[13:20:43] <inflatador>	 <o/
[13:21:39] <ebernhardson>	 o
[13:59:40] <inflatador>	 not sure if you saw this in #security yet, but there's a new log4j CVE: https://github.com/hypnguyen1209/log4j2-rce
[13:59:49] <ebernhardson>	 oh fun, no i hadn't
[14:05:08] <ebernhardson>	 i'm not sure how to judge if opensearch is fine. I know in the normal pathways it avoids java serialization and does it's own thing, but in log4j land not as clear.  I would hope it doesn't
[14:06:58] <ebernhardson>	 but this looks like it requires being able to receive log messages externally, sounds kinda narrow
[14:08:47] <inflatador>	 yeah, that was the conclusion in #security as well. Sounds like our clusters are OK
[15:59:02] <ebernhardson>	 fun oddity: python has str.isalnum which is unicode aware.  So thai characters are alpha-numeric.  But thai combining marks are not, so if you tried to widen a highlight by going to the next non-alphanumeric character you break things
[17:26:32] <ebernhardson>	 redirects are out an appear about as expected. for enwiki that means 12M redirects in NS_MAIN and 66M redirects overall that were previously unsearchable now are
[17:28:00] <ebernhardson>	 i suppose i should count those somewhere else to verify...but our `page_rerender_upsert 
[17:28:22] <ebernhardson>	 `page_rerender_upsert | created` even have fallen off, suggesting the system isn't finding new redirects to create anymore
[22:02:09] <ebernhardson>	 demo of mdeberta highlights against the public dataset research put together: https://people.wikimedia.org/~ebernhardson/mdeberta-highlights/
[22:37:48] <ebernhardson>	 i couldn't find a public version of the final dataset with passage grades though, just the original knn result export from david. the data is in google drive, but restricted so i didn't want to release it here
[23:27:33] <ebernhardson>	 also wrote docs for the cirrus redirect scope, but i feel like it's a bit unclear still: https://www.mediawiki.org/wiki/Help:CirrusSearch#Redirects_in_Search