Can you see what is in nifi-user.log when you make the curl request? ... View more

The issue is that NiFi's Jetty server is being more restrictive about only accepting requests that were intended for its actual host. So in your scenario, NiFi's Jetty server is listening on nifi-0.svc... then when you access https://master-ip:9443/nifi in your browser, the host header is set to master-ip, so when NiFi's Jetty server gets the request it rejects it because master-ip is not nifi-0.svc. NIFI-4671 that Matt pointed out is going to add some way to white-list hostnames that should not be rejected, so you would just put master-ip in the white-list and it would let it through. In the meantime, does Kubernetes have a way to control the headers that get forwarded from the master IP to the NiFi server? For example, if someone had a load balancer in front of NiFi they would access the UI through the load balancer host which would create this same problem, but if you tell the load balancer to strip off the host header then it won't happen. I don't know anything about Kubernetes so I have no idea if this is possible. ... View more

Hi Samir, There shouldn't be an end-less loop... As you can see in the HDFS example diagram, there are two parts to the flow: - ListHDFS -> RPG (this part only runs on primary node) - Input Port -> FetchHDFS -> rest of the flow (this part runs on all nodes) The starting point of your flow should be something that has no input, like ListHDFS, so there can't be a circular loop back to that point. The end of the second part should end with wherever you are sending your data, like PutHDFS for example, after that it is dead end, no loop back to anywhere. If this is not clear, please provide a screen shot or template of your flow so we can see how you have connected the processors. Thanks, Bryan ... View more

Bin packing is the standard strategy that is used when merging together data, it just writes the bytes of each flow file one after another, inserting optional header, footer, and demarcators. Defragment strategy is for when you have previously used one of the "split" processors and want to undo the split back to a single flow file. This mode requires that all of the incoming flow files have the standard "fragment" attributes like fragment.identifier, fragment.index, and fragment.count which are created by the split processors. ... View more

Shut everything down and copy flow.xml.gz from one of the other nodes to HDF1 which will get HDF1 back in sync. If you are running securely and using NiFI's file-based authorizer, then you will also want to copy users.xml and authorizations.xml from one of the other nodes to HDF1, because those are also factored into joining the cluster. ... View more

Sounds good, there is a property in the HBase Client Service where you can specify the location of phoenix-client.jar so it can be located outside NiFi's lib dir. ... View more