Major Description
This email trail needs to be cleaned up, but capturing this to not forget to explore this...
> > > That was one aspect of the action item, but I feel like there was a second.
> > > You mentioned some possibility of a data race on startup of a node after it had been previously failed over. The race involved the existence of pindex data that wasn't expected, and the realization that it might be wanted again. Are we confident that no such race exists?
> > Hard/Graceful failover is one side of the equation, but Delta or Full recovery is the other side.
> > Marty is right, there was a question about race condition when adding the node back into the cluster and whether the FTS indexes would need to be fully rebuilt or whether the data on disk can be re-used when adding a node back in...
> If there is indeed a race, I'd claim it's in the territory of bug, not as-intended. As to whether such a bug exists, not sure... haven't tested it myself.
Ok, all these double checking replies are now making me question my own thinking and what I think I know.
Let me walk back what I said and pop out an MB to not forget.
From: Perry Krug
Sent: Wednesday, March 30, 2016 10:37 AM
To: Marty Schoch
Cc: Steve Yen; Will Gardella; Kirk Kirkconnell; Perry Krug
Subject: Re: FTS failover followup
Hard/Graceful failover is one side of the equation, but Delta or Full recovery is the other side.
Marty is right, there was a question about race condition when adding the node back into the cluster and whether the FTS indexes would need to be fully rebuilt or whether the data on disk can be re-used when adding a node back in...
Perry
Perry Krug
Manager - Solutions Engineering US-West
+1.831.824.4123
perry@couchbase.com
On Wed, Mar 30, 2016 at 10:19 AM, Marty Schoch <marty@couchbase.com> wrote:
Thanks Steve,
That was one aspect of the action item, but I feel like there was a second.
You mentioned some possibility of a data race on startup of a node after it had been previously failed over. The race involved the existence of pindex data that wasn't expected, and the realization that it might be wanted again. Are we confident that no such race exists?
marty
On Wed, Mar 30, 2016 at 1:12 PM, Steve Yen <steve@couchbase.com> wrote:
Hi folks,
Following up on an A/I, I just looked again at the code, and there's no distinction between graceful and hard failover for Watson FTS.
Cheers,
Steve