https://bugzilla.wikimedia.org/show_bug.cgi?id=64140
--- Comment #11 from Sean Pringle <sprin...@wikimedia.org> --- tools-db is labsdb1005. It isn't a production replicant, so this is unrelated to the replication outage. LOAD DATA INFILE importing a lot of data to a transactional storage engine like InnoDB is asking for trouble because the statement is a also a transaction -- an 18GB transaction :-) That means a whole lot of undo log entries are generated in case of rollback, much purge thread lag occurs, other traffic is slowed, and if the undo log exceeds buffer pool capacity it will start thrashing the tablespace on disk. If the transaction gets killed and has to rollback the whole process will be even slower than the original load. All that can be done now is: a) Wait it out. b) Restart mysqld and hope crash recovery is faster (may not be). c) Not nice stuff that will need downtime for everyone. In the future, better to do bulk data loads into ARIA tables then ALTER them to be InnoDB if you need it specifically. Your plan to batch the inserts in 2GB chunks is also good. -- You are receiving this mail because: You are on the CC list for the bug. _______________________________________________ Wikibugs-l mailing list Wikibugs-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikibugs-l
