HTTP Session Failover in WildFly 11 for different machines

[almiller]

Hi, I'm trying to setup session sharing between 2 different machines. (Machine A = 192.168.10.114  node1, node2, Machine B = 192.168.10.7 : node3, node4)

When I run all 4 servers in a standalone mode using ha profile via cmd line:

Machine A = 192.168.10.114:

1.>bin\standalone.bat -b 192.168.10.114 -Djboss.server.base.dir=node1 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=500 -Djboss.node.name=node1

2.>bin\standalone.bat -b 192.168.10.114 -Djboss.server.base.dir=node2 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=1000 -Djboss.node.name=node2

and Machine B = 192.168.10.7:

3.standalone.bat -b 192.168.10.7 -Djboss.server.base.dir=node3 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=200 -Djboss.node.name=node3

4.standalone.bat -b 192.168.10.7 -Djboss.server.base.dir=node4 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=400 -Djboss.node.name=node4

the result: I'm getting 2 clusters (1,2) & (3,4). Each node in a cluster is sharing session only with another one on the same machine but session isn't shared between all the nodes.

I would like to have all 4 nodes as a part of one cluster.

Also tried:

setup domain mode with ha profile as explained in: HTTP Session Failover in WildFly 8 · WildFly . Machine A: running server-group ABC with servers one & two and Machine B is running server-group with the same name  ABC (to try to make it construct one cluster) with servers: three & four.

And again session is shared only between 2 nodes on the same machines.

Please, assist configuring it properly for all servers on both machines.

P.S. I also run jgroups testing tool to verify multicast connection and it was successful in both directions (Sender-Listener) between my 2 machines.

Thanks!

[rhn-support-abhati]

HI,

Are you using UDP based clustering in your setup or TCP based.

For UDP based you need to define a multicast address while starting server like "./standalone.sh -u 230.0.0.5", please try this.

Also make sure that your network is properly configured.

[almiller]

Hi, thanks for your reply.

Tried again to run standalone servers with UDP based clustering:

on Machine A = 192.168.10.114:

1) >bin\standalone.bat -b 192.168.10.114 -u 224.0.0.1 -Djboss.server.base.dir=node1 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=200 -Djboss.node.name=node1

2) >bin\standalone.bat -b 192.168.10.114 -u 224.0.0.1 -Djboss.server.base.dir=node2 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=300 -Djboss.node.name=node2

and on Machine B = 192.168.10.7:

3) >bin\standalone.bat -b 192.168.10.7 -u 224.0.0.1 -Djboss.server.base.dir=node3 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=250 -Djboss.node.name=node3

4) >bin\standalone.bat -b 192.168.10.7 -u 224.0.0.1 -Djboss.server.base.dir=node4 --server-config=standalone-ha.xml -Djboss.socket.binding.port-offset=350 -Djboss.node.name=node4

The result: Cluster is constructed as expected containing 4 nodes: [node1,node2,node3,node4].

But.. I'm getting same session id only for nodes 1,2 and 3,4 but not for 1,2,3,4.  As far as I understand, having <distributable/> in my web.xml enables session replication and that's why nodes 1 & 2 share same session as nodes 3 & 4 do as well.

But how to make all of them share the same session?

Thanks!

[plohia]

Hi ,

Could you please confirm below :

1> Are you deploying the same application on both the machines ?

2> Please share me the configuration files .

3> Could you switch to TCP based clustering and check if it resolves the issue.

-Pranab

[rhn-support-abhati]

Hi Alex,

Can you please confirm if you have deployed same application on all 4 servers ?

Can you elaborate on this "But how to make all of them share the same session?"

If your cluster is formed with all 4 nodes in cluster, then the session id would be replicated and present in all 4 nodes.

You can enable TRACE logging and check in logs for the generated session id.

[almiller]

Hi,

1.Yes, same war file. All it's doing is answering rest GET http request and showing sessionid.

2.      standalone-ha.xml for node 1(machine A) : https://drive.google.com/file/d/12BcEqLPCoCYX-9qO4CTuJbbNlk0jlGjn/view?usp=sharing

settings for another node on Machine B are almost the same, just the IP specified in <interfaces> is the one of Machine B.

3. switched to TCP,  it's the same.

I'm wondering, this probably my lack of knowledge, it would be great if you could confirm and elaborate:

If we try to access our app via different nodes http://192.168.10.114:8180/simple-service-webapp/webapi/myresource/hello  and  http://192.168.10.7:8330/simple-service-webapp/webapi/myresource/hello, I'm getting different session id.

But if I hit:  http://192.168.10.114:8180/simple-service-webapp/webapi/myresource/hello and http://192.168.10.114:8280/simple-service-webapp/webapi/myresource/hello     (only the port is different), I'm getting same session id.

So, my question is why at all I expect to see same session id if we hit different urls? And why when the difference is only in port number it works? (port number is not considered part of url for server in order to decide whether to create new session id or not?)

If this is the issue, may be the solution would be to put a Load Balancer in front of app-servers and that way we would always hit the same url?

Thanks!!

[almiller]

Hi,

- yes, same war file was deployed.

- as I wrote in previous response:

I'm wondering, this probably my lack of knowledge, it would be great if you could confirm and elaborate:

If we try to access our app via different nodes http://192.168.10.114:8180/simple-service-webapp/webapi/myresource/hello  and  http://192.168.10.7:8330/simple-service-webapp/webapi/myresource/hello, I'm getting different session id's.

But if I hit:  http://192.168.10.114:8180/simple-service-webapp/webapi/myresource/hello and http://192.168.10.114:8280/simple-service-webapp/webapi/myresource/hello     (only the port is different), I'm getting same session id.

So, my question is why at all I expect to see same session id if I hit different url's? And why when the difference is only in port number it works? (port number is not considered part of url for server in order to decide whether to create new session id or not?)

If this is the issue, may be the solution would be to put a Load Balancer in front of app-servers and that way we would always hit the same url?

Can you refer to it?

p.s. Tried it. I put a load balancer (another wildfly on Machine A - undertow) in front of  app-servers and try to access: http://192.168.10.114:8080/simple-service-webapp/webapi/myresource/hello  (8080 - is the LB) and I'm getting same session id as when I hit: http://192.168.10.114:8180/simple-service-webapp/webapi/myresource/hello  (directly invoking machine A, means that the request goes through machine A). When I try to hit http://192.168.10.7:8330/simple-service-webapp/webapi/myresource/hello (Machine B), I'm getting different sessionid.

But.. if I turn off  app-server on Machine A, and try to access my app via LB I see that sessionid is kept although machine A is off which means that Machine B responded and sessionid maintained.  Can someone confirm/elaborate if this is an expected behavior?

Thanks a lot!

[plohia]

Hi Alex ,

>> So, my question is why at all I expect to see same session id if I hit different url's? And why when the difference is only in port number it works?

I understand that it's not able to replicate the session among the machines , if the machines are in the different network it may be caused due to firewall issue.

Could you change it to TCP clustering and let me know if this resolves your issue.

-Pranab

[rhusar]

Your test is wrong. Think how browsers work: they can only send cookies to the same domain where they originated, so 192.168.10.114 and 192.168.10.7 are clearly different domains from browsers' perspective, so its not going to send over that cookie at all, thus its a new and different session.

You either need to setup a load-balancer or make both ip addresses resolve for the same domain (e.g. by editing /etc/hosts).

[rhusar]

If your cluster is formed with all 4 nodes in cluster, then the session id would be replicated and present in all 4 nodes.

Not entirely correct: yes, the session will be available on all nodes, but by default, WF11 uses distribution cache mode with default two owners, so the data will be physically always on 2 nodes. If accessed on a node which does not have the data physically, it will be fetched from a node which holds the data.

[almiller]

Hi Radoslav, thanks a lot for your answer.

It confirms my concern "So, my question is why at all I expect to see same session id if I hit different url's? And why when the difference is only in port number it works? "

I'd like to summarize and it would be great if you could confirm:

1) Since we hit different domains (192.168.10.114 & 192.168.10.7) we end up with 2 different jsessionid's generated by wildfly. That's why we use Load Balancer to make server create one session id and reuse it when it's coming from another machine?

2) Regarding the ports difference and why it works, the fact that session id is maintained for 192.168.10.114:8180 & 192.168.10.114:8280 (same machine, different ports) means that browser doesn't distinguish between client requests coming towards different ports and this is an expected behavior (according to some RFC probably)?

Thanks

[almiller]

Hi, thanks for you reply

- It's in the same network

- TCP clustering gives same result.

I think Radoslav's answer exactly refers to the point why it happens (different domains).