Last modified: 2012-12-19 13:44:13 UTC
This is related to https://jira.toolserver.org/browse/TS-549 Certain WMF slaves are using varchar(255) for page.page_title and image.img_name (possibly other columns as well), while other slaves/wiki databses are using varbinary(255). This has to do with MySQL 5 treating varbinary as the same as varchar, from what I understand. The issue that we've run into on the Toolserver is that comparing varchar fields to varbinary fields is incredibly slow. For example, enwiki_p.page.page_title is varbinary while commonswiki_p.image.img_name is varchar, so the following query takes an incredibly long time to run: SELECT page_title FROM page WHERE NOT EXISTS (SELECT img_name FROM commonswiki_p.image WHERE img_name = page_title) LIMIT 1; The Toolserver folks are hesitant to alter the tables on their side, and it would probably be better all around for there to be consistency among the WMF databases and slaves.
all mysql5 or 5.1 slave instances should have their data as binary/ 'varbinary' do note, that 'varbinary(255)' is a 'varchar(255) binary' - so, you should actually tell which character set is used on TS side, rather than leave it untold. inconsistency is on TS side, not on ours - TS uses different charsets, than we do. we will have eventual consistency on our side, though, needs someone (maybe me) to work on it.
Toolserver has labeled TS-549 as a WONTFIX.
Changing LATER to WONTFIX as per comment 1 and 2.
