[Spread-users] spread configuration reload partitions group
jr at vertica.com
Thu Dec 8 14:58:39 EST 2011
Trying to expand a cluster from 20 to 32 daemons.
On the one of 20 running daemons, connected with spmonitor and issued a
'r' to reload the configuration. Later, new 12 daemons were started.
In a group across the original 20 daemons (1 client/daemon), 15 of the
20 saw the other 5 disappear from their group; the unlucky 5 saw all
other 19 disappear.
Is this expected? Is there a way to tune for this scale of cluster to
ride through the transient better?
[I regret to say we have logging directed to /dev/null since slow disk
knocks the daemons over otherwise.]
More information about the Spread-users