Terracotta Discussion Forums

I'm quite interested in Terracotta's capabilities but I've always been turned off by the 'Split Brain' problem, chiefly because our production environment makes this kind of scenario quite likely.

I'd really like to understand what happens to data, and especially recent changes, in the Split Brain scenario.
I've noticed that the HA Architecture description says things like

The new active server "zaps" the original active server

but it's not clear exactly what this means.

So, for the sake of clarity, could someone please explain what happens in this scenario:

0) Two TC servers are on separate networks
1) The connection between the two TC servers goes down but all other functions continue operating
2) The previously passive TC server becomes active
3) Clients are connected to both Active servers, inserting new data into both.
4) The network comes back up, Split Brain is detected, and one of the servers "zaps" the other.

What happens to the data on the one that is "zapped"?
Are all changes to that server since the Split Brain occurred simply wiped, never to be recovered?

I beleive, you have read the HA Architecture Description care fully, it explains that Terracotta cluster can solve almost all split-brain occurrences without loss or corruption of shared data. On the basis of number of cleints, Terracotta decides which active server remain active and zaps the data of other active server. That means all new data is merged and remain at active server. The active server having less connected clients , restarts, wipe its data and start as passive ans sync the data with active server.

hope this will answer ur quires.

Cheers!!

Hi Gyan,

Thank you very much for your response. However, I am still confused about exactly what happens to changes that occur during the split brain. You have said both that one "active server zaps the data of other active server", but also that "all new data is merged". I am assuming that "zaps" here means "deletes". Please correct if that's wrong.

So, in regards to the post-partition changes from the server that will be restarted, are those changes deleted, or are they merged into the other server?

I would also be interested to know exactly what happens to the data here. Presumably the server that is "zapped" simply deletes everything and resynchronizes with the elected one.

Even if you managed to merge somehow the data, taking the most recent for instance, you would probably end up with an incoherent set of data, which is presumably not what you want.

Thanks,
Mark

I think the docs on this page are clear that the zapped server's data is cleared, not integrated into the other server's data. Only then can the zapped server start up as a mirror and sync with the active server's data.