Re: [hobbit] big brother replacement

[Sloan <joe (at) tmsusa.com>]

Tod Hansmann wrote:
> Let me see if I understand.  You have several bb servers at one
> datacenter, each with their twin at the other datacenter, and both sets
> do the tests.  They report to one central display server, but only one
> set reports at a time, depending on failover state, correct?  
>   
You have the basic idea, but there is no single central server, just
pairs of bb servers, one to a data center, in each lan which is being
monitored. For each pair of bb servers, only the server at data center A
does reporting, unless the server in data center B cannot reach the
server in data center A, in which case the server in data center B will
take over the reporting duties until the bb server in data center A
becomes reachable again. While this could theoretically lead to a split
brain condition, the failover condition has only ever triggered when
there was a wan outage.

> Is this failover automatic?  If so, how is this failover determined?
> What if this failover has a false positive?  If not, what is your
> timeframe to swap over?
>   

IIRC It takes one bb cycle to kick in.

We've not seen a false positive, as I mentioned above.

It's just the standard built-in bb failover -

head ~bb/ext/failover follows:

#!/bin/sh

# failover
#
# BIG BROTHER - FAILOVER SCRIPT
# Sean MacGuire
#
# (c) Copyright Quest Software, Inc.  1997-2003  All rights reserved.
#

#
# failover WATCHES BBNET and BBPAGER
#
# IF BBNET OR BBPAGER BECOMES UNAVAILABLE, THEN TAKE OVER UNTIL THEY RETURN
#
# To use, just add failover to the BBEXT variable in etc/bbdef.sh
#
# To configure BBPAGER failover:
# define both the primary and failover machines as BBPAGERS in etc/bb-hosts
# and set bbwarn: FAILOVER in etc/bbwarnsetup.cfg


Joe