Hi Ben, according to my experience, I would recommend not to use DistributedMapCache at all. First of all - its not load balanced / HA and then, it is just a HashMap implementation in Java, talking to clients using simple TCP (=it consumes RAM from NiFi's Java heap space) I guess its in NiFi as a safety net, for simple use cases. For any production / heavy traffic use case, I would go for external caching / databases (Redis?) ZooKeeper is good for any HA / balanced use case, which does not have heavy traffic, so if you have it installed and you know how to do it in ZooKeeper, go for it.
Regards, Michal On Fri, May 26, 2017 at 2:19 AM, 尹文才 <[email protected]> wrote: > Thanks Joe, actually what I want to achieve is one of my processors need to > write some config data to be kept in cluster so that other processors would > be able to easily get the config data from the cluster. > I've considered using the default NIFI state map to do it, but as I know it > could only keep data for a specific processor and it could only be retrieved > by that processor. I've never used Hazelcast, I may > just keep the config data in Zookeeper directly. > > Regards, > ben > > 2017-05-26 8:55 GMT+08:00 Joe Witt <[email protected]>: >> >> So the client would be on each of nodea, nodeb, and nodec. The server >> would be on nodea, nodeb, nodec. Each client would be configured to >> talk to any one of the three servers nodea, nodeb, or nodec. It does >> not offer HA. For more complete behavior it is a good idea to have a >> client/service implementation that talks to a full caching service. >> In the next release you can script out a controller service as a cache >> client. I believe one of the folks in the community has an example >> floating around on how to do this to talk to Hazelcast. >> >> Thanks >> Joe >> >> On Thu, May 25, 2017 at 8:45 PM, 尹文才 <[email protected]> wrote: >> > Thanks Joe, so I need to setup the DistributedMapCacheServer on all >> > nodes, >> > do you mean all DistributedMapCacheClientService should reference only >> > one >> > of the servers(the same one on the same node)? >> > What if the DistributedMapCacheServer goes down on that node(or the node >> > itself goes down), does it support HA? >> > >> > >> > 2017-05-25 20:46 GMT+08:00 Joe Witt <[email protected]>: >> >> >> >> Hello >> >> >> >> You put the DistributedMapCacheServer controller service on as well >> >> and then point at it from the client services. So in your three nodes >> >> all three will have the server service but on all the clients they'll >> >> only point to the server service on one of the nodes. >> >> >> >> Thanks >> >> Joe >> >> >> >> On Thu, May 25, 2017 at 5:29 AM, 尹文才 <[email protected]> wrote: >> >> > Hi guys, I'm currently using NIFI with 3 nodes as a cluster, when >> >> > using >> >> > DistributedMapCacheClientService there's a configuration property >> >> > called 'Server Hostname'. I've tried with localhost on my local NIFI >> >> > standlone node >> >> > and it did work. My question is what should I set inside the NIFI >> >> > cluster? >> >> > Thanks. >> > >> > > >
