Bucardo chokes on large tables — let's delay indexes!

[jessehanley]

For databases with large tables, say >1TB with a handful of indexes on it, a little network hiccup can cause the whole onetimecopy to fail. We can drastically speed up the initial copy by building indexes AFTER the copy instead of during.

By doing this, we can take a ~4TB migration down from 3 days to 10-12 hours. This means someone could easily pull it off in a day on a Heroku dyno without worrying about the daily resets too.

Can we consider adding rebuild_index=1 to the initial sync?

[jessehanley]

And maybe we can have a UI element to check that all indexes are complete/healthy (pg_index.indisvalid) ... we're doing a dry run now with rebuild_index=1 and will post how it goes.