Change Details

The last 5.5 slave in s[1-7] will shortly be upgraded, pending some table partitioning, so it's time to think about 10.0 on masters. * Do we trust 10.0 enough? * Which shard should we do first? * Should we attempt to trial 10.0 under real(ish) master load (thread pool, concurrency, not just serial replicated transactions, etc)? Once we decide to move ahead, there are is other stuff we could achieve at the same time, since we're doing master rotations anyway: * Fix the remaining tables with unique/primary keys on promoted slaves. * Formally switch to MIXED or ROW (currently prod cnf is MIXED) -- and are we ready across the whole tree? * Same for GTID. * Do any shards need rebalancing or splitting (eg, any wikis to move from s3 to ??) Other questions: * Is Sanitarium affected at all? * Do we want to consider any better HA solution for masters? Auto-failiver might be still unwise for Mediawiki, but what else would help?

We already have s2 with mariadb10 masters- implement that upgrade to the other shards. The last 5.5 slave in s[1-7] will shortly be upgraded, pending some table partitioning, so it's time to think about 10.0 on masters. * Do we trust 10.0 enough? * Which shard should we do first? * Should we attempt to trial 10.0 under real(ish) master load (thread pool, concurrency, not just serial replicated transactions, etc)? Once we decide to move ahead, there are is other stuff we could achieve at the same time, since we're doing master rotations anyway: * Fix the remaining tables with unique/primary keys on promoted slaves. * Formally switch to MIXED or ROW (currently prod cnf is MIXED) -- and are we ready across the whole tree? * Same for GTID. * Do any shards need rebalancing or splitting (eg, any wikis to move from s3 to ??) Other questions: * Is Sanitarium affected at all? * Do we want to consider any better HA solution for masters? Auto-failiver might be still unwise for Mediawiki, but what else would help?

We already have s2 with mariadb10 masters- implement that upgrade to the other shards. The last 5.5 slave in s[1-7] will shortly be upgraded, pending some table partitioning, so it's time to think about 10.0 on masters. * Do we trust 10.0 enough? * Which shard should we do first? * Should we attempt to trial 10.0 under real(ish) master load (thread pool, concurrency, not just serial replicated transactions, etc)? Once we decide to move ahead, there are is other stuff we could achieve at the same time, since we're doing master rotations anyway: * Fix the remaining tables with unique/primary keys on promoted slaves. * Formally switch to MIXED or ROW (currently prod cnf is MIXED) -- and are we ready across the whole tree? * Same for GTID. * Do any shards need rebalancing or splitting (eg, any wikis to move from s3 to ??) Other questions: * Is Sanitarium affected at all? * Do we want to consider any better HA solution for masters? Auto-failiver might be still unwise for Mediawiki, but what else would help?