Oleg Bartunov wrote on 12/21/22 12:31
PM:I On Wed, Dec 21, 2022 at 1:12 PM Oleg Bartunov <obartunov@xxxxxxxxxxxxxx> wrote:Dan, it's always good to specify configuration name in a query to avoid recheck, since websearch_to_tsquery(regconfig, text) is immutable, while websearch_to_tsquery(text) is stable. I am glad to hear this because one of my decisions was: * change the query to include configuration name * change the default configuration name I think I'm going to start using websearch_to_tsquery(regconfig, text) immutable function calculates once in planning time, but stable function calculates during running time, so the difference may be very big depending on how many tuples found. Looking at my data, the largest tsvector collection has 453 elements. About 40 have more than 170. Total number of rows is about 2400. Another issue discussed on IRC: why store my tsvector values? Why not just index them instead? At present: I have this column: pkgmessage_textsearchable | tsvector | | | generated always as (to_tsvector('english'::regconfig, pkgmessage)) stored with this index: ports_pkgmessage_textsearchable_idx" gin (pkgmessage_textsearchable) Instead, I could replace that column and index with this index: "testing" gin (to_tsvector('english'::regconfig, pkgmessage)) Simple testing showed it was comparable if not slightly faster. The plan now: implement the index on to_tsvector, not a column, and start specifying the configuration. That's in the near future. I've written up this journey at https://news.freshports.org/2022/12/18/when-tsvector-was-working-as-expected-on-most-hosts-but-not-one/ thank you
|
