Status Update
Comments
cb...@google.com <cb...@google.com> #2
Thank you for opening this report.
I have seen previously this behavior and I believe it is expected. Can you provide me some additional information on what you are trying to achieve?
I have seen previously this behavior and I believe it is expected. Can you provide me some additional information on what you are trying to achieve?
he...@conrad.de <he...@conrad.de> #3
Thank you for your reply.
> Can you provide me some additional information on what you are trying to achieve?
Our real application is not Tomcat, but an Application Server that provides various Web Service endpoints. Lets call those services S1, S2, ... S100.
Some of those service have a workflow, say service S5 internally calls service S15 and service S28.
Because all Applications Servers within the same Instance Group are identical we thought we could use the Load Balancing feature. So in the example above:
(1) Client calls Webservice Service S5. The LoadBalancer directs it to AppServer1
(2) AppServer1 internally calls service S15. It does this via the Loadbalancer. We now hoped that the Loadbalancer would direct it to AppServer2 or AppServer3 (or AppServer1).
But what actually happens is that ALL calls from AppServer1 are directed back to AppServer1. So Loadbalancing INSIDE an Instance Group is not working.
> I have seen previously this behavior and I believe it is expected.
imho this is a bug. Could you please explain the idea behind it?
> Can you provide me some additional information on what you are trying to achieve?
Our real application is not Tomcat, but an Application Server that provides various Web Service endpoints. Lets call those services S1, S2, ... S100.
Some of those service have a workflow, say service S5 internally calls service S15 and service S28.
Because all Applications Servers within the same Instance Group are identical we thought we could use the Load Balancing feature. So in the example above:
(1) Client calls Webservice Service S5. The LoadBalancer directs it to AppServer1
(2) AppServer1 internally calls service S15. It does this via the Loadbalancer. We now hoped that the Loadbalancer would direct it to AppServer2 or AppServer3 (or AppServer1).
But what actually happens is that ALL calls from AppServer1 are directed back to AppServer1. So Loadbalancing INSIDE an Instance Group is not working.
> I have seen previously this behavior and I believe it is expected.
imho this is a bug. Could you please explain the idea behind it?
cb...@google.com <cb...@google.com> #4
Thanks for the additional information and feedback.
I will be merging this issue to an existing report [1]. There is an ongoing investigation to confirm weather this behavior is expected or not.
[1]https://issuetracker.google.com/67366622
I will be merging this issue to an existing report [1]. There is an ongoing investigation to confirm weather this behavior is expected or not.
[1]
Description
- autoscaling: On
- target CPU usage: 80%
- minimum = maximum number of instances = 3
I also created a Loadbalancer (internal load balancer, TCP) and used the Instance Groupd as backend.
This resulted in 3 VM instances, lets call them tom-0001, tom-0002 and tom-0003 – and one Loadbalancer. Each VM instance has an identical application server installed (for this test case I used Tomcat for the sake of simplicity).
I manually created a fourth instance. When I do some “curl -v http://<LB-IP>:8080/demo/index.html“ commands on this instance I can see that the calls are nicely distributed from the Loadbalancer to tom-0001, tom-0002 and tom-0003.
So far, so good.
But when I do the same curl from WITHIN the Instance Group, so for example on tom-0002, the Loadbalancer does not distribute the request any more – they ALL land at tom-0002 (the host where they originated).
Why is Loadbalancing not working in this case? And what can I do to make it working?