Thank You
The current behavior when it runs only from coordinator is good enough.
I have modified you example to run the predicate on nodes to check for example
some file existence .
I think it stateless but for sure not idempotent .
See below it just returns nodes that has some file on the filesystem.
Hope it helps to understand it . Thank You
What I understand the flow is like that :
1. When deployed the service will run this to determine eligible nodes by
looking for this file
2. My service healthcheck code will check periodically if this node is valid
for running the service and do following :
a. Delete file
b. Cancel service
c. Deploy the same service with same configuration and predicate
3. #1 will run again
Is it clear enough ? The current behavior when it runs only from coordinator is
OK as I am using cluster singletons for my services.
Ignite ignite=
IgnitionEx.start("examples/config/example-ignite.xml","ignite-1");
Ignite ignite2=
IgnitionEx.start("examples/config/example-ignite.xml","ignite-2");
ClusterGroup cg=ignite2.cluster().forPredicate(new
IgnitePredicate<ClusterNode>(){
Ignite filterIgnite;
@Override public boolean apply(ClusterNode node){
System.out.println("ignite: "+(isNull(filterIgnite)?null
:filterIgnite.name()));
if(filterIgnite!=null){
return filterIgnite.compute(filterIgnite.cluster()
.forNode(clusterNode))
.apply(new IgniteClosure<String, Boolean>(){
@Override
public Boolean apply(String filename){
File file=new File(filename);
return file.exists();
}
},"/tmp/this_node_is_valid");
}
return false
}
@IgniteInstanceResource
void setFilterIgnite(Ignite filterIgnite){
this.filterIgnite=filterIgnite;
}});
// Deploy services only on server nodes.
ignite.services(cg).deploy(new ServiceConfiguration()
.setMaxPerNodeCount(1)
.setName("my-service")
.setService(new SimpleMapServiceImpl<>())
);
On 5/29/18, 11:38, "aealexsandrov" <[email protected]> wrote:
Hi,
I checked next code:
Ignite ignite =
IgnitionEx.start("examples/config/example-ignite.xml", "ignite-1");
Ignite ignite2 =
IgnitionEx.start("examples/config/example-ignite.xml", "ignite-2");
ClusterGroup cg = ignite2.cluster().forPredicate(new
IgnitePredicate<ClusterNode>() {
Ignite filterIgnite;
@Override public boolean apply(ClusterNode node) {
System.out.println("ignite: " + (isNull(filterIgnite) ? null
: filterIgnite.name()));
return true;
}
@IgniteInstanceResource
void setFilterIgnite(Ignite filterIgnite) {
this.filterIgnite = filterIgnite;
}});
// Deploy services only on server nodes.
ignite.services(cg).deploy(new ServiceConfiguration()
.setMaxPerNodeCount(1)
.setName("my-service")
.setService(new SimpleMapServiceImpl<>())
);
It has the same behavior as nodeFilter predicate: it starts from the
coordinator node. And looks like it should be stateless too because it used
in ServiceConfiguration.
ignite: ignite-1
ignite: ignite-1
Service was initialized: my-service
Service was initialized: my-service
Executing distributed service: my-service
Executing distributed service: my-service
ignite: ignite-1
ignite: ignite-1
ignite: ignite-1
ignite: ignite-1
Maybe I missed something. Could you please provide the code that you propose
to use?
BR,
Andrei
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
