Description
I'm trying to speed up a gh-ost alter of a gigantic table, and I was wondering if you'd considered deferring index creation until after the main backfill? I imagine it'd go a lot faster if the initial table had no indexes, we did the bulk copy of all of the existing rows, then we paused and did a full ALTER TABLE
to create the secondary indexes, then gh-ost replayed the binlog to finish the migration.
I could probably do something like this myself by specifying an alter to gh-ost which drops all indexes, then throttling gh-ost through the control socket, when it gets to 99%, running the ALTER TABLE
on the _gho
table to add the missing indexes, and un-throttling gh-ost once the ALTER finishes in replication, but that seems failure-prone.
Alternatively, if you have any tips for making rewrites go faster on terabyte-scale tables, I'm all ears. :-)