I need an affinityBackupFilter that will prevent backups from running in
the same AWS availability zone. (A single availability zone has the
characteristic that some or all of the EC2 instances in that zone can fail
together due to a single fault. You have no control over the hosts on
which the EC2 instance VMs run on in AWS, except by controlling the
availability zone) .
I could write a few lines of custom code, but then I have to get it
deployed on all nodes in the cluster, and peer class loading will not
work. So I cannot use an of the shelf docker image, for example. So
that code should just be part of Ignite.
I was thinking of adding new class along these lines, where the apply
function will return true only if none of the node's attributes match those
of any of the nodes in the list. This would become part of the code base,
but would only be used if configured as the backupAffinityFunction
ClusterNodeNoAttributesMatchBiPredicate implements
IgniteBiPredicate<ClusterNode,
List<ClusterNode>> {
ClusterNodeNoAttributesMatchBiPredicate(String[] attributeNames)
{....}
For AvailabilityZones, there would be only one attribute examined, but we
have some potential use cases for distributing backups across two
sub-groups of an AZ.
Alternately, we could enhance the RendezvousAffinityFunction to allow one
or more arbitrary attributes to be compared to determine neighbors,
rather than only org.apache.ignite.macs, and to add a setting that
controls whether backups should be placed on neighbors if they can't be
placed anywhere else.
If I have 2 backups and three availability zones (AZ), I want one copy of
the data in each AZ. If all nodes in one AZ fail, I want to be able to
decide to try to get to three copies anyway, increasing the per node
footprint by 50%, or to only run with one backup. This would also give
be a convoluted way to change the number of backups of a cache
dynamically: Start the cache with a large number of backups, but don't
provide a location where the backup would be allowed to run initially.
