This weekend, while an ongoing incident was being handled, I checked and saw several badly performing queries running. These didn't have (I believe) any relation with the incident, but any web request that takes more than 1 or a few seconds (in the worse possible case) is worrying, as they can pile up and create load issues on the database.
While there was several queries with bad performance (mostly related to recentchanges and API), the worst occurring query right now is:
SELECT /* Wikibase\Lib\Store\Sql\SiteLinkTable::getLinks */ ips_site_id, ips_site_page, ips_item_id FROM `wb_items_per_site` WHERE ips_site_id = '?';
where '?' can be any production wiki, but in particular, for enwiki, it selects over 8 million rows. We saw over 30 of those, over all s8 servers, today Mar 8 at 8:18 UTC.
This not only takes over 1 minute to run, but is likely to be not useful due to the amount of memory consumed.
While reading large number of rows can be useful for something like a dump/analytics, it is unlikely to be useful for a web request. Moreso, the application server will struggle to handle 2GB of data at a time, even in memory, that this query returns.
Is it possible some extra WHERE or LIMIT was missing from the query?
Acceptance Criteria🏕️🌟 :
- There are no unlimited selects happening for this query
