com.ibm.streamsx.plumbing > redundant 1.0.1 > com.ibm.streamsx.plumbing.sample.redundant.active > ActiveReplicasUDP.spl
Execution of redundant copies of Flow() using @parallel. Flow is invoked once in the source code but replicated using @parallel. The number of replicas (parallel width) is set by the submission time paramater replicas defaulting to three.
This technique only works for redundant flows if the composite @parallel is applied to has no inputs. If the composite had inputs then each replica (channel) would only receive a subset of the tuples, thus not acting as a redundant flow.
Each replica flow will execute in its own hostpool defined by the host tag rack_n where n is the channel number. This is achieved by the replicateHostTags argument to @parallel. The host tag rack is replicated to N copies where N is the width of the parallel region, with the tags being rack_0 ... rack_N.
The replication of the host tags also drives the replication of host pools that use the tags, thus at runtime there are N host pools corresponding to the replicated host tags.
In this case there is no host pool Rack that corresponds to the tag rack. It would only be created if there was an operator invocation that was placed on hostpool Rack without using replicated hsot tags.
public composite ActiveReplicasUDP
{
param
expression<int32> $replicas : (int32) getSubmissionTimeValue("replicas", "3");
graph
@parallel(width=$replicas, replicateHostTags=["rack"])
() as RedundantFlow = Flow()
{
config placement: host(Rack);
}
config
// Pools are created shared to allow
// other applications to use the same hosts.
hostPool:
Rack = createPool({tags=["rack"]}, Sys.Shared);
}