So, at work, they're setting up a database that needs to be replicated from one part of the country to another. No big deal, there's dozens of methods for doing so.
Unfortunately, it's so infrequent that one sets up such configurations, that, some small things aren't at the top of your head when you do. In this case, the flag that says to the software "both sites are already identical, just declare them synced and only sync what changes at the source site from this point forward."
In and of itself, forgetting this flag isn't inherently horrible. Instead of being instantaneously synced, you have to wait for the devices to do a zero sync. Where it becomes problematic is if you have a LOT of zeros and NOT a lot of bandwidth to ship them.
An initial replication sync of 17TB takes a LOOOONG time if you've got a 5MB/s session limit on the WAN. Specifically, at the rates we were seeing, it was going to take 45.862 days to do that initial sync. The WAN bandwidth was several Gigabits, but, those session limits were a motherfucker.
Oh well. Nothing to be lost by burning-down and redoing. Certainly, it would take sufficiently less time than to do that zero-sync that it's worth re-doing.
Bleah.
No comments:
Post a Comment