es2014 was cloned from es2005, where a test of data compression was performed, hence all the tables have in addition ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=16.
Because of the very small gain obtained, and the newer and bigger machines, we don't need it anymore, hence needs to be reverted, we can get the data from another one of the cluster.
My fault. This just needs iterating over all InnoDB tables and decompress. Compression is clearly not useful for External Storage, but we should seriously think about if for core dbs, plus overally the most optimal row format (dynamic vs compact).
Change 344332 had a related patch set uploaded (by Jcrespo):
[operations/mediawiki-config@master] Pool all es2XXX servers, depool es2014 for maintenance
Change 344332 merged by jenkins-bot:
[operations/mediawiki-config@master] mariadb: Pool all es2XXX servers, depool es2014 for maintenance
Mentioned in SAL (#wikimedia-operations) [2017-03-23T12:04:59Z] <jynus> converting es2014 tables back to uncompressed InnoDB T129350
root@es2014[(none)]> SELECT table_schema, table_name FROM information_schema.tables WHERE Engine='InnodB' AND row_format='compressed';
Empty set (6.40 sec)
Change 344913 had a related patch set uploaded (by Jcrespo):
[operations/mediawiki-config@master] mariadb: repool es2014 after maintenance
Change 344913 merged by jenkins-bot:
[operations/mediawiki-config@master] mariadb: repool es2014 after maintenance