb Rackliffe wrote: > > Some of your are probably familiar with work in the DS fork to improve > > the selection of indexes for SAI queries in > > > https://github.com/datastax/cassandra/commit/eeb33dd62b9b74ecf818a263fd73dbe6714b0df0#diff-2830028723b7f4af5ec7450fae2c206aeefa5a2c3455eff6f4a0734a85cb5424. > > > > > > > While I'm eagerly anticipating working on that in the new year, I'm > > also wondering whether we think some simple CQL extensions to manually > > control index selection would be helpful. Maxwell proposed this a > > while back in CASSANDRA-18112, and I'd like to propose a syntax: > > > > > > ex. Do not use the specified index during the query. > > > > SELECT ... FROM ... WHERE ... WITHOUT INDEX <ks.idx> > > > > This could be helpful for intersection queries where one of the > > provided clauses is not very selective and could simply be handled via > > post-filtering. > > > > ex. Require the specified index to be used. > > > > SELECT ... FROM ... WHERE ... WITH INDEX <ks.idx> > > > > This could be helpful in scenarios where multiple indexes exist on a > > column and was the primary motivation for CASSANDRA-18112. > > > > Thoughts? >
